The integral Picard groups of low-degree Hurwitz spaces

By excision, we know that

Meanwhile, \operatorname{Pic}(\mathscr{X}_{4,g}') is a quotient of \operatorname{Pic}(\mathscr{B}_{4,g}) =\mathbb{Z} (\epsilon a_1) \oplus \mathbb{Z} a_2' \oplus \mathbb{Z} b_2' (see (?)). Hence, using Lemma 5.4, we see \operatorname{Pic}(\mathscr{H}_{4,g}') is a quotient of

Recall that in Lemma 5.1, we showed \operatorname{Pic}(\mathscr{H}_{4,g}) = \operatorname{Pic}(\mathscr{H}_{4,g}') for g \geq 4. By [DP, Proposition 2.15], this group has rank at least 2. Therefore, \operatorname{Pic}(\mathscr{H}_{4,g}) = \mathbb{Z} \oplus \mathbb{Z}, since any quotient would have smaller rank.

By Lemma 5.3, the kernel of the restriction map

lies in the subgroup \langle a_1, a_2' \rangle. We know that \operatorname{Pic}(\mathscr{H}'_{4,g}) is generated by the classes a_1, a_2', b_2', so it follows that \operatorname{Pic}(\mathscr{H}_{4,3}) is also generated by these 3 classes.

Next, we claim that b_2' is integrally expressible in terms of a_1, a_2'. By Lemma 5.4, we know b_2' is expressible in terms of a_1, a_2' in \operatorname{Pic}(\mathscr{H}_{4,3}'). But, the kernel of \operatorname{Pic}(\mathscr{H}_{4,3}) \to \operatorname{Pic}(\mathscr{H}_{4,3}') lies in \langle a_1, a_2' \rangle, so b_2' must also be expressible in terms of a_1, a_2' in \operatorname{Pic}(\mathscr{H}_{4,3}).

It follows that \operatorname{Pic}(\mathscr{H}_{4,3}) is a quotient of \mathbb{Z} a_1 \oplus \mathbb{Z} a_2'. However, by [DP], we know that \operatorname{Pic}(\mathscr{H}_{4,3}) has rank 2. Any further quotient would have lower rank, so we are done.

We have already established that \operatorname{Pic}(\mathscr{H}_{4,2}) is generated by 2a_1 and a_2'. Using Example 3.4, we compute that (see (?) for definitions of the maps f, \alpha, \pi)

From this we see that \lambda and 2a_1 are generators for \operatorname{Pic}(\mathscr{H}_{4,2}). Since \lambda is the generator of \operatorname{Pic}(\mathscr{M}_2) \cong \mathbb{Z}/10\mathbb{Z}, we see that \operatorname{Pic}(\mathscr{H}_{4,2}) is a quotient of \mathbb{Z} \oplus \mathbb{Z}/10.

By the discussion at the start of this section, we know \operatorname{Pic}(\mathscr{H}_{4,2}) has rank at least 1. Thus, it remains to prove that \operatorname{Pic}(\mathscr{M}_2) \to \operatorname{Pic}(\mathscr{H}_{4,2}) is injective.

Let \mathscr{P}^k be the universal Picard variety over \mathscr{M}_2 as in Section ? (the \mathbb{G}_m-rigidification of the universal Picard stack \mathscr{P}ic^k). As in [Mochizuki, Section 6], the natural map \mathscr{H}_{4,2} \to \mathscr{P}^4 factors through a Grassmann fibration. For this, recall that every degree 4 line bundle on a genus 2 curve has a 3-dimensional space of sections. Let \mathscr{G} \to \mathscr{P}^4 be the Grassmann fibration parametrizing two-dimensional subspaces of the space of global sections of a degree 4 line bundle. Then \mathscr{H}_{4,2} sits naturally as an open substack \mathscr{G}.

