Line bundles on the moduli space of parabolic connections over a compact Riemann surface

1. First note that a subbundle and its quotient bundle of a parabolic vector bundle admits a parabolic structure from its original parabolic bundle (see the remark after Definition 1.8 of [MS]). We shall use usual technique of Kernel-Image and Image-Coimage sequences to prove the first part.

   Let \ensuremath{\mathrm{Ker} (\phi)}_* \subset E_* is a parabolic subbundle. Then, \ensuremath{\mathrm{Ker} (\phi)}_* is D-invariant parabolic subbundle of E. Since (E_*, D) is stable, we have p\mu(\ensuremath{\mathrm{Ker} (\phi)}_*) < p\mu(E_*). Since \phi \neq 0, we have \ensuremath{\mathrm{Im} (\phi)} \neq 0. Now, \ensuremath{\mathrm{Im} (\phi)} inherits a parabolic structure from E'_*.

   Consider the Kernel-Image short exact sequence

   0 \to \ensuremath{\mathrm{Ker} (\phi)}_* \to E_* \to \ensuremath{\mathrm{Im} (\phi)}_* \to 0.

   Then, p\mu(E_*) < p\mu(\ensuremath{\mathrm{Im} (\phi)}_*). Next, consider the Image-Coimage short exact sequence

   0 \to \ensuremath{\mathrm{Im} (\phi)}_* \to E'_* \to (E' / \ensuremath{\mathrm{Im} (\phi)})_* \to 0.

   Note that \ensuremath{\mathrm{Im} (\phi)}_* is a D'-invariant parabolic subbundle of E'. Since (E'_*, D') is stable, we have p\mu(\ensuremath{\mathrm{Im} (\phi)}_*) < p\mu (E'_*). Thus, we get that p\mu(E_*) < p\mu(E'_*) which contradicts the assumption that p\mu(E_*) = p\mu (E'_*). Therefore, on underlying vector bundles, we have \ensuremath{\mathrm{Ker} (\phi)} = 0 and \ensuremath{\mathrm{Im} (\phi)} = E'.

2. Let \psi : (E_*, D) \to (E_*, D) be the given endomorphism. Let x \in X, and let \lambda \in \ensuremath{\mathbb{C}} be an eigen value of the linear map \psi(x) : E_x \to E_x. Note that X is a compact Riemann surface, it does not admit any non-constant holomorphic function, therefore the eigen values and their multiplicities are independent of x \in X. Since D is \ensuremath{\mathbb{C}}-linear, \psi - \lambda \ensuremath{\mathbf{1}_{E_*}} is an endomorphism of (E_*, D), that is,

   D \circ (\psi - \lambda \ensuremath{\mathbf{1}_{E_*}}) = (\psi - \lambda \ensuremath{\mathbf{1}_{E_*}}) \otimes \ensuremath{\mathbf{1}_{\Omega^1_X(S)}} \circ D.

   Since (E_*, D) is stable, from (?), \psi - \lambda \ensuremath{\mathbf{1}_{E_*}} is either a zero morphism or an isomorphism. Since \lambda is an eigen value of \psi(x), kernel of \psi - \lambda \ensuremath{\mathbf{1}_{E_*}} is non-trivial. Thus, \psi = \lambda \ensuremath{\mathbf{1}_{E_*}}, where \lambda \in \ensuremath{\mathbb{C}}.

Since E_* is stable parabolic bundle, we have \ensuremath{\mathrm{H}^{0}(X,\ensuremath{\mathrm{ParEnd} (E_*)})} = \ensuremath{\mathbb{C}} \cdot \ensuremath{\mathbf{1}_{E_*}}. Now, Lemma follows from above discussion.

Let (E_*, D) \in Z_{pc}. Then, E_* is not stable. Since \alpha is generic, E_* is not semi-stable. Let

be the Harder-Narasimhan filtration of the parabolic vector bundle E_*. The collection of pairs of integers \{(\ensuremath{\mathrm{rk}_{}(E^i_*)}, p\deg{E^i_*})\}_{i =1}^l is called the Harder-Narasimhan polygon of E_* (see [SS, p.n.178]).

