Introduction
Non-abelian Hodge correspondence for Riemann surfaces
Let C C C be a compact Riemann surface. The non-abelian Hodge correspondence, established through the work of Hitchin [Hit87], Donaldson [Don87], Corlette [Cor88], and Simpson [Sim92], gives a homeomorphism between the moduli space of semistable Higgs bundles of degree zero and the moduli space of semistable flat connections. This groundbreaking work has led to major developments in related fields of mathematics, most notably the study of higher Teichmüller theory through Higgs bundles [Hit92], and the P = W P = W P = W conjecture formulated in [dCHM12], now proved for G L n \mathrm{GL}_n GL n in [MS24] and [HMM26].
A logarithmic version of this non-abelian Hodge correspondence was established by Simpson [Sim90]. In order to incorporate arbitrary residue eigenvalues for logarithmic connections, one needs to introduce parabolic structures on the Higgs bundle side of the correspondence. Simpson’s theorem produces a correspondence between parabolic connections and parabolic Higgs bundles with prescribed residue eigenvalues, parabolic weights and filtration types. The relationship between the parabolic weights and residue eigenvalues on the two sides is encoded in a table, see [Sim90, p. 720].
Non-abelian Hodge correspondence for curves in positive characteristic
Let k k k be an algebraically closed field of characteristic p > 0 p > 0 p > 0 . Compared to the case over the complex numbers, a substantial feature of the characteristic p > 0 p > 0 p > 0 setting is that the algebra of crystalline differential operators has a large center. In fact, it is an Azumaya algebra (i.e. a sheaf of algebras étale-locally isomorphic to a matrix algebra) over the Frobenius twist of the cotangent bundle of the underlying space, see [BMR08]. Based on this Azumaya property, Ogus and Vologodsky [OV07] established a non-abelian Hodge correspondence assuming certain nilpotence and lifting assumptions. Over a smooth projective curve, Groechenig extended Tsen’s theorem to show that the Azumaya algebra of differential operators splits on all spectral curves, thereby establishing a non-abelian Hodge correspondence over the entire Hitchin base; see [Gro16]. The correspondence was generalized to arbitrary reductive groups in [CZ15] under suitable characteristic assumptions. In [dCGZ24], the notion of very good splittings is introduced to obtain a correspondence between semistable Higgs bundles and connections. The matching of semistable objects for general reductive groups is established in [FHZ26].
The generalization of this non-abelian Hodge correspondence over curves to the logarithmic setting was initiated in [She24]. Working over G L n \mathrm{GL}_n GL n , [She24] established a correspondence under a certain regularity condition on the residue eigenvalues for logarithmic connections, see Theorem 2.1. With the same regularity assumption, the matching of semistable objects is established in [dCZ25], using a similar “very good” notion as in [dCGZ24]. Similar to the characteristic zero setting, going beyond this regularity assumption requires the introduction of parabolic structures on the Higgs bundle side, and corresponding results appear in [LS24] and [Bar25].
Main results
Let C C C be a smooth projective curve over an algebraically closed field k k k of characteristic p > 0 p > 0 p > 0 . All theorems in this paper admit straightforward generalizations to any reduced divisor D ⊂ C D \subset C D ⊂ C . We restrict to the case of a single point q ∈ C q \in C q ∈ C for simplicity of exposition. The main goal of this paper is to establish a non-abelian Hodge correspondence between rank n n n semistable parabolic connections on C C C and rank n n n semistable parabolic Higgs bundles on its Frobenius twist C ′ C' C ′ , allowing arbitrary prescribed residue eigenvalues, parabolic weights and filtration types, with no restriction on p p p . The relationship between the parabolic weights and residue eigenvalues on the two sides is encoded in the following table analogous to Simpson’s table in [Sim90].
Table 1.
Here ν \nu ν is a fixed map from k k k to { − p + 1 , … , − 1 , 0 } \{-p+1,\ldots,-1,0\} { − p + 1 , … , − 1 , 0 } whose restriction to each fiber of the Artin–Schreier map b ↦ b p − b b \mapsto b^p-b b ↦ b p − b is a bijection; see § 3 for its precise definition. Since parabolic weights lie in [ 0 , 1 ) [0,1) [ 0 , 1 ) , the above table offers a bijection between pairs ( a , α ) (a,\alpha) ( a , α ) and ( b , β ) (b,\beta) ( b , β ) in k × [ 0 , 1 ) k \times[0,1) k × [ 0 , 1 ) .
Before giving the precise statement of our main theorem, we introduce the following notation. Let a \boldsymbol{a} a be an unordered n n n -tuple of elements in k k k , and let α \boldsymbol{\alpha} α be a set of parabolic weights. We introduce the notion of an assignment a ↦ α \boldsymbol{a} \mapsto\boldsymbol{\alpha} a ↦ α , which assigns a parabolic weight in α \boldsymbol{\alpha} α to each entry of a \boldsymbol{a} a , counted with multiplicity; see Theorem 4.2. Let H i g g s a ↦ α ( C ) \mathit{Higgs}^{\boldsymbol{a}\mapsto\boldsymbol{\alpha}}(C) Higgs a ↦ α ( C ) (resp. C o n n a ↦ α ( C ) \mathit{Conn}^{\boldsymbol{a}\mapsto\boldsymbol{\alpha}}(C) Conn a ↦ α ( C ) ) be the moduli stack of parabolic Higgs bundles (resp. connections) whose residue at q q q acts on each graded quotient with eigenvalues prescribed by a ↦ α \boldsymbol{a} \mapsto\boldsymbol{\alpha} a ↦ α ; see Theorem 4.3. Let a ↦ α \boldsymbol{a} \mapsto\boldsymbol{\alpha} a ↦ α and b ↦ β \boldsymbol{b} \mapsto\boldsymbol{\beta} b ↦ β be two assignments related through the table above; see Section 5.1 for the precise relation. The moduli stack H i g g s a ↦ α ( C ′ ) \mathit{Higgs}^{\boldsymbol{a}\mapsto\boldsymbol{\alpha}}(C') Higgs a ↦ α ( C ′ ) admits the Hitchin morphism h ′ h' h ′ , and C o n n b ↦ β ( C ) \mathit{Conn}^{\boldsymbol{b}\mapsto\boldsymbol{\beta}}(C) Conn b ↦ β ( C ) admits the de Rham–Hitchin morphism h d R h_{\mathrm{dR}} h dR to the same Hitchin base:
Figure 1.
Our main theorem establishes an isomorphism between C o n n b ↦ β ( C ) \mathit{Conn}^{\boldsymbol{b}\mapsto\boldsymbol{\beta}}(C) Conn b ↦ β ( C ) and H i g g s a ↦ α ( C ′ ) \mathit{Higgs}^{\boldsymbol{a}\mapsto\boldsymbol{\alpha}}(C') Higgs a ↦ α ( C ′ ) up to twisting by a certain Picard torsor, and parabolic semistability is preserved under this correspondence.
Let b b b be a regular tuple in the sense of Theorem 2.1 that maps to a a a under the componentwise Artin–Schreier map. The moduli stack H b \mathcal{H}_b H b of regular¹ logarithmic connections with eigenvalues of residue b b b introduced in [She24] plays a crucial role in the logarithmic non-abelian Hodge correspondence per se, as well as in subsequent works [LS24, Bar25] and [dCZ25]. In particular, it was shown in [She24] that when p > n p > n p > n , H b \mathcal{H}_b H b maps surjectively to the corresponding Hitchin base A a A_a A a , and is a torsor for the relative Picard stack P a \mathcal{P}_a P a of spectral curves over A a A_a A a . We provide a slightly different local construction in this paper to remove this restriction on the characteristic; see Theorem 2.2.
In order to match semistable objects in the parabolic setting, we generalize the notion of “very good regular connections” introduced in [dCGZ24] and [dCZ25] by incorporating a residue-dependent degree condition; see Theorem 3.1. We show that this “very good” substack H b v g ⊂ H b \mathcal{H}_b^{\mathrm{vg}} \subset\mathcal{H}_b H b vg ⊂ H b maps surjectively to A a A_a A a and is a P a 0 \mathcal{P}_a^0 P a 0 -torsor, where P a 0 ⊂ P a \mathcal{P}_a^0 \subset\mathcal{P}_a P a 0 ⊂ P a is the substack parametrizing line bundles having degree zero on each integral component; see Theorem 3.3. Up to canonical equivalence via Hecke transformations, this P a 0 \mathcal{P}_a^0 P a 0 -torsor H b v g \mathcal{H}_b^{\mathrm{vg}} H b vg is independent of the choice of a regular Artin–Schreier preimage b b b , and we denote it by H a v g \mathcal{H}_a^{\mathrm{vg}} H a vg .
Now we are ready to state the main theorem of the paper. Note that the tuple b b b underlying b ↦ β b \mapsto\beta b ↦ β need not be regular. Let H i g g s a → α , s s ( C ′ , d ) ⊂ H i g g s a → α ( C ′ ) \mathop{\mathrm{Higgs}}\nolimits^{a\to\alpha,\mathrm{ss}}(C',d) \subset\mathop{\mathrm{Higgs}}\nolimits^{a\to\alpha}(C') Higgs a → α , ss ( C ′ , d ) ⊂ Higgs a → α ( C ′ ) (resp. C o n n b → β , s s ( C , d ) ⊂ C o n n b → β ( C ) \mathop{\mathrm{Conn}}\nolimits^{b\to\beta,\mathrm{ss}}(C,d) \subset\mathop{\mathrm{Conn}}\nolimits^{b\to\beta}(C) Conn b → β , ss ( C , d ) ⊂ Conn b → β ( C ) ) be the substack classifying α \alpha α -semistable (resp. β \beta β -semistable) objects of parabolic degree d d d . The following is proved in Theorem 5.3 and Theorem 5.2.
Theorem 1.1. Let a ↦ α a \mapsto\alpha a ↦ α and b ↦ β b \mapsto\beta b ↦ β be two assignments related through the table above. There exists an isomorphism of A a A_a A a -stacks
H a v g × P a 0 H i g g s a → α ( C ′ ) ≅ C o n n b → β ( C ) . \mathcal{H}_a^{\mathrm{vg}} \mathbin{\times}_{\mathcal{P}_a^0} \mathop{\mathrm{Higgs}}\nolimits^{a\to\alpha}(C') \cong\mathop{\mathrm{Conn}}\nolimits^{b\to\beta}(C). H a vg × P a 0 Higgs a → α ( C ′ ) ≅ Conn b → β ( C ) . For any d ∈ R d \in\mathbb{R} d ∈ R , this isomorphism restricts to an isomorphism
H a v g × P a 0 H i g g s a → α , s s ( C ′ , d ) ≅ C o n n b → β , s s ( C , p d ) . \mathcal{H}_a^{\mathrm{vg}} \mathbin{\times}_{\mathcal{P}_a^0} \mathop{\mathrm{Higgs}}\nolimits^{a\to\alpha,\mathrm{ss}}(C',d) \cong\mathop{\mathrm{Conn}}\nolimits^{b\to\beta,\mathrm{ss}}(C,pd). H a vg × P a 0 Higgs a → α , ss ( C ′ , d ) ≅ Conn b → β , ss ( C , p d ) .