Its complement Z = \mathscr{G} \smallsetminus \mathscr{H}_{4,2} is the locus of pencils with a base point. Note that Z has 1-dimensional irreducible fibers over \mathscr{P}^4, so Z is irreducible. Since Z meets each fiber of \mathscr{G} \to \mathscr{P}^4, it is not equivalent to the pullback of a divisor on \mathscr{P}^4. In particular, the map \operatorname{Pic}(\mathscr{P}^4) \to \operatorname{Pic}(\mathscr{H}_{4,2}) must be injective. Using Lemma 4.1, we conclude that \operatorname{Pic}(\mathscr{M}_2) \to \operatorname{Pic}(\mathscr{H}_{4,2}) is also injective, completing the proof.

By Lemma 6.1, \operatorname{Pic}(\mathscr{H}_{5,g})\cong \operatorname{Pic}(\mathscr{H}_{5,g}') for g\geq 4. We have that \operatorname{Pic}(\mathscr{H}'_{5,g}) is a quotient of \operatorname{Pic}(\mathscr{X}'_{5,g})\cong \operatorname{Pic}(\mathscr{B}'_{5,g}) by classes supported on \mathscr{D}_{5,g}. By Lemma 6.3, one such class is (10g+36)a_1-7a_2'-b_2'. Therefore, \operatorname{Pic}(\mathscr{H}'_{5,g}) is a quotient of

By [DP, Proposition 2.15], the rank of \operatorname{Pic}(\mathscr{H}_{5,g})\otimes \mathbb{Q} is at least 2, so we must have that \operatorname{Pic}(\mathscr{H}_{5,g})\cong \mathbb{Z}(\epsilon a_1)\oplus \mathbb{Z} a_2'.

By Lemma 6.2, the kernel of the restriction map

lies in the subgroup \langle 2a_1, a_2' \rangle. We know that \operatorname{Pic}(\mathscr{H}'_{5,g}) is generated by the classes a_1, a_2', b_2', so it follows that \operatorname{Pic}(\mathscr{H}_{5,3}) is also generated by these 3 classes.

Next, we claim that b_2' is integrally expressible in terms of a_1, a_2'. By Lemma 6.3, we know b_2' is expressible in terms of a_1, a_2' in \operatorname{Pic}(\mathscr{H}_{5,3}'). But, the kernel of \operatorname{Pic}(\mathscr{H}_{5,3}) \to \operatorname{Pic}(\mathscr{H}_{5,3}') lies in \langle a_1, a_2' \rangle, so b_2' must also be expressible in terms of a_1, a_2' in \operatorname{Pic}(\mathscr{H}_{5,3}).

It follows that \operatorname{Pic}(\mathscr{H}_{5,3}) is a quotient of \mathbb{Z} a_1 \oplus \mathbb{Z} a_2'. However, by [DP], we know that \operatorname{Pic}(\mathscr{H}_{5,3})\otimes \mathbb{Q} has rank 2, so \operatorname{Pic}(\mathscr{H}_{5,3})\cong \mathbb{Z}\oplus \mathbb{Z}.

We have already established that a_1, a_2' are generators for \operatorname{Pic}(\mathscr{H}_{5,2}). We compute directly \lambda = 5a_1 - a_2' as in Section ?, from which we see \lambda and a_1 are generators for \operatorname{Pic}(\mathscr{H}_{5,2}). Arguing as in Section ?, it suffices to show that the map \operatorname{Pic}(\mathscr{M}_2) \to \operatorname{Pic}(\mathscr{H}_{5,2}) is injective.

Every degree 5 line bundle on a genus 2 curve has a 4-dimensional space of sections. The Hurwitz space \mathscr{H}_{5,2} then sits naturally as an open inside the Grassmann fibration \mathscr{G} \to \mathscr{P}^5 parametrizing 2-dimensional subspaces of the space of global sections of a degree 5 line bundle. The complement of \mathscr{H}_{5,2} \subset \mathscr{G} is the locus of pencils with a base point, which we again see is irreducible and not equivalent to the pullback of a divisor on \mathscr{P}^5. In particular, the map \operatorname{Pic}(\mathscr{P}^5) \to \operatorname{Pic}(\mathscr{H}_{5,2}) must be injective. Applying Lemma 4.1, we conclude that \operatorname{Pic}(\mathscr{M}_2) \to \operatorname{Pic}(\mathscr{H}_{5,2}) is also injective, completing the proof.