Now, analogous to the techniques in [BM7, Lemma 3.1] (p.n. 303), the space of all isomorphism classes of parabolic vector bundles over X whose Harder-Narasimhan polygon coincides with the given parabolic vector bundle E_* is of dimension at most

We have already computed that the dimension of the space of all isomorphic classes of parabolic connections , lying in \ensuremath{\mathcal{M}}_{pc}(r, d, \alpha), on any given parabolic vector bundle E'_* (assuming that parabolic connection exists on E'_*) which is

The subvariety of \ensuremath{\mathcal{M}}_{pc}(r,d, \alpha) parametrizing all pairs of the form (E'_*, D') \in \ensuremath{\mathcal{M}}_{pc}(r,d, \alpha) such that the Harder-Narasimhan polygon of the parabolic vector bundle E'_* coincides with that of E_* is at most of dimension

Since \text{dim} \ensuremath{\mathcal{M}}_{pc}(r, d, \alpha) = 2r^2(g-1) +2 + m(r^2-r), we have

Thus, \ensuremath{\mathrm{codim}(Z_{pc}, \ensuremath{\mathcal{M}}_{pc}(r, d, \alpha))} \geq (r-1)(g-2) + 1.

The morphism \iota^* defined in (?) is an isomorphism follows from the fact that \ensuremath{\mathrm{codim}(Z_{pc}, \ensuremath{\mathcal{M}}_{pc}(r, d, \alpha))} \geq 2 as proved in Lemma 3.2.

Now, we show that \pi^* is an isomorphism. From Lemma 2.3, the moduli space \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha) is T^*\ensuremath{\mathcal{M}}(r, d, \alpha)-torsor. We use this fact to compactify the moduli space \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha), and using this compactification, we show that the Picard group \ensuremath{\mathrm{Pic} (\ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha))} is isomorphic to \ensuremath{\mathrm{Pic} (\ensuremath{\mathcal{M}}(r, d, \alpha))}.

We have seen that for any E_* \in \ensuremath{\mathcal{M}}(r, d, \alpha), the fibre \pi^{-1}(E_*) is an affine space modelled on \ensuremath{\mathrm{H}^{0}(X,\Omega^1_X(S) \otimes \ensuremath{\mathrm{SParEnd} (E_*)})}. We know that the dual of an affine space is a vector space, therefore the dual

is a vector space over \ensuremath{\mathbb{C}}.

Let

be the algebraic vector bundle such that for every Zariski open subset U of \ensuremath{\mathcal{M}}(r, d, \alpha), a section of \ensuremath{\mathcal{W}} over U is an algebraic function f: \pi^{-1}(U) \to \ensuremath{\mathbb{C}} whose restriction to each fiber \pi^{-1}(E_*), is an element of \pi^{-1}(E_*)^{\vee}.

Thus, a fibre \ensuremath{\mathcal{W}}(E_*) = \psi^{-1}(E_*) of \ensuremath{\mathcal{W}} at E_* \in \ensuremath{\mathcal{M}}(r, d, \alpha) is \pi^{-1}(E_*)^{\vee}. The dimension of \pi^{-1}(E_*)^{\vee} is equal to \text{dim}_{\ensuremath{\mathbb{C}}} \pi^{-1}(E_*) + 1, and since E_* is stable, the dimension of \pi^{-1}(E_*)^{\vee} is equal to

follows from Lemma 3.1.

Let (E_*,D) \in \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha), and define a map

by \Psi_{(E_*, D)}(\varphi) = \varphi[(E_*, D)], that is in fact the evaluation map. Now, the kernel \ensuremath{\mathrm{Ker} (\Psi_{(E_*, D)})} defines a hyperplane in \pi^{-1}(E_*)^{\vee} denoted by H_{(E_*, D)}.

Let \ensuremath{\mathbf{P}} (\ensuremath{\mathcal{W}}) be the projective bundle defined by hyperplanes in the fibre \pi^{-1}(E_*)^{\vee}, that is, we have

induced from \psi.

Next, define a map

by sending (E_*, D) to the equivalence class of H_{(E_*, D)}, which is an open embedding. Set

Then \tilde{\psi}^{-1}(E_*) \cap {\bf H} is a projective hyperplane in \tilde{\psi}^{-1}(E_*) for every E_* \in \ensuremath{\mathcal{M}}(r, d, \alpha), and hence {\bf H} is a hyperplane at infinity.