A key ingredient in the proof of Theorem 1.1 is the establishment of a Cartier descent formalism for parabolic connections, building on work of Li and Sun [LS24]. Let τ \tau τ be an unordered n n n -tuple of elements in F p \mathbb{F}_p F p . We denote by C o n n 0 τ → β ( C ) ⊂ C o n n τ → β ( C ) \mathop{\mathrm{Conn}}\nolimits_0^{\tau\to\beta}(C) \subset\mathop{\mathrm{Conn}}\nolimits^{\tau\to\beta}(C) Conn 0 τ → β ( C ) ⊂ Conn τ → β ( C ) the substack of connections with zero p p p -curvature, and by B u n α ( C , m ′ ) \mathop{\mathrm{Bun}}\nolimits^\alpha(C,\mathbf{m}') Bun α ( C , m ′ ) the moduli stack of parabolic bundles with parabolic weights α \alpha α and filtration type m ′ \mathbf{m}' m ′ . The following is proved in Theorem 5.7.
Proposition 1.2. For any assignment τ ↦ β \tau\mapsto\beta τ ↦ β , there exists an isomorphism
C o n n 0 τ → β ( C ) ≅ B u n α ( C , m ′ ) , \mathop{\mathrm{Conn}}\nolimits_0^{\tau\to\beta}(C) \cong\mathop{\mathrm{Bun}}\nolimits^\alpha(C,\mathbf{m}'), Conn 0 τ → β ( C ) ≅ Bun α ( C , m ′ ) , where the parabolic weights α \alpha α and filtration type m ′ \mathbf{m}' m ′ arise from the assignment α ↦ α \alpha\mapsto\alpha α ↦ α constructed from τ ↦ β \tau\mapsto\beta τ ↦ β through the table above.
A crucial step towards the proof of Theorem 1.2 is to show that every object in C o n n 0 τ → β ( C ) \mathop{\mathrm{Conn}}\nolimits_0^{\tau\to\beta}(C) Conn 0 τ → β ( C ) is locally isomorphic to a certain model object; see Theorem 5.5. This is a parabolic analogue of [LS24, Lemma 5.21], and we offer an independent proof using Lagrange interpolation polynomials.
Utilizing the correspondence between parabolic bundles and filtered sheaves, we obtain a particularly clean description of this Cartier descent formalism, as stated in the following theorem proved in Theorem 6.4 .
Theorem 1.3. Under the isomorphism in Theorem 1.2 , a filtered connection E = ( { E x } x ∈ R , ∇ E ) \mathcal{E} = (\{\mathcal{E}_x\}_{x \in\mathbb{R}}, \nabla^{\mathcal{E}}) E = ({ E x } x ∈ R , ∇ E ) corresponds to the filtered sheaf F = { F x } x ∈ R \mathcal{F} = \{\mathcal{F}_x\}_{x \in\mathbb{R}} F = { F x } x ∈ R defined by
F x : = ( F r C ) ∗ ker ( ∇ d R E : E d R ⟶ E d R ⊗ ω C ( q ) ) , (1) \mathcal{F}_x := (\mathrm{Fr}_C)_* \ker\left(\nabla^{\mathcal{E}}_{\mathrm{dR}} : \mathcal{E}_{\mathrm{dR}} \longrightarrow\mathcal{E}_{\mathrm{dR}} \otimes\omega_C(q)\right), \tag*{(1)} F x := ( Fr C ) ∗ ker ( ∇ dR E : E dR ⟶ E dR ⊗ ω C ( q ) ) , (1) where F r C : C ⟶ C ′ \mathrm{Fr}_C : C \longrightarrow C' Fr C : C ⟶ C ′ is the relative Frobenius.
The inverse of this construction is provided in Theorem 6.6. Note that this formulation of Cartier descent for filtered sheaves through (1) treats all choices of residue eigenvalues in F p \mathbb{F}_p F p and parabolic weights simultaneously, and the description involves neither root stacks nor parahoric group schemes. This formulation also potentially extends to higher dimensional varieties. We plan to investigate this generalization in future work.
Organization of the paper
In Section 2, we review various constructions related to logarithmic Higgs bundles and connections. We show the substack of regular connections maps surjectively to the corresponding Hitchin base, and is a torsor for the relative Picard stack of spectral curves. In Section 3, we first introduce the residue-dependent degree condition that defines the substack H b v g \mathcal{H}^{\mathrm{vg}}_b H b vg of “very good logarithmic connections”. Then we show H b v g \mathcal{H}^{\mathrm{vg}}_b H b vg is a P a o \mathcal{P}^o_a P a o -torsor independent of the choice of the Artin–Schreier preimage b b b , up to canonical equivalence. In Section 4, we first review the definition of parabolic bundles, Higgs bundles, and connections, as well as their parabolic semistability conditions. Then we introduce the notion of an assignment of parabolic weights to eigenvalues. In Section 5, we state and prove the main parabolic non-abelian Hodge correspondence Theorem 1.1 of the paper, after introducing the analogue of Simpson’s table that encodes the relationship between the parabolic weights and residue eigenvalues on the two sides. We also establish the parabolic Cartier descent formalism Theorem 1.2 . In Section 6, working with filtered sheaves, we provide another description of this Cartier descent formalism by proving Theorem 1.3 .
Acknowledgments
The author thanks Siqing Zhang for contributing to the proof of Theorem 3.3, and for helpful discussions on related subjects. The author was partially supported by the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No. 101034413 and the Austrian Science Fund (FWF), grant 10.55776/P35847.
Declaration of AI usage
The author used GPT-5.6 sol (very high), GPT-6 (very high) and Claude Opus 5.5 (extra) to proofread and improve the exposition of the manuscript. The manuscript was written entirely by the author, who takes full responsibility for its content.
Logarithmic Higgs bundles and logarithmic connections
Logarithmic Higgs bundles
Let k k k be an algebraically closed field. Let C C C be a smooth projective curve over k k k , and let q ∈ C q \in C q ∈ C be a marked point. Let ω C \omega_C ω C be the sheaf of differentials on C C C . A logarithmic Higgs bundle over ( C , q ) (C,q) ( C , q ) is a pair ( E , ϕ ) (E,\phi) ( E , ϕ ) where E E E is a vector bundle over C C C and ϕ \phi ϕ is an O C \mathcal{O}_C O C -linear morphism
ϕ : E ⟶ E ⊗ ω C ( q ) . \phi: E \longrightarrow E \otimes\omega_C(q). ϕ : E ⟶ E ⊗ ω C ( q ) .
Let H i g g s ( C , n ) \mathcal{Higgs}(C,n) H i g g s ( C , n ) be the moduli stack of rank n n n logarithmic Higgs bundles over ( C , q ) (C,q) ( C , q ) . For a k k k -scheme S S S , H i g g s ( C , n ) ( S ) \mathcal{Higgs}(C,n)(S) H i g g s ( C , n ) ( S ) is the groupoid of pairs ( E , ϕ ) (E,\phi) ( E , ϕ ) , where E E E is a rank n n n vector bundle on C × S C \times S C × S and ϕ \phi ϕ is an O C × S \mathcal{O}_{C \times S} O C × S -linear morphism
ϕ : E ⟶ E ⊗ p C ∗ ω C ( q ) , \phi: E \longrightarrow E \otimes p_C^*\omega_C(q), ϕ : E ⟶ E ⊗ p C ∗ ω C ( q ) ,
where ρ C : C × S → C \rho_C:C\times S\to C ρ C : C × S → C is the projection map. By taking the coefficients of the characteristic polynomial of ϕ \phi ϕ , we obtain the so-called Hitchin morphism, which we denote by
h : H i g g s ( C , n ) ⟶ A ( ω C ( q ) ) : = ⨁ i = 1 n H 0 ( C , ω C i ( i q ) ) . h:\mathit{Higgs}(C,n)\longrightarrow A(\omega_C(q)):=\bigoplus_{i=1}^{n}H^0(C,\omega_C^i(iq)). h : Higgs ( C , n ) ⟶ A ( ω C ( q )) := i = 1 ⨁ n H 0 ( C , ω C i ( i q )) .
Residues of logarithmic Higgs bundles
Recall that the fiber ω C ( q ) ∣ q \omega_C(q)|_q ω C ( q ) ∣ q carries a canonical trivialization induced by the residue map
ω C ( q ) ∣ q → ∼ O q ≅ k . \omega_C(q)|_q\xrightarrow{\sim}\mathcal{O}_q\cong k. ω C ( q ) ∣ q ∼ O q ≅ k .
For any choice of a local coordinate x x x of C C C vanishing at q q q , this trivialization maps the section d x x \frac{dx}{x} x d x to 1 1 1 . Let ( E , ϕ ) (E,\phi) ( E , ϕ ) be a logarithmic Higgs bundle over ( C , q ) (C,q) ( C , q ) . The residue of ϕ \phi ϕ at q q q
res q ( ϕ ) ∈ H 0 ( q , E n d ( E ) ⊗ ω C ( q ) ) ≅ H 0 ( q , E n d ( E ∣ q ) ) \operatorname{res}_q(\phi)\in H^0(q,\mathcal{E}nd(E)\otimes\omega_C(q))\cong H^0(q,\mathcal{E}nd(E|_q)) res q ( ϕ ) ∈ H 0 ( q , E n d ( E ) ⊗ ω C ( q )) ≅ H 0 ( q , E n d ( E ∣ q ))
is the restriction of ϕ \phi ϕ to q q q , where the second isomorphism is induced by the residue trivialization.