Using excision, for g \geq 3, we have

Now observe that

The matrix on the left of (Proof) is invertible over \mathbb{Z}. Thus, \operatorname{Pic}(\mathscr{H}_{3,g}^s) is the sum of two cyclic groups with orders given by the diagonal entries of the matrix on the right of (Proof). This completes the proof for g \geq 3.

Finally, for g = 2, we already know a_2'= 2a_1 by (?) and that 10a_1 = 0 in \operatorname{Pic}(\mathscr{H}_{3,g}). When we remove T, this creates one additional relation 0 = T = 84a_1-24a_2' = 36a_1 in \operatorname{Pic}(\mathscr{H}_{3,g}^s). We have \gcd(10,36) = 2, so \operatorname{Pic}(\mathscr{H}_{3,2}^s) = \mathbb{Z}/2\mathbb{Z}.

By excision, we have \operatorname{Pic}(\mathscr{H}_{4,g}^s) = \operatorname{Pic}(\mathscr{H}_{4,g})/\langle T, D \rangle. Row operations over \mathbb{Z} diagonalize the change of basis matrix for \epsilon a_1, a_2' to T, D, namely we have

where the matrix on the left is invertible over \mathbb{Z}. By its definition in (?), we have \epsilon = 1 when g is odd and \epsilon = 2 when g is even, so the corollary follows for g \geq 3.

Finally, when g = 2, the above tells us 18(2a_1) = 0 and 12a_2' = 0, and we have the additional relation

Using generators 2a_1 + a_2' and 2a_1, we see that they generate cyclic groups of order 2 and 18 respectively.

By excision \operatorname{Pic}(\mathscr{H}_{5,g}^s) = \operatorname{Pic}(\mathscr{H}_{5,g})/\langle T, D \rangle. Again, row operations over \mathbb{Z} diagonalize the change of basis matrix for \epsilon a_1, a_2' to T, D:

For g \geq 3, these are the only relations, so \operatorname{Pic}(\mathscr{H}_{5,g}^s) is the sum of two cyclic groups of orders equal to the diagonal entries above.

In genus 2, the above gives 44a_1 = 0 and 12a_2' =0, and we have the additional relation

The generators a_1 and 3a_1+a_2' generate cyclic groups of order 44 and 2 respectively.

The proof is the same as in [BV, Theorem 4.4], except that instead of taking P\rightarrow S a \mathbb{P}^1-fibration in the definition of the various stacks, we take P=\mathbb{P} V\rightarrow S where V is a rank 2 vector bundle with trivial determinant.

Recall that \operatorname{Pic}(\mathrm{B}G) is naturally identified with the character group of G because it is identified with Mumford’s functorial Picard group. The exact sequence of groups

induces a left exact sequence

The Picard group \operatorname{Pic}(\mathrm{B}\mu_2) is isomorphic to \mathbb{Z}/2\mathbb{Z}. Let h be a generator of \operatorname{Pic}(\mathrm{B} \mu_2). Recall that the map \mu_2 \to G_{r,d,s,e} \times \operatorname{SL}_2 sends -1 to (-\operatorname{Id}_d, \operatorname{Id}_{r+d}, -\operatorname{Id}_e, \operatorname{Id}_{s+e}, -\operatorname{Id}_2). The generator t_1 \in \operatorname{Pic}(\mathrm{B}(G_{r,d,s,e} \times \operatorname{SL}_2)) corresponds to the determinant of the rank d matrix. Thus, the right-hand map in (Proof) sends t_1 to dh. Similarly, u_1 and w_1 are sent to zero, and v_1 to eh. The kernel is thus the subgroup generated by the classes listed in (3.5).