We first show that \pi^* is injective. Let \eta \to \ensuremath{\mathcal{M}}(r, d, \alpha) be a line bundle such that \pi^* \eta is a trivial line bundle over \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha). A trivialization of \pi^* \eta is equivalent to have a nowhere vanishing section of \pi^* \eta over \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha). Fix s \in \ensuremath{\mathrm{H}^{0}(\ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha),\pi^* \eta)} a nowhere vanishing section. Choose a point E_* \in \ensuremath{\mathcal{M}}(r, d, \alpha). Then, from the following commutative diagram

we get

a nowhere vanishing map. Notice that \pi^{-1}(E_*) \cong \ensuremath{\mathbb{C}}^N and \eta(E_*) \cong \ensuremath{\mathbb{C}}, where N = r^2(g-1)+\frac{m(r^2-r)}{2} + 1. Now, any nowhere vanishing algebraic function on an affine space \ensuremath{\mathbb{C}}^N is a constant function, that is, s|_{\pi^{-1}(E_*)} is a constant function and hence corresponds to a non-zero vector \alpha_{E_*} \in \eta(E_*). Since s is constant on each fiber of \pi, the trivialization s of \pi^*\eta descends to a trivialization of the line bundle \eta over \ensuremath{\mathcal{M}}(r, d, \alpha), and hence giving a nowhere vanishing section of \eta over \ensuremath{\mathcal{M}}(r, d, \alpha). Thus, \eta is a trivial line bundle over \ensuremath{\mathcal{M}}(r, d, \alpha).

It remains to show that \pi^* is surjective. Let \vartheta \to \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha) be an algebraic line bundle. Since \ensuremath{\mathbf{P}} (\ensuremath{\mathcal{W}}) is a smooth compactification of \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha) follows from the embedding i : \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha) \hookrightarrow \ensuremath{\mathbf{P}} (\ensuremath{\mathcal{W}}) in (?), the homomorphism of Picard groups

defined by pull back of line bundles via i in (?), is a surjective homomorphism. Therefore, we can extend \vartheta to a line bundle \vartheta' over \ensuremath{\mathbf{P}} (\ensuremath{\mathcal{W}}). Again from the morphism \tilde{\psi}: \ensuremath{\mathbf{P}}(\ensuremath{\mathcal{W}}) \to \ensuremath{\mathcal{M}}(r, d, \alpha) in (?) and from [H], Chapter III, Exercise 12.5, p.n. 291, we have

Therefore,

where \Lambda is a line bundle over \ensuremath{\mathcal{M}} and l \in \ensuremath{\mathbb{Z}}. Since {\bf H} = \ensuremath{\mathbf{P}} (\ensuremath{\mathcal{W}}) \setminus \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha) is the hyperplane at infinity, using (?) the line bundle \ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}} (\ensuremath{\mathcal{W}})}}({\bf H}) associated to the divisor {\bf H} can be expressed as

for some line bundle \Gamma over \ensuremath{\mathcal{M}}(r, d, \alpha). Now, from (?) and (?), we get

Since, the restriction of the line bundle \ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}} (\ensuremath{\mathcal{W}})}}({\bf H}) to the complement

is the trivial line bundle and restriction of \tilde{\psi} to \ensuremath{\mathcal{M}}'_{pc}(r, d, \alpha) is the map \pi defined in (?), therefore, we have

This completes the proof of the theorem.

Consider the dual of the Atiyah exact sequence in (?), that is,

Consider \ensuremath{\mathcal{O}_{\ensuremath{\mathcal{M}}}} as the trivial line bundle \ensuremath{\mathcal{M}} \times \ensuremath{\mathbb{C}}. Let

be a holomorphic section of the trivial line bundle defined by x \mapsto (x,1). Let Y = \ensuremath{\mathrm{Im} (\alpha)} \subset \ensuremath{\mathcal{M}} \times \ensuremath{\mathbb{C}} be the image of \alpha. Then

In fact, for every open subset U \subset \ensuremath{\mathcal{M}}, a holomorphic section of \ensuremath{\mathcal{C}}(L)|_{U} over U gives a holomorphic splitting of the Atiyah exact sequence (?), associated to the holomorphic vector bundle L| _{U} \to U. For instance, suppose \tau: U \to \ensuremath{\mathcal{C}}(L)| _{U} is a holomorphic section. Then \tau will be a holomorphic section of \ensuremath{\mathrm{At} (L)}^*|_{U} over U, because \ensuremath{\mathcal{C}}(L) = {\imath^*}^{-1}Y \subset \ensuremath{\mathrm{At} (L)}^*. Since

so we get a holomorphic splitting \tau of the Atiyah exact sequence (?) associated to L|_{U}. Thus, L|_{U} admits a holomorphic connection. Conversely, given any splitting of Atiyah exact sequence (?) over an open subset U \subset \ensuremath{\mathcal{M}}, we get a holomorphic section of \ensuremath{\mathcal{C}}(L)|_{U} over U.