Let t ⊂ g l n \mathfrak{t}\subset\mathfrak{gl}_n t ⊂ gl n be the Cartan subalgebra given by diagonal matrices, and let W W W be the Weyl group acting on t \mathfrak{t} t . Consider the evaluation-at q q q morphism
ev q : A ( ω C ( q ) ) = ⨁ i = 1 n H 0 ( C , ω C i ( i q ) ) ⟶ ⨁ i = 1 n H 0 ( q , ω C i ( i q ) ) → ∼ Γ ( q , O q ⊕ n ) ≅ t ( k ) / W . (2) \operatorname{ev}_q:A(\omega_C(q))=\bigoplus_{i=1}^{n}H^0(C,\omega_C^i(iq))\longrightarrow\bigoplus_{i=1}^{n}H^0(q,\omega_C^i(iq))\xrightarrow{\sim}\Gamma(q,\mathcal{O}_q^{\oplus n})\cong\mathfrak{t}(k)/W. \tag*{(2)} ev q : A ( ω C ( q )) = i = 1 ⨁ n H 0 ( C , ω C i ( i q )) ⟶ i = 1 ⨁ n H 0 ( q , ω C i ( i q )) ∼ Γ ( q , O q ⊕ n ) ≅ t ( k ) / W . (2)
where the isomorphism ⨁ i = 1 n H 0 ( q , ω C i ( i q ) ) → ∼ Γ ( q , O q ⊕ n ) \bigoplus_{i=1}^{n}H^0(q,\omega_C^i(iq))\xrightarrow{\sim}\Gamma(q,\mathcal{O}_q^{\oplus n}) ⨁ i = 1 n H 0 ( q , ω C i ( i q )) ∼ Γ ( q , O q ⊕ n ) is induced by the tensor powers of the residue trivialization ω C ( q ) ∣ q → ∼ O q \omega_C(q)|_q\xrightarrow{\sim}\mathcal{O}_q ω C ( q ) ∣ q ∼ O q , and the last isomorphism is given by the standard identification A n ≅ t / W A^n\cong\mathfrak{t}/W A n ≅ t / W via the coefficients of the characteristic polynomial. The Hitchin morphism induces a morphism
Γ ( q , E n d ( E ∣ q ) ) ⟶ Γ ( q , O q ⊕ n ) ≅ t ( k ) / W \Gamma(q,\mathcal{E}nd(E|_q))\longrightarrow\Gamma(q,\mathcal{O}_q^{\oplus n})\cong\mathfrak{t}(k)/W Γ ( q , E n d ( E ∣ q )) ⟶ Γ ( q , O q ⊕ n ) ≅ t ( k ) / W
given by sending an endomorphism to the coefficients of its characteristic polynomial, which determine and are determined by the eigenvalues of the endomorphism. We parametrize the closed points of t / W \mathfrak{t}/W t / W by unordered n n n -tuples of elements in k k k , written as
a = a 1 n 1 a 2 n 2 ⋯ a s n s , (3) a=a_1^{n_1}a_2^{n_2}\cdots a_s^{n_s}, \tag*{(3)} a = a 1 n 1 a 2 n 2 ⋯ a s n s , (3)
with a j ∈ k , n j ∈ N > 0 a_j\in k,n_j\in\mathbb{N}_{>0} a j ∈ k , n j ∈ N > 0 and ∑ j = 1 s n j = n \sum_{j=1}^{s}n_j=n ∑ j = 1 s n j = n .
Logarithmic connections
Now we assume k k k is of characteristic p > 0 p>0 p > 0 . A logarithmic connection over ( C , q ) (C,q) ( C , q ) is a pair ( E , ∇ ) (E,\nabla) ( E , ∇ ) where E E E is a vector bundle over C C C and ∇ : E → E ⊗ ω C ( q ) \nabla:E\to E\otimes\omega_C(q) ∇ : E → E ⊗ ω C ( q ) is a k k k -linear map that satisfies the Leibniz rule. Let Conn ( C , n ) \operatorname{Conn}(C,n) Conn ( C , n ) be the moduli stack of rank n n n logarithmic connections over ( C , q ) (C,q) ( C , q ) . For a k k k -scheme S S S , Conn ( C , n ) ( S ) \operatorname{Conn}(C,n)(S) Conn ( C , n ) ( S ) is the groupoid of pairs ( E , ∇ ) (E,\nabla) ( E , ∇ ) , where E E E is a rank n n n vector bundle on C × S C\times S C × S and ∇ \nabla ∇ is an O S \mathcal{O}_S O S -linear morphism
∇ : E ⟶ E ⊗ ρ C ∗ ω C ( q ) \nabla:E\longrightarrow E\otimes\rho_C^*\omega_C(q) ∇ : E ⟶ E ⊗ ρ C ∗ ω C ( q )
that satisfies the Leibniz rule.
The de Rham–Hitchin morphism
Let T C \mathcal{T}_{C} T C be the tangent sheaf of C C C . Let C ′ = C × k , F k k C^{\prime}=C\times_{k,F_{k}}k C ′ = C × k , F k k be the Frobenius twist of C C C , and let Fr C : C → C ′ \operatorname{Fr}_{C}:C\to C^{\prime} Fr C : C → C ′ be the relative Frobenius morphism. Let q ′ = Fr C ( q ) ∈ C ′ q^{\prime}=\operatorname{Fr}_{C}(q)\in C^{\prime} q ′ = Fr C ( q ) ∈ C ′ . For any logarithmic connection ( E , ∇ ) (E,\nabla) ( E , ∇ ) , we associate with ∇ \nabla ∇ its p p p -curvature
ψ ∇ : Fr C ∗ ( T C ′ ( − q ′ ) ) ⟶ E n d ( E ) , \psi_{\nabla}:\operatorname{Fr}_{C}^{*}(\mathcal{T}_{C^{\prime}}(-q^{\prime}))\longrightarrow\mathcal{E}nd(E), ψ ∇ : Fr C ∗ ( T C ′ ( − q ′ )) ⟶ E n d ( E ) ,
which is the O C \mathcal{O}_{C} O C -linear morphism corresponding to the p p p -linear morphism
ψ ∇ : T C ( − q ) ⟶ E n d ( E ) , ψ ∇ ( ∂ ) = ∇ ∂ p − ∇ ∂ [ p ] \psi_{\nabla}:\mathcal{T}_{C}(-q)\longrightarrow\mathcal{E}nd(E),\qquad\psi_{\nabla}(\partial)=\nabla_{\partial}^{p}-\nabla_{\partial^{[p]}} ψ ∇ : T C ( − q ) ⟶ E n d ( E ) , ψ ∇ ( ∂ ) = ∇ ∂ p − ∇ ∂ [ p ]
for any ∂ ∈ T C ( − q ) ( U ) \partial\in\mathcal{T}_{C}(-q)(U) ∂ ∈ T C ( − q ) ( U ) and U ⊆ C U\subseteq C U ⊆ C open. The p p p -curvature ψ ∇ \psi_{\nabla} ψ ∇ can also be viewed as a twisted Higgs field
ψ ∇ : E ⟶ E ⊗ Fr C ∗ ( ω C ′ ( q ′ ) ) . \psi_{\nabla}:E\longrightarrow E\otimes\operatorname{Fr}_{C}^{*}(\omega_{C^{\prime}}(q^{\prime})). ψ ∇ : E ⟶ E ⊗ Fr C ∗ ( ω C ′ ( q ′ )) .
The coefficients of the characteristic polynomial of ψ ∇ \psi_{\nabla} ψ ∇ define a point a ∈ A ( Fr C ∗ ( ω C ′ ( q ′ ) ) ) . a\in A(\operatorname{Fr}_{C}^{*}(\omega_{C^{\prime}}(q^{\prime}))). a ∈ A ( Fr C ∗ ( ω C ′ ( q ′ ))) . By the same argument as in [LP01, Proposition 3.2], this point lies in the image of the pullback map Fr C ∗ : A ( ω C ′ ( q ′ ) ) → A ( Fr C ∗ ( ω C ′ ( q ′ ) ) ) . \operatorname{Fr}_{C}^{*}:A(\omega_{C^{\prime}}(q^{\prime}))\to A(\operatorname{Fr}_{C}^{*}(\omega_{C^{\prime}}(q^{\prime}))). Fr C ∗ : A ( ω C ′ ( q ′ )) → A ( Fr C ∗ ( ω C ′ ( q ′ ))) . Thus we obtain a morphism
h d R : Conn ( C , n ) ⟶ A ( ω C ′ ( q ′ ) ) h_{\mathrm{dR}}:\operatorname{Conn}(C,n)\longrightarrow A(\omega_{C^{\prime}}(q^{\prime})) h dR : Conn ( C , n ) ⟶ A ( ω C ′ ( q ′ ))
which we call the de Rham–Hitchin morphism.
Residues of logarithmic connections
Given a logarithmic connection ( E , ∇ ) (E,\nabla) ( E , ∇ ) over ( C , q ) (C,q) ( C , q ) , the residue trivialization of ω C ( q ) \omega_{C}(q) ω C ( q ) gives rise to the residue res q ( ∇ ) ∈ Γ ( q , E n d ( E ∣ q ) ) \operatorname{res}_{q}(\nabla)\in\Gamma(q,\mathcal{E}nd(E|_{q})) res q ( ∇ ) ∈ Γ ( q , E n d ( E ∣ q )) , and the Frobenius pullback of the residue trivialization of ω C ′ ( q ′ ) \omega_{C^{\prime}}(q^{\prime}) ω C ′ ( q ′ ) gives rise to the residue res q ( ψ ∇ ) ∈ Γ ( q , E n d ( E ∣ q ) ) . \operatorname{res}_{q}(\psi_{\nabla})\in\Gamma(q,\mathcal{E}nd(E|_{q})). res q ( ψ ∇ ) ∈ Γ ( q , E n d ( E ∣ q )) . We have the identity of residues in Γ ( q , E n d ( E ∣ q ) ) \Gamma(q,\mathcal{E}nd(E|_{q})) Γ ( q , E n d ( E ∣ q ))
res q ( ψ ∇ ) = res q ( ∇ ) p − res q ( ∇ ) , \operatorname{res}_{q}(\psi_{\nabla})=\operatorname{res}_{q}(\nabla)^{p}-\operatorname{res}_{q}(\nabla), res q ( ψ ∇ ) = res q ( ∇ ) p − res q ( ∇ ) ,
see [She24, Lemma 4.1]. Thus we have the identity of eigenvalues in t ( k ) / W \mathfrak{t}(k)/W t ( k ) / W :
a ( res q ( ψ ∇ ) ) = a ( res q ( ∇ ) ) p − a ( res q ( ∇ ) ) . (4) a(\operatorname{res}_{q}(\psi_{\nabla}))=a(\operatorname{res}_{q}(\nabla))^{p}-a(\operatorname{res}_{q}(\nabla)). \tag*{(4)} a ( res q ( ψ ∇ )) = a ( res q ( ∇ ) ) p − a ( res q ( ∇ )) . (4)
Let AS : t ( k ) / W → t ( k ) / W \operatorname{AS}:\mathfrak{t}(k)/W\to\mathfrak{t}(k)/W AS : t ( k ) / W → t ( k ) / W be the morphism given by a ↦ a p − a a\mapsto a^{p}-a a ↦ a p − a , which is induced by the étale componentwise Artin–Schreier morphism on A n \mathbb{A}^{n} A n and the quotient morphism by the natural action of the permutation group W = S n W=S_{n} W = S n .
