The intrinsic cohomology ring of the universal compactified Jacobian over the moduli space of stable curves

We consider the universal curve

over the moduli of stable marked curves (or rather, the base change to the moduli of stable marked curves with a level structure constructed in [ACV]). Since we assume n\geq 1, there is a section through the smooth locus of the family. Using this section, universal fine compactified Jacobians of different degrees are identified. So we only need to prove that

with \phi, \phi' nondegenerate. In view of Theorem 2.13, it suffices to check that

1. (i) the universal curve (Proof) is \delta-regular,

2. (ii) the universal fine compactified Jacobian \overline{J}^{0,\phi}_{g,n} is nonsingular, and

3. (iii) the morphism \pi: \overline{J}^{0,\phi}_{g,n} \to \overline{{\mathcal M}}_{g,n} has full support,

where we assume in (ii, iii) that \phi is nondegenerate.

Indeed, (ii) is given by [KP, Corollary 4.4]; (i) and (iii) are given by [MSV, Fact 2.4 and Theorem 1.9]. This completes the proof.

We can use a nonsingular closed point on the reduced locally planar curve C_0 to reduce to the case where both \phi, \phi' are nondegenerate stability conditions of degree 0 for C_0.

Under the assumption of Theorem 2.13, for any closed fiber C_b of the family C \to B there is a multiplicative perverse filtration

so that the associated graded {\mathbb{H}}_{C_b}^{\phi_b} is a bigraded {\mathbb{Q}}-algebra; for two nondegenerate stability conditions \phi, \phi', we further have an isomorphism of bigraded {\mathbb{Q}}-algebras:

Therefore, to prove Theorem 0.8, it suffices to check that for any reduced locally planar curve C_0, there is a \delta-regular family C\to B containing C_0 as a closed fiber, such that

1. (i) any nondegenerate stability condition of degree 0 for C_0 can be extended to a nondegenerate stability condition for C\to B,

2. (ii) the relative fine compactified Jacobian associated with the nondegenerate stability condition given by (i) is nonsingular, and the morphism to the base has full support.

Indeed, both (i) and (ii) are guaranteed by taking a versal deformation C\to B of C_0 as in [MSV, Facts 2.3, 2.4, and Theorem 2.12], and the proof is complete.

Since the convolution kernel {\mathcal K}^d in (?) is shown in [MSY, Proposition 3.2 and Section 4.4] to be supported in codimension g, the component {\mathfrak{C}}^d_{\mathrm{red}} is simply given by the fundamental class of the codimension g support of {\mathcal K}^d. On the other hand, [MSY, Corollary 4.5] shows that étale locally the {\mathcal K}^d only differ by tensoring a line bundle, and hence have the same support. In particular, the class {\mathfrak{C}}^d_{\mathrm{red}} is independent of d.

Essentially all three parts are consequences of the Fourier vanishing. We begin with part (i). In view of the realization Theorem 1.2, for the first statement of (i) it suffices to show

where {\mathfrak{q}}^d_{\geq l + 1} is defined in (?). Expanding the left-hand side of both equations

we see that (Proof) (resp. (Proof)) follows directly from (FV2) (resp. (FV1)). The second statement of (i) is parallel.

To prove (?) we first observe that

hence ({\mathfrak{F}}^d)^{-1}_{2g - k} descends to a well-defined morphism

Similarly, {\mathfrak{F}}^d_k descends to a well-defined morphism

To show that they are inverse to each other, we consider the identities

where the second identity holds both by (FV1) and for dimension reasons. Now the condition i > 2g - k translates to 2g - i < k, which implies that the entire second term in (?) factors through

Therefore we have

The other direction is parallel. This proves (i).

For (ii) we consider via (?) the identity

For dimension reasons we then have

Again, in view of the realization Theorem 1.2, to prove (ii) it suffices to show

where {\mathfrak{q}}^\vee_{\geq k + l + 1 - 2g} is defined in (?). We expand *_{\mathrm{red}} using (?), and find

where the second identity uses (FV2). This time the conditions i \geq 2g - k and j \geq 2g - l translate to

But then we have

for such i, j, k, l by (FV2). This proves (ii).