The translation between (3.5) and (3.5) follows from Example 3.4. We explain the case d, e both odd, the other cases being similar but simpler. Since d and e are both odd, the following change of basis matrix has integer coefficients

The determinant of the 4 \times 4 matrix above is 1 so the entries of the two column vectors generate the same subgroup with \mathbb{Z}-coefficients.

It suffices to treat the case when X' \subset X is the complement of an irreducible divisor D. (Removing any component of codimension two or more from X does not change \operatorname{Pic}(X); if X' is the complement of a reducible divisor, then we just apply the irreducible case to each component in turn.)

Because f: Y \to X is smooth, Y' \subset Y is the complement of the irreducible divisor f^{-1}(D), which has class f^*[D]. Let \langle [D] \rangle denote the subgroup of \operatorname{Pic}(X) generated by the fundamental class of D and similarly for \langle f^*[D] \rangle inside \operatorname{Pic}(Y). We therefore have a diagram of exact sequences where the left vertical map is surjective.

The result now follows from the snake lemma.

Under the correspondence between \mathbb{G}_m-torsors and line bundles, X is the complement of the zero section in the total space of the line bundle \mathcal{L}\rightarrow Y. We thus have the exact sequence

After identifying \operatorname{Pic}(\mathcal{L}) with \operatorname{Pic}(Y), the first map in the sequence is given by multiplication by the first Chern class c_1(\mathcal{L}), and the result follows.

There are natural isomorphisms \mathscr{P}^k \cong \mathscr{P}^{k+2} (given by tensoring with the canonical), so it suffices to prove the claim for k = 2 and k = 3. When k = 2, the canonical line bundle gives a section \mathscr{M}_2 \to \mathscr{P}^k, so the pullback map must be injective.

Now consider the case k = 3. Every degree 3 line bundle on a genus 2 curve has 2 sections. Therefore, \mathscr{H}_{3,2} is naturally an open substack inside \mathscr{P}^3. (It is the complement of the universal curve \mathscr{C} \hookrightarrow \mathscr{P}^3 embedded by summing each point with a canonical divisor.) The isomorphism \operatorname{Pic}(\mathscr{M}_2) \to \operatorname{Pic}(\mathscr{H}_{3,2}) factors through \operatorname{Pic}(\mathscr{M}_2) \to \operatorname{Pic}(\mathscr{P}^3), so the latter must also be injective.

Following the notation of [DP], let M(E, F) \subset \mathscr{H}_{4,g} denote the locus of covers \alpha with E_\alpha \cong E and F_\alpha \cong F. The complement of \mathscr{H}'_{4,g}\subset \mathscr{H}_{4,g} is the union of M(E, F) such that

If E = \mathcal{O}(e_1) \oplus \mathcal{O}(e_2) \oplus \mathcal{O}(e_3) with e_1 \leq e_2 \leq e_3 and F = \mathcal{O}(f_1) \oplus \mathcal{O}(f_2) with f_1 \leq f_2, then (Proof) is equivalent to 2e_1 - f_1 \leq -2.

Next, let E_{\operatorname{gen}} and F_{\operatorname{gen}} denote the balanced bundles of rank 3 and 2 and degree g+3. First note that h^1(F_{\operatorname{gen}}^\vee \otimes \operatorname{Sym}^2 E_{\operatorname{gen}}) = 0: this is equivalent to 2 \lfloor \frac{g+3}{3} \rfloor - \lceil \frac{g+3}{2} \rceil \geq -1. This says that M(E_{\operatorname{gen}}, F_{\operatorname{gen}}) \subseteq \mathscr{H}_{4,g}'. Hence, any divisorial component of \mathscr{H}_{4,g} \smallsetminus \mathscr{H}_{4,g}' is contained in a divisorial component of \mathscr{H}_{4,g} \smallsetminus M(E_{\operatorname{gen}}, F_{\operatorname{gen}}).