Spectral description of logarithmic connections
For the rest of the paper, we write A : = A ( ω C ′ ( q ′ ) ) . A:=A(\omega_{C^{\prime}}(q^{\prime})). A := A ( ω C ′ ( q ′ )) . Let S S S be the universal spectral curve over A A A . Let P ( S / A ) \mathcal{P}(S/A) P ( S / A ) be the relative Picard stack for S / A S/A S / A . Let T F ( S / A ) \mathcal{TF}(S/A) T F ( S / A ) be the A A A -stack parametrizing torsion free rank 1 sheaves on S S S over A A A in the sense of Schaub [Sch98], see also [dC17, §2.4]. By the Beauville–Narasimhan–Ramanan (BNR) correspondence as in [Sch98, Propositions 2.1 and 5.1], there is a canonical isomorphism of A A A -stacks
H i g g s ( C ′ , n ) ≅ T F ( S / A ) , \mathcal{H}iggs(C^{\prime},n)\cong\mathcal{TF}(S/A), H i g g s ( C ′ , n ) ≅ T F ( S / A ) ,
mapping a logarithmic Higgs bundle to its spectral sheaf.
Let S F r : = C × Fr C , C ′ S S^{\mathrm{Fr}}:=C\times_{\operatorname{Fr}_{C},C^{\prime}}S S Fr := C × Fr C , C ′ S be the Frobenius pullback of the universal spectral curve S S S . Let P ( S F r / A ) \mathcal{P}(S^{\mathrm{Fr}}/A) P ( S Fr / A ) be the relative Picard stack for S F r / A S^{\mathrm{Fr}}/A S Fr / A . Let T F ( S F r / A ) \mathcal{TF}(S^{\mathrm{Fr}}/A) T F ( S Fr / A ) be the A A A -stack parametrizing torsion free rank 1 sheaves on S F r S^{\mathrm{Fr}} S Fr over A A A . Taking the spectral sheaf of the p p p -curvature ψ ∇ \psi_{\nabla} ψ ∇ , we obtain a canonical morphism of A A A -stacks
Conn ( C , n ) ⟶ T F ( S F r / A ) . (5) \operatorname{Conn}(C,n)\longrightarrow\mathcal{TF}(S^{\mathrm{Fr}}/A). \tag*{(5)} Conn ( C , n ) ⟶ T F ( S Fr / A ) . (5)
Let
H : = Conn ( C , n ) × T F ( S F r / A ) P ( S F r / A ) \mathcal{H}:=\operatorname{Conn}(C,n)\mathbin{\times_{T_F(S^{\mathrm{Fr}}/A)}}\mathcal{P}(S^{\mathrm{Fr}}/A) H := Conn ( C , n ) × T F ( S Fr / A ) P ( S Fr / A )
be the open substack of Conn ( C , n ) \operatorname{Conn}(C,n) Conn ( C , n ) characterized by the spectral sheaf of ψ ∇ \psi_{\nabla} ψ ∇ being invertible. To characterize H \mathcal{H} H , we introduce the following:
Definition 2.1. Let b = b 1 n 1 b 2 n 2 ⋯ b s n s ∈ t ( k ) / W \mathbf{b}=b_1^{n_1}b_2^{n_2}\cdots b_s^{n_s}\in\mathfrak{t}(k)/W b = b 1 n 1 b 2 n 2 ⋯ b s n s ∈ t ( k ) / W be parametrized as in (3) . We say b \mathbf{b} b is a regular tuple if for each pair j , j ′ j,j' j , j ′ with j ≠ j ′ j\ne j' j = j ′ , we have b j − b j ′ ∉ F p b_j-b_{j'}\notin\mathbb{F}_p b j − b j ′ ∈ / F p .
Given a connection ( E , ∇ ) (E,\nabla) ( E , ∇ ) with eigenvalues of residue b \mathbf{b} b , if b \mathbf{b} b is not regular, then by writing residues in Jordan normal forms, the p p p -curvature ψ ∇ \psi_{\nabla} ψ ∇ is not regular in the Lie-theoretic sense, i.e. the corresponding Higgs bundle ( E , ψ ∇ ) (E,\psi_{\nabla}) ( E , ψ ∇ ) does not land in the open Picard stack inside the Higgs moduli, see [CZ17, Proposition 2.6.1.(1)] and [dCZ25, §2.2.8]. By [dCZ25, Proof of Lemma 3.15], objects in H \mathcal{H} H can only have regular eigenvalues of residues. For any regular tuple b \mathbf{b} b , we define H b ⊂ H \mathcal{H}_{\mathbf{b}}\subset\mathcal{H} H b ⊂ H to be the substack characterized by the condition that the residue of the connection has eigenvalues b \mathbf{b} b . Let a = AS ( b ) \mathbf{a}=\operatorname{AS}(\mathbf{b}) a = AS ( b ) , and let A a : = ev ‾ q ′ − 1 ( a ) ⊂ A \mathcal{A}_{\mathbf{a}}:=\overline{\operatorname{ev}}_{q'}^{-1}(\mathbf{a})\subset\mathcal{A} A a := ev q ′ − 1 ( a ) ⊂ A , where ev ‾ q ′ \overline{\operatorname{ev}}_{q'} ev q ′ is the evaluation-at-q ′ q' q ′ morphism (2) . It follows from (4) that h D R ( H b ) ⊆ A a h_{\mathrm{DR}}(\mathcal{H}_{\mathbf{b}})\subseteq\mathcal{A}_{\mathbf{a}} h DR ( H b ) ⊆ A a .
For the rest of the paper, we write P : = P ( S / A ) \mathcal{P}:=\mathcal{P}(S/A) P := P ( S / A ) and P a : = P ( S / A a ) \mathcal{P}_{\mathbf{a}}:=\mathcal{P}(S/\mathcal{A}_{\mathbf{a}}) P a := P ( S / A a ) . Recall that by [She24, Lemma 4.11], each H b \mathcal{H}_{\mathbf{b}} H b admits a P a \mathcal{P}_{\mathbf{a}} P a -action, and becomes a pseudo P a \mathcal{P}_{\mathbf{a}} P a -torsor under this action, see [She24, Proposition 4.13]. We now show the following
Proposition 2.2. For any regular tuple b \mathbf{b} b , H b \mathcal{H}_{\mathbf{b}} H b is a P a \mathcal{P}_{\mathbf{a}} P a -torsor.
Proof. It is enough to show that H b → A a \mathcal{H}_{\mathbf{b}}\to\mathcal{A}_{\mathbf{a}} H b → A a is surjective. When char ( k ) = p \operatorname{char}(k)=p char ( k ) = p satisfies p > n p>n p > n , this is shown in [She24, Proposition 4.6]. The construction of local models in Step 2 of the proof of [She24, Proposition 4.6] relies on the assumption p > n p>n p > n . We now provide a slightly different argument that does not rely on this assumption. Let b = b 1 n 1 b 2 n 2 ⋯ b s n s \mathbf{b}=b_1^{n_1}b_2^{n_2}\cdots b_s^{n_s} b = b 1 n 1 b 2 n 2 ⋯ b s n s be parametrized as in (3) , and let AS ( b ) = a = a 1 n 1 a 2 n 2 ⋯ a s n s \operatorname{AS}(\mathbf{b})=\mathbf{a}=a_1^{n_1}a_2^{n_2}\cdots a_s^{n_s} AS ( b ) = a = a 1 n 1 a 2 n 2 ⋯ a s n s , with AS ( b j ) = a j \operatorname{AS}(b_j)=a_j AS ( b j ) = a j . Let D q = Spec ( k [ [ x ] ] ) D_q=\operatorname{Spec}(k[[x]]) D q = Spec ( k [[ x ]]) be the formal disc around q ∈ C q\in C q ∈ C , and set y = x p ∂ x y=x^p\partial_x y = x p ∂ x . Let a ∈ A a a\in\mathcal{A}_{\mathbf{a}} a ∈ A a . When restricted to D q D_q D q , the Frobenius pullback of the spectral curve decomposes as
S a F r × C D q = ∏ j = 1 s S a , a j F r . S_a^{\mathrm{Fr}}\times_C D_q=\prod_{j=1}^{s}S_{a,a_j}^{\mathrm{Fr}}. S a Fr × C D q = j = 1 ∏ s S a , a j Fr .
Each S a , a j F r S_{a,a_j}^{\mathrm{Fr}} S a , a j Fr is of the form
S a , a j F r ≃ Spec ( k [ [ x ] ] [ y ] / ( f j ) ) , S_{a,a_j}^{\mathrm{Fr}}\simeq\operatorname{Spec}(k[[x]][y]/(f_j)), S a , a j Fr ≃ Spec ( k [[ x ]] [ y ] / ( f j )) ,
where f j ∈ k [ [ x p ] ] [ y ] f_j\in k[[x^p]][y] f j ∈ k [[ x p ]] [ y ] and satisfies f ‾ j = ( y − a j ) n j ∈ k [ [ x p ] ] [ y ] / x p k [ [ x p ] ] [ y ] \overline{f}_j=(y-a_j)^{n_j}\in k[[x^p]][y]/x^pk[[x^p]][y] f j = ( y − a j ) n j ∈ k [[ x p ]] [ y ] / x p k [[ x p ]] [ y ] . Our goal is to construct for each j j j a logarithmic connection ( E j , ∇ j ) (E_j,\nabla_j) ( E j , ∇ j ) over D q D_q D q that satisfies the following three conditions:
the eigenvalues of res q ( ∇ j ) \operatorname{res}_q(\nabla_j) res q ( ∇ j ) are all b j b_j b j ,
the spectral curve of ψ ∇ j \psi_{\nabla_j} ψ ∇ j is S a , a j F r S_{a,a_j}^{\mathrm{Fr}} S a , a j Fr ,
the spectral sheaf of ψ ∇ j \psi_{\nabla_j} ψ ∇ j is an invertible sheaf on S a , a j F r S_{a,a_j}^{\mathrm{Fr}} S a , a j Fr .