For (iii) we consider the following diagram

where the top and bottom rows are given by (?) and (?), respectively, and the vertical arrows are isomorphisms by (?). We claim that the diagram commutes. Indeed, decomposing

according to the codimension, we have *^d_g = *_{\mathrm{red}}. There are identities

where the second identity uses both (FV1) and dimension constraints. We observe that (?) contains the term

and all the other terms satisfy

By (i), this means that all the other terms in (?) factor through

and hence descend to zero. This shows the commutativity of the diagram (?).

Finally, for any integer d, we conclude from (i) that there are mutually inverse graded isomorphisms

Moreover, by the commutative diagram (?), the above isomorphisms are compatible with the reduced convolution product \overline{*}_{\mathrm{red}} on the left-hand side, and the cup-product \overline{\cup} on the right-hand side. This proves (iii) since by Proposition 1.4, the product \overline{*}_{\mathrm{red}} on the left-hand side is independent of d. The proof of Theorem 1.5 is now complete.

The proof is essentially identical to the one given for compactified Jacobians associated with integral curves in Section ?. For a dualizable abelian fibration

we show that the graded ring object {\mathcal H}_M is isomorphic via the Fourier transform to \widetilde{{\mathcal H}}_{M^\vee} endowed with the convolution product induced by {\mathfrak{C}}_{\mathrm{red}}, as in (?). The proof of Theorem 1.5 works verbatim, since the only geometric inputs required are

1. (i) the dimension bound for the support of the convolution kernel {\mathcal K},

2. (ii) the Fourier vanishing, and

3. (iii) the full support property of \pi, \pi^\vee,

all of which remain valid in the general setting. The rest of the argument is entirely formal.

The proof is parallel to that of [MSY, Proposition 3.2]; the arguments there need to be generalized from integral curves to reduced curves, and this was already worked through in [MRV2] (in order to establish Theorem 2.9). For the reader’s convenience, in the following, we explain in more detail that the \delta-regularity assumption not only guarantees that the support of {\mathcal K} is of codimension g as in [MSY, Proposition 3.2], but also determines the support completely as stated here in the proposition.

We consider

where the p_{ijk}, p_{ij} are the natural projections from \overline{J}_C^{\phi'} \times_B \overline{J}_C^{\phi'}\times_B \overline{J}_C^{\phi'} \times_B \overline{J}_C^\phi. The convolution kernel (?) is governed by {\mathcal K}_3; it suffices to show that the restriction of {\mathcal K}_3 over B\setminus B^\circ has codimension >g.

We consider b\in B\setminus B^\circ (in particular, \delta(b) >0), and a point

By an identical argument as in the proof of [MSY, Proposition 3.2], we have

where C^{\mathrm{reg}}_b \subset C_b is the regular locus; we note that [MSY, Proposition 3.2] only handles integral locally planar curves, but the proof of [MRV2, Proposition 6.3] generalizes all the relevant arguments to reduced ones.

Next, we recall the natural action

It is a smooth morphism of relative dimension g. Hence we have

Let

be the composition of the natural inclusion \mu_b^{-1}(Z_b) \hookrightarrow J_{C_b}^{\underline{0}}\times (\overline{J}_{C_b}^{\phi'_b} )^{\times 3}, and the projection to the second factor; every closed fiber of \sigma_b lies in the irreducible group variety J^{\underline{0}}_{C_b}. By the proof of [A2, Corollary 7.6], (?) implies that any closed fiber of \sigma_b has dimension \leq \delta(b), while a fiber over any closed point lying in the Zariski dense open subset

is 0-dimensional. In particular, for each b\in B we have

which by (?) further implies

Consequently, we deduce from the \delta-regularity of C\to B that there is no irreducible codimension g component contained in the support of {\mathcal K}_3 over B\setminus B^\circ. This completes the proof of the proposition.