Next, let us define bundles that are “one-off" from balanced

In [DP, p. 20], Deopurkar–Patel enumerate the divisorial components of \mathscr{H}_{4,g} \smallsetminus M(E_{\operatorname{gen}}, F_{\operatorname{gen}}) and show that, for g \neq 3, they are always of the form

Note that we are using the irreducibility of M and CE in [DP, Propositions 4.5 and 4.7] to write these divisors as the closures above. Meanwhile, when g = 3, the stratum

is also a divisor. This divisor does not lie in \mathscr{H}_{4,g}'.

One readily checks that

Hence, when they are defined, \mathscr{H}_{4,g}' contains each of

and, for g \neq 3, all other possible M(E, F) have codimension at least 2.

Continuing the notation of the previous lemma, first note that if

then by [part1, Proposition 5.6(2)], we have f_1 \leq 2 and f_2 \leq 4 and f_1 + f_2 = 6, hence f_1 = 2 and f_2 = 4. The divisor in (?) can therefore be viewed as the locus were the universal \mathscr{E} (pulled back along \mathscr{H}_{4,3} \to \mathscr{B}_{4,3}) has splitting type (1, 2, 3) on fibers of the universal \mathbb{P}^1-fibration. As a splitting locus for \mathscr{E}, this divisor occurs in the expected codimension, so its fundamental class is determined by the universal splitting loci formulas of [L]. In particular, it can be expressed in terms of the classes \epsilon a_1, a_2'.

Base changing (?) by \operatorname{BSL}_2 \to \operatorname{BPGL}_2, we obtain the stacks and morphisms considered in [part1, Section 5.2] (below \Delta_{4,g}' is the complement of the open inclusion \mathcal{H}_{4,g}' \hookrightarrow \mathcal{X}_{4,g}' of [part1, Lemma 5.3]):

We claim some combination of components of \Delta_{4,g}' has class

This will establish the lemma since the map \operatorname{Pic}(\mathscr{X}'_{4,g}) \to \operatorname{Pic}(\mathcal{X}'_{4,g}) sends the class of a component of \mathscr{D}_{4,g} to the class of the corresponding component of \Delta_{4,g}'. Because \operatorname{Pic}(\mathscr{X}'_{4,g}) \to \operatorname{Pic}(\mathcal{X}'_{4,g}) is injective (by Lemma 3.6), classes are represented by the same formulas in either group.

By [part2, Lemma 5.2 and Equation 5.7], we know that (8g + 20)a_1 - 8a_2' - b_2' = 0 in \operatorname{Pic}(\mathcal{H}_{4,g}), so this relation must also hold on the open substack \mathcal{H}_{4,g}' \subset \mathcal{H}_{4,g}. But, we also know \mathcal{H}_{4,g}' \cong \mathcal{X}'_{4,g} \smallsetminus \Delta_{4,g}'. By excision, every relation among a_1, a_2', b_2' restricted to \operatorname{Pic}(\mathcal{H}_{4,g}') comes from a class supported on \Delta_{4,g}' \subset \mathcal{X}'_{4,g}. That is, some combination of components of \Delta_{4,g}' has class (8g + 20)a_1 - 8a_2' - b_2' = 0. The corresponding combination of components of \mathscr{D}_{4,g} will have the same class.

Following the notation of [DP], let M(E, F) \subset \mathscr{H}_{5,g} denote the locus of covers \alpha with E_\alpha \cong E and F_\alpha \cong F. The complement of \mathscr{H}'_{5,g}\subset \mathscr{H}_{5,g} is the union of M(E, F) such that

If E = \mathcal{O}(e_1) \oplus \cdots \oplus \mathcal{O}(e_4) with e_1 \leq \cdots \leq e_4 and F = \mathcal{O}(f_1) \oplus \cdots \oplus \mathcal{O}(f_5) with f_1 \leq \cdots \leq f_5, then (Proof) is equivalent to e_1 + f_1 + f_2 - (g+4) \leq -2.