Since S a , a j F r S_{a,a_j}^{\mathrm{Fr}} S a , a j Fr is the spectrum of a local ring, (3) implies that the spectral sheaf of ψ ∇ j \psi_{\nabla_j} ψ ∇ j is isomorphic to O S a , a j F r \mathcal{O}_{S_{a,a_j}^{\mathrm{Fr}}} O S a , a j Fr . Let e e e be its generator. A logarithmic connection with spectral curve S a , a j F r S_{a,a_j}^{\mathrm{Fr}} S a , a j Fr is determined by the connection acting on e e e , which can be written as ∇ ( e ) = g d x x \nabla(e)=g\frac{dx}{x} ∇ ( e ) = g x d x , g ∈ k [ [ x ] ] [ y ] / ( f j ) g\in k[[x]][y]/(f_j) g ∈ k [[ x ]] [ y ] / ( f j ) . Assuming g ∈ k [ [ x p ] ] [ y ] / ( f j ) g\in k[[x^p]][y]/(f_j) g ∈ k [[ x p ]] [ y ] / ( f j ) , conditions (1) and (2) are equivalent to
g p − g = y , g ≡ b j ( m o d ( x p , y − a j ) ) . (6) g^p-g=y,\qquad g\equiv b_j\pmod{(x^p,y-a_j)}. \tag*{(6)} g p − g = y , g ≡ b j ( mod ( x p , y − a j )) . (6)
Since b j b_j b j is a simple root of z p − z = a j z^p-z=a_j z p − z = a j , the equations (6) admit a unique solution by Hensel’s lemma. Indeed, this solution can be written explicitly as
g = b j − ∑ m = 0 ∞ ( y − a j ) p m . g=b_j-\sum_{m=0}^{\infty}(y-a_j)^{p^m}. g = b j − m = 0 ∑ ∞ ( y − a j ) p m .
The rest of the argument proceeds as in Steps 1 and 3 of the proof of [She24, Proposition 4.6], using Beauville–Laszlo gluing to obtain the desired global logarithmic connection. □
The stack of very good logarithmic connections
Let b ∈ t ( k ) / W b\in t(k)/W b ∈ t ( k ) / W be a regular tuple as defined in Theorem 2.1. The goal of this section is to define the substack of so-called very good logarithmic connections H b v G ⊂ H b \mathcal{H}_b^{\mathrm{vG}}\subset\mathcal{H}_b H b vG ⊂ H b , which plays a crucial role in our main theorem on the non-abelian Hodge correspondence for semistable parabolic connections.
Before defining the very good substack, we fix the following notation and conventions. Let σ : k → F p \sigma:k\to\mathbb{F}_p σ : k → F p be an F p \mathbb{F}_p F p -linear retraction of the inclusion F p ↪ k \mathbb{F}_p\hookrightarrow k F p ↪ k . We fix this choice throughout this paper. Let π : F p → Z \pi:\mathbb{F}_p\to\mathbb{Z} π : F p → Z be the set-theoretic section of the quotient map Z → F p \mathbb{Z}\to\mathbb{F}_p Z → F p with image in { − p + 1 , − p + 2 , … , − 1 , 0 } \{-p+1,-p+2,\ldots,-1,0\} { − p + 1 , − p + 2 , … , − 1 , 0 } . Let
ν : = τ ∘ σ : k → Z \nu:=\tau\circ\sigma:k\to\mathbb{Z} ν := τ ∘ σ : k → Z
be the composition. Since σ \sigma σ is an F p \mathbb{F}_p F p -linear retraction, ν ( b + c ) = τ ( σ ( b ) + c ) \nu(b+c)=\tau(\sigma(b)+c) ν ( b + c ) = τ ( σ ( b ) + c ) for c ∈ F p c\in\mathbb{F}_p c ∈ F p , so ν \nu ν restricts to a bijection from each Artin–Schreier fiber b + F p b+\mathbb{F}_p b + F p onto { − p + 1 , … , − 1 , 0 } \{-p+1,\ldots,-1,0\} { − p + 1 , … , − 1 , 0 } .
Let b = b 1 n 1 b 2 n 2 ⋯ b s n s b=b_1^{n_1}b_2^{n_2}\cdots b_s^{n_s} b = b 1 n 1 b 2 n 2 ⋯ b s n s be parametrized as in (3), and let AS ( b ) = a = a 1 n 1 a 2 n 2 ⋯ a s n s \operatorname{AS}(b)=a=a_1^{n_1}a_2^{n_2}\cdots a_s^{n_s} AS ( b ) = a = a 1 n 1 a 2 n 2 ⋯ a s n s , with AS ( b j ) = a j \operatorname{AS}(b_j)=a_j AS ( b j ) = a j . It follows from Theorem 2.2 that h d R ( H b ) = A a h_{\mathrm{dR}}(\mathcal{H}_b)=A_a h dR ( H b ) = A a . Let a ∈ A a a\in A_a a ∈ A a . Given an object E : = ( E , ∇ ) \mathcal{E}:=(E,\nabla) E := ( E , ∇ ) in the fiber groupoid ( H b ) a (\mathcal{H}_b)_a ( H b ) a , let L E \mathcal{L}_{\mathcal{E}} L E be the spectral sheaf of the p p p -curvature ψ ∇ \psi_\nabla ψ ∇ , i.e. the line bundle on S a F r S_a^{\mathrm{Fr}} S a Fr given by the image of the morphism (5). The corresponding spectral curve S a F r S_a^{\mathrm{Fr}} S a Fr decomposes into integral components
S a F r = ∑ t = 1 N n t S a , t F r . S_a^{\mathrm{Fr}}=\sum_{t=1}^{N}n_tS_{a,t}^{\mathrm{Fr}}. S a Fr = t = 1 ∑ N n t S a , t Fr .
where S a , t F r S_{a,t}^{\mathrm{Fr}} S a , t Fr are the distinct reduced irreducible components and n t n_t n t are their multiplicities. Let π F r : S a F r → C \pi^{\mathrm{Fr}}:S_a^{\mathrm{Fr}}\to C π Fr : S a Fr → C be the projection. We denote
( π F r ) − 1 ( q ) × S a F r S a , t F r = ∑ l = 1 M t c t , l a t , l (\pi^{\mathrm{Fr}})^{-1}(q)\times_{S_a^{\mathrm{Fr}}}S_{a,t}^{\mathrm{Fr}}=\sum_{l=1}^{M_t}c_{t,l}a_{t,l} ( π Fr ) − 1 ( q ) × S a Fr S a , t Fr = l = 1 ∑ M t c t , l a t , l
with c t , l ∈ N > 0 c_{t,l}\in\mathbb{N}_{>0} c t , l ∈ N > 0 and a t , l a_{t,l} a t , l a closed point in the fiber S a , t F r × C q ⊂ Tot ( Fr ∗ c ( ω C ( q ′ ) ) ) × C q S_{a,t}^{\mathrm{Fr}}\times_Cq\subset\operatorname{Tot}(\operatorname{Fr}_*^c(\omega_C(q')))\times_Cq S a , t Fr × C q ⊂ Tot ( Fr ∗ c ( ω C ( q ′ ))) × C q . The latter is canonically identified with A k 1 A_k^1 A k 1 through the Frobenius pullback of the residue trivialization, so that we can write a t , l ∈ k u n a m b i g u o u s l y . S i n c e a_{t,l}\in k unambiguously. Since a t , l ∈ k u nambi g u o u s l y . S in ce b$ is a regular tuple, there is a unique b j b_j b j in the decomposition b = b 1 n 1 b 2 n 2 ⋯ b s n s b=b_1^{n_1}b_2^{n_2}\cdots b_s^{n_s} b = b 1 n 1 b 2 n 2 ⋯ b s n s satisfying
b j p − b j = a t , l ∈ k . b_j^p-b_j=a_{t,l}\in k. b j p − b j = a t , l ∈ k .
We write this b j b_j b j as b t , l b_{t,l} b t , l .
Definition 3.1. We say that E \mathcal{E} E is very good if, for each integral component S a , t F r S_{a,t}^{\mathrm{Fr}} S a , t Fr of S a F r S_a^{\mathrm{Fr}} S a Fr , the following equality holds in Z \mathbb{Z} Z :
deg S a , t F r ( L E ∣ S a , t F r ) + ∑ l = 1 M t c t , l ν ( b t , l ) = 0. (7) \deg_{S_{a,t}^{\mathrm{Fr}}}(\mathcal{L}_{\mathcal{E}}|_{S_{a,t}^{\mathrm{Fr}}})+\sum_{l=1}^{M_t}c_{t,l}\nu(b_{t,l})=0. \tag*{(7)} deg S a , t Fr ( L E ∣ S a , t Fr ) + l = 1 ∑ M t c t , l ν ( b t , l ) = 0. (7)
Definition 3.2. Let H b v g \mathcal{H}_{b}^{\mathrm{vg}} H b vg be the substack of H b \mathcal{H}_{b} H b classifying logarithmic connections E \mathcal{E} E that restrict to a very good logarithmic connection in the sense of Theorem 3.1 at each geometric point of the test scheme.
Let P a ∘ ⊆ P a \mathcal{P}_{a}^{\circ} \subseteq\mathcal{P}_{a} P a ∘ ⊆ P a be the substack characterized by the condition that, at each geometric point of the test scheme with image a ∈ A a a \in A_{a} a ∈ A a , the corresponding line bundle on S a S_{a} S a has degree zero on each integral component of the spectral curve S a S_{a} S a .
Proposition 3.3. The P a \mathcal{P}_{a} P a -action on H b \mathcal{H}_{b} H b restricts to a P a ∘ \mathcal{P}_{a}^{\circ} P a ∘ -action on H b v g \mathcal{H}_{b}^{\mathrm{vg}} H b vg . Moreover, H b v g \mathcal{H}_{b}^{\mathrm{vg}} H b vg is a P a ∘ \mathcal{P}_{a}^{\circ} P a ∘ -torsor.