The proof is identical to that in [MSY, Section 3.5] for integral curves. The \delta-regularity is used here in a parallel way as in the proof of Proposition 2.10. Indeed, instead of {\mathcal K}_3, we consider another sequence of objects

where i: \overline{J}^{\phi'}_C \times_B \overline{J}^{\phi'}_C \hookrightarrow \overline{J}^{\phi'}_C \times \overline{J}^{\phi'}_C is the closed embedding. These objects are supported on \overline{J}^{\phi'}_C \times_B \overline{J}^{\phi'}_C, and we can deduce the codimension estimate

by a similar proof as for Proposition 2.10 using \delta-regularity. Then an argument via the Adams operations as in [MSY, Section 3.5.2] yields the desired (FV1).

The other half (FV2) can be deduced similarly; see [MSY2, Proposition 2.9].

For (i), we first note that

is a dualizable abelian fibration; here we add subscripts for the Poincaré sheaf {\mathcal P} and the convolution kernel {\mathcal K} to indicate the dependence on the stability conditions. Indeed, as we mentioned before Proposition 2.12, (a1, a2) of Definition 2.2 are satisfied, and (b) is also satisfied by the assumption of the theorem. Therefore, we obtain (i) from Theorem 2.5 and Proposition 2.12.

Now we prove (ii). By setting \phi = \phi' in (Proof), we obtain another dualizable abelian fibration

Proposition 2.10 implies that the reduced convolution classes associated with {\mathcal K}_{\phi\phi'} and {\mathcal K}_{\phi'\phi'} respectively coincide; both are given by

Therefore, we obtain (ii) by applying Theorem 2.6 to (Proof) and (?).

By [FV], the Chow group \mathrm{CH}^1(J_{g,n}) is spanned by tautological classes. Since \overline{J}_{g,n}^{d,\phi} is nonsingular, the excision sequence implies that R^1(\overline{J}_{g,n}^{d,\phi}) = \mathrm{CH}^1(\overline{J}_{g,n}^{d,\phi}). By the Lefschetz (1,1) theorem, it suffices to show that H^{2,0}(\overline{J}_{g,n}^{d,\phi}) = 0.

The rational Abel–Jacobi map \overline{\mathcal{M}}_{g,n+g}\dashrightarrow \overline{J}_{g,n}^{d,\phi} with respect to the last g markings admits a resolution a: \overline{\mathcal{M}}_{g,n+g}'\to\overline{J}_{g,n}^{d,\phi} via a birational modification \overline{\mathcal{M}}_{g,n+g}' \to \overline{\mathcal{M}}_{g,n+g} with \overline{\mathcal{M}}_{g,n+g}' nonsingular. Since a is proper and surjective, the pullback map

is injective. As birational transformations do not affect H^{2, 0}(-), the vanishing of H^{2, 0}(\overline{J}^{d,\phi}_{g,n}) follows from the fact that H^{2,0}(\overline{\mathcal{M}}_{g,n+g}) = 0 by [AC, Theorem 2.2].

Part (i) is from [BMP, Proposition 6.4]. When d=0, (ii) is from [Pixton, Sp]. For general d \in {\mathbb{Z}}, we consider the isomorphism given by twisting by the first marking

Then (ii) follows from the functoriality of \mathsf{uniDR}^c_g(b;\mathsf{a}) under \tau_d^*; see [BHPSS, Section 7.4]. Part (iii) is from [BHPSS, Theorem 0.8].

Since \pi_{i*}(\Theta^g)=g!, we have

where the second equality follows from (?) and the last equality follows from the perversity bound of Proposition 3.2 and the vanishing given by Lemma 3.3. We conclude that a^g = c.

By a similar argument using the perversity bound of Proposition 3.2, we have

Therefore s=0.

We first prove that [\mathsf{DR}^1_g(b;a_1)]_{b^2} is a nonzero class in H^2(\overline{\mathcal{M}}_{g,1},{\mathbb{Q}}). Restricting to {\mathcal M}_{g,1}\subset \overline{\mathcal{M}}_{g,1}, we have

Let p : {\mathcal M}_{g,1} \to {\mathcal M}_g be the forgetful map. Then we have

Therefore, we deduce from Theorem 3.6 that

On the other hand, by (?) combined with (?), we obtain that

from which we deduce

Here we applied Lemma 3.7 and (Proof). The desired identity is concluded.