Let \ensuremath{\mathbf{P}}(\ensuremath{\mathrm{At} (L)}) be the projectivization of \ensuremath{\mathrm{At} (L)}, that is, \ensuremath{\mathbf{P}}(\ensuremath{\mathrm{At} (L)}) parametrises hyperplanes in \ensuremath{\mathrm{At} (L)}. Let \ensuremath{\mathbf{P}}({T \ensuremath{\mathcal{M}}}) be the projectivization of the tangent bundle T\ensuremath{\mathcal{M}}. Notice that \ensuremath{\mathbf{P}}(T\ensuremath{\mathcal{M}}) is a subvariety of \ensuremath{\mathbf{P}} (\ensuremath{\mathrm{At} (L)}), and \ensuremath{\mathbf{P}}(T\ensuremath{\mathcal{M}}) is the zero locus of the of a section of the tautological line bundle \ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}} (\ensuremath{\mathrm{At} (L)})}}(1). Now, observe that

So we have

Since for any finite dimensional vector space V over \ensuremath{\mathbb{C}} and for every k \geq 0, we have \ensuremath{\mathrm{H}^{0}(\ensuremath{\mathbf{P}}(V),\ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}}(V)}}(k))} = \ensuremath{\mathcal{S}}ym^k(V), where \ensuremath{\mathcal{S}}ym^k(V) denote the k-th symmetric powers of V. We get a natural isomorphism

where \ensuremath{\mathcal{S}}ym^k\ensuremath{\mathrm{At} (L)} denote the k-the symmetric powers of \ensuremath{\mathrm{At} (L)}, and hence from (?), we get

The symbol operator

as described in (?), induces a morphism

on k-th symmetric powers of bundles. In view of the following composition

we have

In fact, we have the symbol exact sequence associated to L over \ensuremath{\mathcal{M}},

In other words, we get a filtration

such that

To prove (4.2), it is enough to show that

We have the following commutative diagram

which gives the following commutative diagram of long exact sequences

In order to show (4.2), in view of (?), it is enough to prove that the boundary operator \delta'_k is injective for all k \geq 1 and which is equivalent to showing that the boundary operator

is injective for every k \geq 1.

Further a connecting homomorphism can be expressed as the cup product by the extension class of the corresponding short exact sequence. Denote the extension class of the following short exact sequence

by \gamma_{k}.

Moreover, \gamma_{k} can be expressed in terms of the first Chern class c_1(L), because the first Chern class of c_1(L) is nothing but the extension class of the following Atiyah exact sequence (see [A])

and the short exact sequence (?) is the k-th symmetric power of the Atiyah exact sequence (?).

Thus, the connecting homomorphism \delta_{k} can be described using the first Chern class c_1(L) \in \ensuremath{\mathrm{H}^{1}(\ensuremath{\mathcal{M}}, T^* \ensuremath{\mathcal{M}})} of the line bundle L.

The cup product with kc_1(L) gives rise to a homomorphism

Also, we have a canonical homomorphism of vector bundles

which induces a morphism of C-vector spaces

So, we get a morphism

Then \tilde{\mu} = \delta_{k}. Now, it is enough to show that \tilde{\mu} is injective.

Consider the natural projection

where T^*\ensuremath{\mathcal{M}} denote the cotangent bundle of \ensuremath{\mathcal{M}}. Then, we have

Moreover, using the projection formula we have

Using (?) and (?) we get

Now, to compute \ensuremath{\mathrm{H}^{0}(T^*\ensuremath{\mathcal{M}},\ensuremath{\mathcal{O}_{T^*\ensuremath{\mathcal{M}}}})} we use the Hitchin fibration for the moduli space of strongly parabolic Higgs bundles as defined in (?). Recall that for any generic point a \in \ensuremath{\mathcal{H}}_P, we have h_P^{-1}(a) = A, where A is some abelian variety (for more details see [SWW], [BGL]), and we will be using this fact showing that \tilde{\mu} is injective.