Proof. Let a ∈ A a a \in A_{a} a ∈ A a . Let W : S a F r = C × F r C , C ′ S a → S a W:S_{a}^{\mathrm{Fr}}=C \times_{\mathrm{Fr}_{C},C'} S_{a} \to S_{a} W : S a Fr = C × Fr C , C ′ S a → S a and π F r : S a F r = C × F r C , C ′ S a → C \pi^{\mathrm{Fr}}:S_{a}^{\mathrm{Fr}}=C \times_{\mathrm{Fr}_{C},C'} S_{a} \to C π Fr : S a Fr = C × Fr C , C ′ S a → C be the projections. Let E = ( E , ∇ E ) ∈ H b \mathcal{E}=(E,\nabla_{E}) \in\mathcal{H}_{b} E = ( E , ∇ E ) ∈ H b such that h d R ( E ) = a ∈ A a h_{\mathrm{dR}}(\mathcal{E})=a \in A_{a} h dR ( E ) = a ∈ A a , and let L E \mathcal{L}_{\mathcal{E}} L E be the spectral sheaf of its p p p -curvature. Let L ′ L' L ′ be a line bundle on S a S_{a} S a . The P a \mathcal{P}_{a} P a -action on H b \mathcal{H}_{b} H b maps ( E , L ′ ) (\mathcal{E},L') ( E , L ′ ) to
( π ∗ F r ( L E ⊗ W ∗ L ′ ) , ∇ ⊗ ) , \left(\pi_{*}^{\mathrm{Fr}}(L_{\mathcal{E}} \otimes W^{*}L'),\nabla_{\otimes}\right), ( π ∗ Fr ( L E ⊗ W ∗ L ′ ) , ∇ ⊗ ) ,
where ∇ ⊗ \nabla_{\otimes} ∇ ⊗ is the tensor product connection of ∇ E \nabla_{E} ∇ E acting on L E L_{\mathcal{E}} L E and the canonical connection on W ∗ L ′ W^{*}L' W ∗ L ′ , see [She24, Lemma 4.9]. Now we assume E ∈ H b v g \mathcal{E} \in\mathcal{H}_{b}^{\mathrm{vg}} E ∈ H b vg . Note that the logarithmic connection resulting from the action of L ′ L' L ′ on E \mathcal{E} E remains in H b v g \mathcal{H}_{b}^{\mathrm{vg}} H b vg if and only if W ∗ L ′ W^{*}L' W ∗ L ′ has degree zero on each integral component of S a , t F r S_{a,t}^{\mathrm{Fr}} S a , t Fr of S a F r S_{a}^{\mathrm{Fr}} S a Fr . Let S a , t r e d S_{a,t}^{\mathrm{red}} S a , t red be the integral component of S a S_{a} S a corresponding to S a , t F r S_{a,t}^{\mathrm{Fr}} S a , t Fr under the universal homeomorphism W : S a F r → S a W:S_{a}^{\mathrm{Fr}} \to S_{a} W : S a Fr → S a . Then we have ( C × F r C , C ′ S a , t r e d ) r e d = S a , t F r (C \times_{\mathrm{Fr}_{C},C'} S_{a,t}^{\mathrm{red}})^{\mathrm{red}}=S_{a,t}^{\mathrm{Fr}} ( C × Fr C , C ′ S a , t red ) red = S a , t Fr , therefore
deg ( W ∗ L ′ ∣ S a , t F r ) = 0 ⟺ deg ( L ′ ∣ S a , t r e d ) = 0. \deg(W^{*}L'|_{S_{a,t}^{\mathrm{Fr}}})=0 \Longleftrightarrow\deg(L'|_{S_{a,t}^{\mathrm{red}}})=0. deg ( W ∗ L ′ ∣ S a , t Fr ) = 0 ⟺ deg ( L ′ ∣ S a , t red ) = 0.
Since all integral components of S a S_{a} S a arise this way for some t t t , it follows that H b v g \mathcal{H}_{b}^{\mathrm{vg}} H b vg is a pseudo P a ∘ \mathcal{P}_{a}^{\circ} P a ∘ -torsor.
Now we show that H b v g \mathcal{H}_{b}^{\mathrm{vg}} H b vg maps surjectively to A a A_{a} A a . Fix a ∈ A a a \in A_{a} a ∈ A a . By Theorem 2.2, there exists E = ( E , ∇ E ) ∈ H b \mathcal{E}=(E,\nabla_{E}) \in\mathcal{H}_{b} E = ( E , ∇ E ) ∈ H b such that h d R ( E ) = a h_{\mathrm{dR}}(\mathcal{E})=a h dR ( E ) = a . Denote Z t : = C × F r C , C ′ S a , t r e d ⊆ S a F r Z_{t}:=C \times_{\mathrm{Fr}_{C},C'} S_{a,t}^{\mathrm{red}} \subseteq S_{a}^{\mathrm{Fr}} Z t := C × Fr C , C ′ S a , t red ⊆ S a Fr . Let π t : S a , t r e d → C ′ \pi_{t}:S_{a,t}^{\mathrm{red}} \to C' π t : S a , t red → C ′ and π t F r : Z t → C \pi_{t}^{\mathrm{Fr}}:Z_{t} \to C π t Fr : Z t → C be the projection maps. Note that Z t = e t S a , t F r Z_{t}=e_{t}S_{a,t}^{\mathrm{Fr}} Z t = e t S a , t Fr for some e t ∈ N > 0 e_{t} \in\mathbb{N}_{>0} e t ∈ N > 0 . Since Z t Z_{t} Z t is defined through Frobenius pullback, its defining ideal is preserved by the connection ∇ E \nabla_{E} ∇ E . Therefore ∇ E \nabla_{E} ∇ E restricts to a logarithmic connection on E t : = ( π t F r ) ∗ ( L E ∣ Z t ) E_{t}:=(\pi_{t}^{\mathrm{Fr}})_{*}(L_{\mathcal{E}}|_{Z_{t}}) E t := ( π t Fr ) ∗ ( L E ∣ Z t ) whose residue has eigenvalues given by ∏ l = 1 M t b t , l e t c t , l \prod_{l=1}^{M_{t}} b_{t,l}^{e_{t}c_{t,l}} ∏ l = 1 M t b t , l e t c t , l . By the residue–degree formula, we have
deg ( E t ) + ∑ l = 1 M t e t c t , l b t , l = 0 in k . \deg(E_{t})+\sum_{l=1}^{M_{t}} e_{t}c_{t,l}b_{t,l}=0 \quad\text{in } k. deg ( E t ) + l = 1 ∑ M t e t c t , l b t , l = 0 in k .
It follows that ∑ l = 1 M t e t c t , l b t , l \sum_{l=1}^{M_{t}} e_{t}c_{t,l}b_{t,l} ∑ l = 1 M t e t c t , l b t , l lies in F p \mathbb{F}_{p} F p . Since σ \sigma σ is an F p \mathbb{F}_{p} F p -linear retraction and ν = τ ∘ σ \nu=\tau\circ\sigma ν = τ ∘ σ , we have
∑ l = 1 M t e t c t , l ν ( b t , l ) = τ ( ∑ l = 1 M t e t c t , l b t , l ) ( m o d p ) . \sum_{l=1}^{M_{t}} e_{t}c_{t,l}\nu(b_{t,l})=\tau\left(\sum_{l=1}^{M_{t}} e_{t}c_{t,l}b_{t,l}\right) \pmod p. l = 1 ∑ M t e t c t , l ν ( b t , l ) = τ ( l = 1 ∑ M t e t c t , l b t , l ) ( mod p ) .
By Riemann–Roch, we have
deg ( E t ) = deg ( L E ∣ Z t ) + p deg ( ( π t ) ∗ O S a , t r e d ) . \deg(E_{t})=\deg(L_{\mathcal{E}}|_{Z_{t}})+p\deg((\pi_{t})_{*}O_{S_{a,t}^{\mathrm{red}}}). deg ( E t ) = deg ( L E ∣ Z t ) + p deg (( π t ) ∗ O S a , t red ) .
It follows that
deg ( L E ∣ Z t ) + ∑ l = 1 M t e t c t , l ν ( b t , l ) = 0 ( m o d p ) . \deg(L_{\mathcal{E}}|_{Z_{t}})+\sum_{l=1}^{M_{t}} e_{t}c_{t,l}\nu(b_{t,l})=0 \pmod p. deg ( L E ∣ Z t ) + l = 1 ∑ M t e t c t , l ν ( b t , l ) = 0 ( mod p ) .
Set d t = 1 p ( deg ( L E ∣ Z t ) + ∑ l = 1 M t e t c t , l ν ( b t , l ) ) d_{t}=\frac{1}{p}\left(\deg(L_{\mathcal{E}}|_{Z_{t}})+\sum_{l=1}^{M_{t}}e_{t}c_{t,l}\nu(b_{t,l})\right) d t = p 1 ( deg ( L E ∣ Z t ) + ∑ l = 1 M t e t c t , l ν ( b t , l ) ) , and let L ′ L' L ′ be a line bundle on S a S_{a} S a with degree − d t -d_{t} − d t on each integral component S a , t r e d S_{a,t}^{\mathrm{red}} S a , t red . Let E ′ \mathcal{E}' E ′ be the logarithmic connection arising from the action of L ′ L' L ′ on E \mathcal{E} E . Since L E ′ = L E ⊗ W ∗ L ′ L_{\mathcal{E}'}=L_{\mathcal{E}}\otimes W^*L' L E ′ = L E ⊗ W ∗ L ′ , we have
deg ( L E ′ ∣ Z t ) + ∑ l = 1 M t e t c t , l ν ( b t , l ) = 0 in Z . \deg(L_{\mathcal{E}'}|Z_t)+\sum_{l=1}^{M_t}e_tc_{t,l}\nu(b_{t,l})=0\quad\text{in }\mathbb{Z}. deg ( L E ′ ∣ Z t ) + l = 1 ∑ M t e t c t , l ν ( b t , l ) = 0 in Z .
Combining with
deg ( L E ′ ∣ Z t ) = e t deg ( L E ′ ∣ S a , t F r ) , \deg(L_{\mathcal{E}'}|Z_t)=e_t\deg(L_{\mathcal{E}'}|S^{\mathrm{Fr}}_{a,t}), deg ( L E ′ ∣ Z t ) = e t deg ( L E ′ ∣ S a , t Fr ) ,
we obtain the desired equality (7) in Theorem 3.1.
Proposition 3.4. Let b ′ ∈ Γ ( q , Q q ⊕ n ) ≃ Z ( k ) / W \mathbf{b}'\in\Gamma(q,Q_q^{\oplus n})\simeq\mathbb{Z}(k)/W b ′ ∈ Γ ( q , Q q ⊕ n ) ≃ Z ( k ) / W be another regular tuple such that AS ( b ′ ) = AS ( b ) = a \operatorname{AS}(\mathbf{b}')=\operatorname{AS}(\mathbf{b})=\mathbf{a} AS ( b ′ ) = AS ( b ) = a . Then we have an isomorphism H e c k e b ′ , b : H b → ∼ H b ′ v g \mathcal{H}\!ecke_{\mathbf{b}',\mathbf{b}}:\mathcal{H}_{\mathbf{b}}\xrightarrow{\sim}\mathcal{H}_{\mathbf{b}'}^{\mathrm{vg}} H ec k e b ′ , b : H b ∼ H b ′ vg of P a o P_a^o P a o -torsors induced by Hecke transformation. Given a third regular tuple b ′ ′ \mathbf{b}'' b ′′ mapping to a \mathbf{a} a under AS \operatorname{AS} AS , we have a canonical 2-isomorphism
H e c k e b ′ ′ , b ′ ∘ H e c k e b ′ , b ≃ H e c k e b ′ ′ , b . \mathcal{H}\!ecke_{\mathbf{b}'',\mathbf{b}'}\circ\mathcal{H}\!ecke_{\mathbf{b}',\mathbf{b}}\simeq\mathcal{H}\!ecke_{\mathbf{b}'',\mathbf{b}}. H ec k e b ′′ , b ′ ∘ H ec k e b ′ , b ≃ H ec k e b ′′ , b .