We first prove that

We pull back the relation (?) to \overline{J}_{g,1}^{0,\phi_i} and then push it forward to \overline{\mathcal{M}}_{g,1}. The left-hand side calculates (Proof). Now we consider the right-hand side: The first term does not contribute since the stability condition forces d = d' = 0. By dimension considerations, the only potentially contributing term from the second term is when g_1 = 0; however, this term also vanishes because its coefficient is a multiple of g_1. Therefore, (Proof) is concluded.

By (?), we obtain

Expanding the right-hand side, only the terms

contribute by (?) and Lemma 3.3. We conclude the desired formula for \beta' by comparing the coefficients of the terms in (?) in the equation

combined with Lemma 3.8, (Proof), and the fact

given by the projection formula.

By (?) and Lemma 3.7, we obtain

Expanding the right-hand side, only the terms

contribute by (?) and Lemma 3.3.

The second term of (Proof) simplifies to

where the first equality follows from the projection formula and the second follows from (Proof) and Lemma 3.9. By Theorem 3.6, the third term of (Proof) equals to

We conclude the desired formula by comparing the coefficients of the terms in (Proof) and (Proof) respectively.

The genus g universal quasistable curve over \overline{J}^{-m,\widetilde{\phi}}_{\widetilde{\Gamma}_\delta} induces a morphism

For each w\in V(\Gamma), we have the natural projection

Consider the following relation in \mathrm{CH}^*(\mathfrak{Pic}_{g,n,d}):

When w \in V(\Gamma) is a stable vertex, then \varphi^*p_w^* \Theta = \Theta[w]. Therefore, it is enough to consider the case when w is an unstable vertex of \Gamma. When w is an unstable vertex, the degree at w should be 1 by the stability condition. The relation \mathsf{uniDR}^1_0(0;a+1,-a)=0 on \mathfrak{Pic}_{0,2,1} yields

After pulling back this relation along p_w\circ \varphi, the other terms do not contribute due to the stability conditions. Therefore, the result holds for m=1. The general case of (i) follows by a similar argument.

The argument for (ii) is parallel. It suffices to consider \kappa_{0,1}[w] with w an unstable vertex of \Gamma, and then it follows from the relation \mathsf{uniDR}^1_0(b;a, -a) = 0 on \mathfrak{Pic}_{0,2,1}.

Part (iii) follows from the relation \mathsf{uniDR}^1_0(a+1,-a)=0 on \mathfrak{Pic}_{0,2,-1}; see [BMP, (34)].

Consider the stable graph \widetilde{\Gamma} with g vertices labeled u_0, u_1, \dots ,u_{g-1}, where g(u_0)=0 and g(u_1)=\cdots=g(u_{g-1})=1. For i=0, \dots, g, the vertex u_i is connected to u_{i-1} by one edge (with u_g := u_0), and the first marking is attached to u_0.

We show that the values \phi(z)(u_1), \dots, \phi(z)(u_{g-1}) are all equal. For each 1 \le i \le g-2, consider the contraction of graphs \widetilde{\Gamma} \to \widetilde{\Gamma}_i in which the vertices u_k for k \ne i,i+1 are contracted to u_0. The graph \widetilde{\Gamma}_i admits a symmetry interchanging the vertices u_i and u_{i+1}. Since \phi(z) is compatible with both contractions and the symmetries of stable graphs, we obtain

Therefore we have \phi(z) (u_1) = \cdots = \phi(z) (u_{g-1}).

On the other hand, there is a contraction \widetilde{\Gamma} \to \Gamma_0 contracting the vertices u_i for 1 \le i \le g-1 to a single vertex. Compatibility of \phi(z) with contractions then yields

Since \phi(z) is a stability condition of degree 0, we have

and the desired equality follows from the compatibility of \phi(z) with contractions.