Let g: T^*\ensuremath{\mathcal{M}} \to \ensuremath{\mathbb{C}} be an algebraic function. Since T^*\ensuremath{\mathcal{M}} is an open subset of \ensuremath{\mathcal{M}}_{Higgs }(r, d, \alpha) such that the complement has codimension at least 2 (see (?)), from Hartog’s theorem, the algebraic function g is extended to an algebraic function

Then its restriction \tilde{g}|_{h_P^{-1}(a)}: h_P^{-1}(a) \to \ensuremath{\mathbb{C}} to h_P^{-1}(a) is an algebraic function. Since h_P^{-1}(a) is an abelian variety, \tilde{g} is a constant function. Therefore, on generic fibre h^{-1}(a), \tilde{g} is constant, and h_P is proper, hence gives an algebraic function on \ensuremath{\mathcal{H}}_P. Thus, any algebraic function on T^*\ensuremath{\mathcal{M}} descends to an algebraic function on \ensuremath{\mathcal{H}}_P.

Set \ensuremath{\mathcal{P}} = \mbox{d}( \ensuremath{\mathrm{H}^{0}(\ensuremath{\mathcal{H}}_P,\ensuremath{\mathcal{O}_{\ensuremath{\mathcal{H}}_P}})}) \subset \ensuremath{\mathrm{H}^{0}(\ensuremath{\mathcal{H}}_P,\Omega^1_{\ensuremath{\mathcal{H}}_P})} the space of all exact algebraic 1-form. Define a map

by g \mapsto d \tilde{g}, where \tilde{g} is the function which is defined by descent of g. Then \theta is an isomorphism.

From (?) and (?), we have

which is an isomorphism. Let us denote the restriction of h_P on T^*\ensuremath{\mathcal{M}} by h'_{P}.

Let T_{h'_P} = T_{T^* \ensuremath{\mathcal{M}} / \ensuremath{\mathcal{H}}_P} = \ensuremath{\mathcal{K}er (\mbox{d}h'_P)} be the relative tangent sheaf on T^*\ensuremath{\mathcal{M}}, where

morphism of bundles.

Note that \ensuremath{\mathrm{H}^{0}(\ensuremath{\mathcal{H}}_P,\Omega^1_{\ensuremath{\mathcal{H}}_P})} \subset \ensuremath{\mathrm{H}^{0}(T^*\ensuremath{\mathcal{M}},T_{h'_P})}, and hence from (?), we have an injective homomorphism

Consider the morphism

defined by taking cup product with the first Chern class c_1(p^* L) \in \ensuremath{\mathrm{H}^{1}(T^*\ensuremath{\mathcal{M}}, T^* T^* \ensuremath{\mathcal{M}})}.

Using the pairing

we get a homomorphism

Since c_1(p^* L) = p^*c_1 (L), we have

for all \omega_k \in \ensuremath{\mathrm{H}^{0}(\ensuremath{\mathcal{M}},\ensuremath{\mathcal{S}}ym^{k} T \ensuremath{\mathcal{M}})}). Since \nu and \theta are injective homomorphisms, it is enough to show that \eta|_{\nu(\ensuremath{\mathcal{P}})} is injective homomorphism.

Let \omega \in \ensuremath{\mathcal{P}} \setminus \{0\} be a non-zero exact 1-form. Choose a \in \ensuremath{\mathcal{H}}_P such that \omega(a) \neq 0. In view of (?) and (?), the generic fibre has the form

where A is an abelian variety and F is a subvariety of A such that \mbox{codim}(F,A) \geq 2. Now, \eta(\nu(\omega)) \in \ensuremath{\mathrm{H}^{1}(T^*\ensuremath{\mathcal{M}},\ensuremath{\mathcal{O}_{T^*\ensuremath{\mathcal{M}}}})} and we have restriction map

Since \omega(a) \neq 0, \eta(\nu(\omega)) \in \ensuremath{\mathrm{H}^{1}({h'_P}^{-1}(a),\ensuremath{\mathcal{O}_{{h'_P}^{-1}(a)}})}. Because of the following isomorphisms

it follows that \eta(\nu(\omega)) \neq 0. This completes the proof.

Recall that the map \pi_{\xi} defined in (?) is a T^*\ensuremath{\mathcal{M}}(r, \alpha, \xi)-torsor. Let

be the algebraic vector bundle such that for every Zariski open subset U of \ensuremath{\mathcal{M}}(r, \alpha, \xi), a section of \ensuremath{\mathcal{V}} over U is an algebraic function f: \pi_{\xi}^{-1}(U) \to \ensuremath{\mathbb{C}} whose restriction to each fiber \pi_{\xi}^{-1}(E_*), is an element of \pi_{\xi}^{-1}(E_*)^{\vee}. Now, the rest of the proof is exactly similar to a part of the proof of the Theorem 3.3.