Proof. Let b = b 1 n 1 b 2 n 2 ⋯ b s n s \mathbf{b}=b_1^{n_1}b_2^{n_2}\cdots b_s^{n_s} b = b 1 n 1 b 2 n 2 ⋯ b s n s , and let AS ( b ) = a = a 1 n 1 a 2 n 2 ⋯ a s n s \operatorname{AS}(\mathbf{b})=\mathbf{a}=a_1^{n_1}a_2^{n_2}\cdots a_s^{n_s} AS ( b ) = a = a 1 n 1 a 2 n 2 ⋯ a s n s , with AS ( b j ) = a j \operatorname{AS}(b_j)=a_j AS ( b j ) = a j . After possibly reordering the indices, b ′ \mathbf{b}' b ′ can be written as b ′ = ( b 1 ′ ) n 1 ( b 2 ′ ) n 2 ⋯ ( b s ′ ) n s \mathbf{b}'=(b'_1)^{n_1}(b'_2)^{n_2}\cdots(b'_s)^{n_s} b ′ = ( b 1 ′ ) n 1 ( b 2 ′ ) n 2 ⋯ ( b s ′ ) n s with AS ( b j ′ ) = AS ( b j ) = a j \operatorname{AS}(b'_j)=\operatorname{AS}(b_j)=a_j AS ( b j ′ ) = AS ( b j ) = a j . Let E = ( E , ∇ ) ∈ H b \mathcal{E}=(E,\nabla)\in\mathcal{H}_{\mathbf{b}} E = ( E , ∇ ) ∈ H b such that h d R ( E ) = a ∈ A a h_{\mathrm{dR}}(\mathcal{E})=\mathbf{a}\in A_{\mathbf{a}} h dR ( E ) = a ∈ A a . The spectral correspondence for ( E , ψ ∇ ) (E,\psi\nabla) ( E , ψ ∇ ) gives rise to a line bundle L E L_{\mathcal{E}} L E on S a F r S^{\mathrm{Fr}}_{\mathbf{a}} S a Fr such that π F r ∗ L E ≅ E \pi^*_{\mathrm{Fr}}L_{\mathcal{E}}\cong E π Fr ∗ L E ≅ E . Let x x x be a local parameter of C C C at q q q , so that O ^ C , q ≃ k [ [ x ] ] \widehat{\mathcal{O}}_{C,q}\simeq k[[x]] O C , q ≃ k [[ x ]] and K q ≃ k ( ( x ) ) K_q\simeq k((x)) K q ≃ k (( x )) . Let D q : = Spec O ^ C , q ≃ Spec k [ [ x ] ] D_q:=\operatorname{Spec}\widehat{\mathcal{O}}_{C,q}\simeq\operatorname{Spec}k[[x]] D q := Spec O C , q ≃ Spec k [[ x ]] be the formal disc around q ∈ C q\in C q ∈ C , and let D ˚ q : = Spec K q ≃ Spec k ( ( x ) ) \mathring{D}_q:=\operatorname{Spec}K_q\simeq\operatorname{Spec}k((x)) D ˚ q := Spec K q ≃ Spec k (( x )) be the punctured disc. By definition of A a A_{\mathbf{a}} A a , we have
( π F r ) − 1 ( q ) = ∑ j = 1 s n j a j . (\pi^{\mathrm{Fr}})^{-1}(q)=\sum_{j=1}^{s}n_ja_j. ( π Fr ) − 1 ( q ) = j = 1 ∑ s n j a j .
It follows that S a F r × C D q S^{\mathrm{Fr}}_{\mathbf{a}}\times_C D_q S a Fr × C D q decomposes as a disjoint union
S a F r × C D q = ∐ j = 1 s S a , a j F r . S^{\mathrm{Fr}}_{\mathbf{a}}\times_C D_q=\coprod_{j=1}^{s}S^{\mathrm{Fr}}_{\mathbf{a},a_j}. S a Fr × C D q = j = 1 ∐ s S a , a j Fr .
Therefore E ∣ D q E|_{D_q} E ∣ D q decomposes as
E ∣ D q = ⨁ j = 1 s M j , E|_{D_q}=\bigoplus_{j=1}^{s}M_j, E ∣ D q = j = 1 ⨁ s M j ,
with M j M_j M j supported on S a , a j F r S^{\mathrm{Fr}}_{\mathbf{a},a_j} S a , a j Fr for each j j j . Note that the connection ∇ \nabla ∇ preserves each M j M_j M j , and res q ( ∇ ) \operatorname{res}_q(\nabla) res q ( ∇ ) has eigenvalue b j b_j b j when acting on M j M_j M j . Consider the k [ [ x ] ] k[[x]] k [[ x ]] -lattice M ′ ⊂ E ∣ D ˚ q M'\subset E|_{\mathring{D}_q} M ′ ⊂ E ∣ D ˚ q defined by
M ′ = x ν ( b 1 ′ ) − ν ( b 1 ) M 1 ⊕ ⋯ ⊕ x ν ( b s ′ ) − ν ( b s ) M s . (8) M'=x^{\nu(b'_1)-\nu(b_1)}M_1\oplus\cdots\oplus x^{\nu(b'_s)-\nu(b_s)}M_s. \tag*{(8)} M ′ = x ν ( b 1 ′ ) − ν ( b 1 ) M 1 ⊕ ⋯ ⊕ x ν ( b s ′ ) − ν ( b s ) M s . (8)
By the Beauville–Laszlo theorem [BL95], E ∣ C ∖ q E|_{C\setminus q} E ∣ C ∖ q and M ′ M' M ′ glue to a new vector bundle E ′ E' E ′ on C C C through their canonical identification over D ˚ q \mathring{D}_q D ˚ q . The logarithmic connection ∇ \nabla ∇ on E E E induces a logarithmic connection on E ′ E' E ′ , which we still denote by ∇ \nabla ∇ . Now we define
H e c k e b ′ , b : H b ⟶ H b ′ , ( E , ∇ ) ⟼ ( E ′ , ∇ ) . \mathcal{H}\!ecke_{\mathbf{b}',\mathbf{b}}:\mathcal{H}_{\mathbf{b}}\longrightarrow\mathcal{H}_{\mathbf{b}'},\qquad(E,\nabla)\longmapsto(E',\nabla). H ec k e b ′ , b : H b ⟶ H b ′ , ( E , ∇ ) ⟼ ( E ′ , ∇ ) .
Note that for each 1 ≤ j ≤ s 1\leq j\leq s 1 ≤ j ≤ s , res q ( ∇ ) \operatorname{res}_q(\nabla) res q ( ∇ ) has eigenvalue ν ( b j ′ ) − ν ( b j ) + b j = b j ′ \nu(b'_j)-\nu(b_j)+b_j=b'_j ν ( b j ′ ) − ν ( b j ) + b j = b j ′ when acting on x ν ( b j ′ ) − ν ( b j ) M j x^{\nu(b'_j)-\nu(b_j)}M_j x ν ( b j ′ ) − ν ( b j ) M j . When restricted to D q ⊂ C D_q\subset C D q ⊂ C , each integral component S a , t F r S^{\mathrm{Fr}}_{a,t} S a , t Fr of S a F r S^{\mathrm{Fr}}_a S a Fr decomposes as
S a , t F r × C D q = ∐ l = 1 M t S a , t , a t , l F r . S^{\mathrm{Fr}}_{a,t}\times_C D_q=\coprod_{l=1}^{M_t}S^{\mathrm{Fr}}_{a,t,a_{t,l}}. S a , t Fr × C D q = l = 1 ∐ M t S a , t , a t , l Fr .
Let L ε , t , l \mathcal{L}_{\varepsilon,t,l} L ε , t , l denote the restriction of L ε \mathcal{L}_{\varepsilon} L ε to S a , t , a t , a t , l F r S^{\mathrm{Fr}}_{a,t,a_t,a_{t,l}} S a , t , a t , a t , l Fr , then the modification (8) corresponds to
L ε , t , l ↦ L ε , t , l ( ( ν ( b t , l ) − ν ( b t , l ′ ) ) div S a , t , a t , l F r ( x ) ) = L ε , t , l ( ( ν ( b t , l ) − ν ( b t , l ′ ) ) c t , l a t , l ) . \mathcal{L}_{\varepsilon,t,l}\mapsto\mathcal{L}_{\varepsilon,t,l}\bigl((\nu(b_{t,l})-\nu(b'_{t,l}))\operatorname{div}S^{\mathrm{Fr}}_{a,t,a_t,l}(x)\bigr)=\mathcal{L}_{\varepsilon,t,l}\bigl((\nu(b_{t,l})-\nu(b'_{t,l}))c_{t,l}a_{t,l}\bigr). L ε , t , l ↦ L ε , t , l ( ( ν ( b t , l ) − ν ( b t , l ′ )) div S a , t , a t , l Fr ( x ) ) = L ε , t , l ( ( ν ( b t , l ) − ν ( b t , l ′ )) c t , l a t , l ) .
It follows that H e c k e b ′ , b \mathcal{H}ecke_{b',b} H ec k e b ′ , b maps H b v g \mathcal{H}^{\mathrm{vg}}_b H b vg to H b ′ v g \mathcal{H}^{\mathrm{vg}}_{b'} H b ′ vg , and is equivariant with respect to the P a o \mathcal{P}^o_a P a o -action. An inverse of H e c k e b ′ , b \mathcal{H}ecke_{b',b} H ec k e b ′ , b is given by H e c k e b , b ′ \mathcal{H}ecke_{b,b'} H ec k e b , b ′ , therefore we have an isomorphism
H e c k e b ′ , b : H b v g ⟶ ∼ H b ′ v g . \mathcal{H}ecke_{b',b}:\mathcal{H}^{\mathrm{vg}}_b\mathbin{\stackrel{\sim}{\longrightarrow}}\mathcal{H}^{\mathrm{vg}}_{b'}. H ec k e b ′ , b : H b vg ⟶ ∼ H b ′ vg .
of P o \mathcal{P}^o P o -torsors. The composition 2-isomorphism follows directly from the definition.
It follows that, up to canonical isomorphism, the P a o \mathcal{P}^o_a P a o -torsor H b v g \mathcal{H}^{\mathrm{vg}}_b H b vg is independent of the choice of the regular Artin–Schreier preimage b b b . From now on we denote this torsor by H a v g \mathcal{H}^{\mathrm{vg}}_a H a vg .