The universal double ramification cycle relations provide an algorithm to calculate the classes of the form

which leads to the proof of both identities above. Since the full proof is lengthy but the idea is straightforward, we summarize the computational scheme here and leave the details to Section ?.

A class (Proof) must be a multiple of the class [\Gamma_{g_1}], and it suffices to determine the coefficient. First, we use Lemma 3.11 to distribute the monomial \Theta^{k}\cup \kappa^l_{0,1} among the vertices of \Gamma_{g_1}^\pm; then we apply the relations in Section ? at each vertex. For dimension reasons, the only terms that can contribute nontrivially to the pushforward are those corresponding to the stratum \Gamma_{g_1}', decorated by monomials of \Theta divisors on stable vertices. These terms can be computed by successive applications of the universal double ramification cycle relations.

We multiply \tfrac{1}{3}\kappa_{0,1} on each side of (?). By (Proof) and the projection formula, the first term vanishes:

The third and the fourth terms also do not contribute to the pushforward. To see this, by the vanishing Lemma 3.13(ii), we have for any g_1 \ge 1 the vanishing

When g_1 = 0, the corresponding term vanishes because the coefficient is a multiple of g_1. A similar argument shows that the fourth term also does not contribute to the pushforward.

Thus only the second term contributes to the pushforward. By Lemma 3.13 and (?), the contribution from the stratum \Gamma_{g_1}^+ pushes forward to

Since \pi^*[\Gamma_{g_1}] = [\Gamma_{g_1}^+] + [\Gamma_{g_1}^-], the projection formula implies

Therefore, by (?), the contribution from the stratum \Gamma_{g_1}^- is

Adding the two contributions yields the formula for a_{g_1}(\phi(z)).

We proceed by explicit intersection theory. When g_1\neq i, the vanishing

holds for dimension reasons. On the other hand, we have

which is nonzero by [FP, (28)].

By a general result of Raynaud [Ray, Corollary XIII 2.4 and Proposition XIII 2.6], for any regular and quasicompact base B and any abelian scheme A \to B, the first étale cohomology H^1_{\text{\'et}}(B, A) is a torsion group. This implies the existence of a positive integer N for which the pushforward torsor

via the “multiplication by N” map [N]: A \to A is isomorphic to the trivial torsor A. Here the A-action on the second factor A is defined through [N].

The natural morphism T \to [N]_*T then yields a morphism

which is fiberwise an isogeny up to translation over B.

We consider the morphism of relative Chow motives

Since (?) is induced by the pullback of a morphism, it automatically preserves the cup-product, and it is an isomorphism by the conservativity result [AHPL, Lemma A.6].

More precisely, we observe that it is enough to prove the isomorphism after base change to the algebraic closure of the base field. When the base field is algebraically closed (or more generally a perfect field), the category of relative Chow motives over B, in the sense of either [DM] or [CH], admits a fully faithful embedding in the category of Beilinson motives \mathrm{DM}_{\mathord{\mathchoice {\mbox{\fontsize\tf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} {\mbox{\fontsize\tf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} {\mbox{\fontsize\sf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} {\mbox{\fontsize\ssf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} }}(B) of [CD]; see e.g. [Jin]. By the comparison theorem [CD, Theorem 16.2.18] between \mathrm{DM}_{\mathord{\mathchoice {\mbox{\fontsize\tf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} {\mbox{\fontsize\tf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} {\mbox{\fontsize\sf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} {\mbox{\fontsize\ssf@size\z@\usefont{T2A}{\rmdefault}{m}{n}\CYRB}} }}(B) and the category of étale motives \mathrm{DA}_{\text{\'et}}(B, {\mathbb{Q}}) with \mathbb{Q}-coefficients (and without transfers), and by [AHPL, Lemma A.6], it suffices to verify that (?) is an isomorphism when pulled back to any geometric point of B. The latter statement is immediate since \varphi is fiberwise an isogeny up to translation.

Finally, the last claim follows from the construction of the multiplicative Chow–Künneth decomposition for abelian schemes in [DM].