Let \ensuremath{\mathcal{N}}_{\ensuremath{\mathbf{P}}(\ensuremath{\mathcal{V}})/\bf{H}_0} denote the normal bundle of the divisor {\bf H}_0 \subset \ensuremath{\mathbf{P}} (\ensuremath{\mathcal{V}}), where

is the vector bundle (see (5.1)) in Proposition 5.1.

Recall that the effective divisor \bf{H}_0 is numerically effective if and only if the restriction of the line bundle \ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}} (\ensuremath{\mathcal{V}})}}(\bf{H}_0) to \bf{H}_0 is numerically effective. From Poincaré adjunction formula we have the following

Therefore, \bf{H}_0 is numerically effective if and only if the normal bundle \ensuremath{\mathcal{N}}_{\ensuremath{\mathbf{P}}(\ensuremath{\mathcal{V}})/\bf{H}_0} is numerically effective. Recall that the tangent bundle T \ensuremath{\mathcal{M}}_\xi is numerically effective if and only if the tautological line bundle \ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}} (T \ensuremath{\mathcal{M}}_\xi)}}(1) is numerically effective.

Thus, to prove the proposition it is enough to show that the normal bundle \ensuremath{\mathcal{N}}_{\ensuremath{\mathbf{P}}(\ensuremath{\mathcal{V}})/\bf{H}_0} is canonically isomorphic to the tautological line bundle \ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}} (T \ensuremath{\mathcal{M}}_\xi)}}(1).

First we show that the divisor \bf{H}_0 is canonically isomorphic to projective bundle \ensuremath{\mathbf{P}} (T \ensuremath{\mathcal{M}}_\xi). Let

be the projective bundle. Let E_* \in \ensuremath{\mathcal{M}}_\xi, and

Then, \theta represents a hyperplane in the fibre \ensuremath{\mathcal{V}}(E_*) = \pi_{\xi}^{-1}(E_*)^{\vee} of the vector bundle \ensuremath{\mathcal{V}}, where \pi_\xi is defined in (?). Let H_{\theta} denote this hyperplane represented by \theta. Note that H_{\theta} \subset \pi_{\xi}^{-1}(E_*)^{\vee} and \pi_{\xi}^{-1}(E_*) is the affine space modelled over the vector space \ensuremath{\mathrm{H}^{0}(X,\Omega^1_X(S) \otimes \ensuremath{\mathrm{SParEnd'} (E_*)})}, therefore for any f \in H_{\theta}, we have

Since \theta \in \bf{H}_0, and H_\theta is a hyperplane, the subspace of T_{E_*} \ensuremath{\mathcal{M}}_{\xi} generated by \{\text{d}f\}_{f \in H_{\theta}} is a hyperplane in T_{E_*} \ensuremath{\mathcal{M}}_{\xi}. Let \tilde{\theta} \in \ensuremath{\mathbf{P}} (T_{E_*} \ensuremath{\mathcal{M}}_{\xi}) denote this hyperplane. Thus, we get a canonical isomorphism

by sending \theta to \tilde{\theta}.

Next, note that the fibre \ensuremath{\mathcal{N}}_{\ensuremath{\mathbf{P}}(\ensuremath{\mathcal{V}})/\bf{H}_0}(\theta) of the normal bundle \ensuremath{\mathcal{N}}_{\ensuremath{\mathbf{P}}(\ensuremath{\mathcal{V}})/\bf{H}_0} is canonically isomorphic to the quotient \ensuremath{\mathcal{V}}(E_*) / H_{\theta}. Consider the morphism

of vector spaces defined by sending f \mapsto \text{d} f. Since image of the hyperplane H_{\theta} is contained in \tilde{\theta}, we have well defined morphism on quotients

which is an isomorphism. Recall that the fibre of the tautological line bundle \ensuremath{\mathcal{O}_{\ensuremath{\mathbf{P}} (T \ensuremath{\mathcal{M}}_{\xi})}}(1) at E_* is canonically identified with the quotient T_{E_*} \ensuremath{\mathcal{M}}_{\xi}/ \tilde{\theta}. This completes the proof.