Next, let E_{\operatorname{gen}}, respectively F_{\operatorname{gen}}, denote the balanced bundle of rank 4 and degree g+4, respectively rank 5 and degree 2g+8. First note that h^1(E_{\operatorname{gen}} \otimes \det E_{\operatorname{gen}}^\vee \otimes \wedge^2 F_{\operatorname{gen}}^\vee) = 0, as in this case, e_1 + f_1 + f_2 - (g+4) \geq \lfloor \frac{g+4}{4} \rfloor + 2 \lfloor \frac{2(g+4)}{5} \rfloor - (g+4) \geq -1 (except when g = 3, in which case f_2 = \lceil \frac{2(g+4)}{5}\rceil, so we still have e_1 + f_1+f_2 \geq -1.) This says that M(E_{\operatorname{gen}}, F_{\operatorname{gen}}) \subseteq \mathscr{H}_{5,g}'. Hence, any divisorial component of \mathscr{H}_{5,g} \smallsetminus \mathscr{H}_{5,g}' is contained in a divisorial component of \mathscr{H}_{5,g} \smallsetminus M(E_{\operatorname{gen}}, F_{\operatorname{gen}}).

Again, let us define bundles that are “one-off" from balanced

In [DP, p. 25], Deopurkar–Patel enumerate the divisorial components of \mathscr{H}_{5,g} \smallsetminus M(E_{\operatorname{gen}}, F_{\operatorname{gen}}) and show that, for g \neq 3, they are always of the form

Note that we are using the irreducibility of M and CE in [DP, Propositions 5.1 and 5.2] to write these divisors as the closures above. One readily checks that

Hence, when they are defined, \mathscr{H}_{5,g}' contains each of

and all other possible M(E, F) have codimension at least 2.

The divisorial component of the complement of \mathscr{H}_{5,3}' inside \mathscr{H}_{5,3} is the locus

By [DP, Proposition 5.2], this locus is precisely the preimage of the hyperelliptic locus under the natural morphism \mathscr{H}_{5,3}\rightarrow \mathscr{M}_3. By [DL], the hyperelliptic locus in \mathscr{M}_3 has class 9\lambda. The class \lambda = c_1(f_*\omega_f) = c_1(\pi_* \mathscr{E} \otimes \omega_{\pi}) \in \operatorname{Pic}(\mathscr{H}_{5,3}) is pulled back from \operatorname{Pic}(\mathscr{B}_{5,3}). Since \operatorname{Pic}(\mathscr{B}_{5,3}) includes into \operatorname{Pic}(\mathcal{B}_{5,3}), we can determine this class via a calculation on the \operatorname{SL}_2 quotient, as in Example 3.4:

which is in the span of 2a_1 and a_2'.

Base changing (?) by \operatorname{BSL}_2 \to \operatorname{BPGL}_2, we obtain the stacks and morphisms considered in [part1, Section 5.3] (below \Delta_{5,g}' is the complement of the open inclusion \mathcal{H}_{5,g}' \hookrightarrow \mathcal{X}_{5,g}' of [part1, Lemma 5.11]):

By [part2, Lemma 6.6 and Equation 6.19], the relation \frac{(10g+36)}{\epsilon}(\epsilon a_1)-7a_2'-b_2'=0 holds in \operatorname{Pic}(\mathcal{H}_{5,g}), so it must also hold on the open substack \mathcal{H}'_{5,g}. Since \mathcal{H}'_{5,g}=\mathcal{X}'_{5,g}\smallsetminus \Delta_{5,g}', any relation among a_1, a_2', b_2' on \mathcal{H}'_{5,g} must come from a class supported on \Delta_{5,g}'. Therefore, some combination of components of \Delta_{5,g}' has class (10g+36)a_1-7a_2'-b_2'. The corresponding combination of components on \mathscr{D}_{5,g} has the same class.