Parabolic structures and residue assignments
Parabolic Higgs bundles and connections
Let m = ( m 1 , m 2 , … , m r ) \boldsymbol{m}=(m_1,m_2,\ldots,m_r) m = ( m 1 , m 2 , … , m r ) be an ordered tuple of positive integers that sum to n n n . A quasi-parabolic bundle of filtration type m \boldsymbol{m} m is a vector bundle E E E together with a filtration of k k k -vector spaces
0 = E 0 ⊂ E 1 ⊂ ⋯ ⊂ E r − 1 ⊂ E r = E q : = E ∣ q , 0=E_0\subset E_1\subset\cdots\subset E_{r-1}\subset E_r=E_q:=E|_q, 0 = E 0 ⊂ E 1 ⊂ ⋯ ⊂ E r − 1 ⊂ E r = E q := E ∣ q ,
such that dim k ( E j / E j − 1 ) = m j \dim_k(E_j/E_{j-1})=m_j dim k ( E j / E j − 1 ) = m j for j = 1 , 2 , … , r j=1,2,\ldots,r j = 1 , 2 , … , r . We denote such a filtration by E ∙ E_\bullet E ∙ .
A quasi-parabolic Higgs bundle is a triple ( E , E ∙ , ϕ ) (E,E_\bullet,\phi) ( E , E ∙ , ϕ ) , where ( E , E ∙ ) (E,E_\bullet) ( E , E ∙ ) is a quasi-parabolic bundle and ϕ \phi ϕ is a logarithmic Higgs field whose residue res q ( ϕ ) \operatorname{res}_q(\phi) res q ( ϕ ) preserves the filtration E ∙ E_\bullet E ∙ : res q ( ϕ ) ( E j ) ⊆ E j \operatorname{res}_q(\phi)(E_j)\subseteq E_j res q ( ϕ ) ( E j ) ⊆ E j . Similarly, a quasi-parabolic connection is a triple ( E , E ∙ , ∇ ) (E,E_\bullet,\nabla) ( E , E ∙ , ∇ ) , where ( E , E ∙ ) (E,E_\bullet) ( E , E ∙ ) is a quasi-parabolic bundle and ∇ \nabla ∇ is a logarithmic connection whose residue res q ( ∇ ) \operatorname{res}_q(\nabla) res q ( ∇ ) preserves the filtration E ∙ E_\bullet E ∙ : res q ( ∇ ) ( E j ) ⊆ E j \operatorname{res}_q(\nabla)(E_j)\subseteq E_j res q ( ∇ ) ( E j ) ⊆ E j . Let H i g g s ( C , m ) \mathit{Higgs}(C,\boldsymbol{m}) Higgs ( C , m ) (resp. C o n n ( C , m ) ) \mathit{Conn}(C,\boldsymbol{m})) Conn ( C , m )) denote the moduli stack of quasi-parabolic Higgs bundles (resp. connections) of filtration type m \boldsymbol{m} m .
A parabolic bundle ( E , E ∙ , α ) (E,E_\bullet,\boldsymbol{\alpha}) ( E , E ∙ , α ) is a quasi-parabolic bundle ( E , E ∙ ) (E,E_\bullet) ( E , E ∙ ) together with a set of parabolic weights α ⊂ [ 0 , 1 ) \boldsymbol{\alpha}\subset[0,1) α ⊂ [ 0 , 1 ) , enumerated in decreasing order as
α = ( α 1 , α 2 , … , α r ) , 1 > α 1 > α 2 > ⋯ > α r ≥ 0. \boldsymbol{\alpha}=(\alpha_1,\alpha_2,\ldots,\alpha_r),\qquad1>\alpha_1>\alpha_2>\cdots>\alpha_r\geq0. α = ( α 1 , α 2 , … , α r ) , 1 > α 1 > α 2 > ⋯ > α r ≥ 0.
Let B u n α ( C , m ) \mathit{Bun}^{\boldsymbol{\alpha}}(C,\boldsymbol{m}) Bun α ( C , m ) , H i g g s α ( C , m ) \mathit{Higgs}^{\boldsymbol{\alpha}}(C,\boldsymbol{m}) Higgs α ( C , m ) , and C o n n α ( C , m ) \mathit{Conn}^{\boldsymbol{\alpha}}(C,\boldsymbol{m}) Conn α ( C , m ) denote the moduli stacks of parabolic bundles, Higgs bundles, and connections with parabolic weights α \boldsymbol{\alpha} α and filtration type m \boldsymbol{m} m , respectively.
Remark 4.1. Let θ ∈ t ( R ) \theta\in\mathfrak{t}(\mathbb{R}) θ ∈ t ( R ) . For a root γ \gamma γ of G L n \mathrm{GL}_n GL n , set
m γ : = [ − γ ( θ ) ] = min { m ∈ Z ∣ γ ( θ ) + m ≥ 0 } . m_\gamma:=[-\gamma(\theta)]=\min\{m\in\mathbb{Z}\mid\gamma(\theta)+m\geq0\}. m γ := [ − γ ( θ )] = min { m ∈ Z ∣ γ ( θ ) + m ≥ 0 } .
We denote by G θ G_\theta G θ the smooth affine group scheme over C C C which is isomorphic to G L n \mathrm{GL}_n GL n over C ∖ q C\setminus q C ∖ q and whose group of k [ [ x ] ] k[[x]] k [[ x ]] -points is generated by T ( k [ [ x ] ] ) T(k[[x]]) T ( k [[ x ]]) together with the root subgroups
U γ ( x m γ k [ [ x ] ] ) . U_\gamma(x^{m_\gamma}k[[x]]). U γ ( x m γ k [[ x ]]) .
Equivalently,
G θ ( R [ [ x ] ] ) = { ( g i j ) ∈ G L n ( R ( ( x ) ) ) | g i j ∈ x [ θ j − θ i ] R [ [ x ] ] for all i , j , det ( g ) ∈ R [ [ x ] ] × } . G_\theta(R[[x]])=\left\{(g_{ij})\in\mathrm{GL}_n(R((x)))\ \middle|\ g_{ij}\in x^{[\theta_j-\theta_i]}R[[x]]\ \text{for all }i,j,\ \det(g)\in R[[x]]^\times\right\}. G θ ( R [[ x ]]) = { ( g ij ) ∈ GL n ( R (( x ))) g ij ∈ x [ θ j − θ i ] R [[ x ]] for all i , j , det ( g ) ∈ R [[ x ] ] × } .
Different orderings of θ \theta θ give isomorphic group schemes, with the isomorphism induced by conjugation by the corresponding permutation matrix. Hence the isomorphism class of G θ \mathcal{G}_{\theta} G θ only depends on the class of θ \theta θ in t ( R ) / W \mathfrak{t}(\mathbb{R})/W t ( R ) / W .
Now let
α = ( α 1 , α 2 , … , α r ) , m = ( m 1 , m 2 , … , m r ) , \boldsymbol{\alpha}=(\alpha_1,\alpha_2,\ldots,\alpha_r),\qquad\boldsymbol{m}=(m_1,m_2,\ldots,m_r), α = ( α 1 , α 2 , … , α r ) , m = ( m 1 , m 2 , … , m r ) ,
with 1 > α 1 > α 2 > ⋯ > α r ≥ 0 1>\alpha_1>\alpha_2>\cdots>\alpha_r\geq0 1 > α 1 > α 2 > ⋯ > α r ≥ 0 and ∑ j = 1 r m j = n \sum_{j=1}^{r}m_j=n ∑ j = 1 r m j = n . We associate to ( α , m ) (\boldsymbol{\alpha},\boldsymbol{m}) ( α , m ) the element
θ ( α , m ) = diag ( α 1 ( m 1 ) , α 2 ( m 2 ) , … , α r ( m r ) ) ∈ t ( R ) , \theta(\boldsymbol{\alpha},\boldsymbol{m})=\operatorname{diag}(\alpha_1^{(m_1)},\alpha_2^{(m_2)},\ldots,\alpha_r^{(m_r)})\in\mathfrak{t}(\mathbb{R}), θ ( α , m ) = diag ( α 1 ( m 1 ) , α 2 ( m 2 ) , … , α r ( m r ) ) ∈ t ( R ) ,
where α j ( m j ) \alpha_j^{(m_j)} α j ( m j ) means that α j \alpha_j α j is repeated m j m_j m j times. Let θ = ( θ 1 , θ 2 , … , θ n ) ∈ t ( R ) \theta=(\theta_1,\theta_2,\ldots,\theta_n)\in\mathfrak{t}(\mathbb{R}) θ = ( θ 1 , θ 2 , … , θ n ) ∈ t ( R ) be an element in the same Weyl group orbit of θ ( α , m ) \theta(\boldsymbol{\alpha},\boldsymbol{m}) θ ( α , m ) . A parabolic vector bundle on C C C with parabolic weights α \boldsymbol{\alpha} α and filtration type m \boldsymbol{m} m is equivalent to a G θ \mathcal{G}_{\theta} G θ -torsor on C C C . We describe this correspondence as follows. Fix a model object F 0 = ( F 0 , F 0 ) \mathcal{F}^{0}=(F^{0},\mathbf{F}^{0}) F 0 = ( F 0 , F 0 ) , where F 0 = ⨁ i = 1 r O C F^{0}=\bigoplus_{i=1}^{r}\mathcal{O}_{C} F 0 = ⨁ i = 1 r O C , and the parabolic structure is obtained by assigning the parabolic weight θ i \theta_i θ i to the i i i -th summand ( O C ) q ⊂ F q 0 (\mathcal{O}_{C})_q\subset F_q^{0} ( O C ) q ⊂ F q 0 . The sheaf of automorphisms Aut ( F 0 ) \operatorname{Aut}(\mathcal{F}^{0}) Aut ( F 0 ) is equal to G θ \mathcal{G}_{\theta} G θ . For any parabolic bundle F = ( F , F ) \mathcal{F}=(F,\mathbf{F}) F = ( F , F ) with parabolic weights α \boldsymbol{\alpha} α and filtration type m \boldsymbol{m} m , we associate with it the sheaf Isom ( F 0 , F ) \operatorname{Isom}(\mathcal{F}^{0},\mathcal{F}) Isom ( F 0 , F ) of isomorphisms from F 0 \mathcal{F}^{0} F 0 to F \mathcal{F} F . Note that Isom ( F 0 , F ) \operatorname{Isom}(\mathcal{F}^{0},\mathcal{F}) Isom ( F 0 , F ) is an Aut ( F 0 ) \operatorname{Aut}(\mathcal{F}^{0}) Aut ( F 0 ) -torsor through precomposition, and the evaluation map induces
Isom ( F 0 , F ) × Aut ( F 0 ) F 0 ≅ F . \operatorname{Isom}(\mathcal{F}^{0},\mathcal{F})\times^{\operatorname{Aut}(\mathcal{F}^{0})}\mathcal{F}^{0}\cong\mathcal{F}. Isom ( F 0 , F ) × Aut ( F 0 ) F 0 ≅ F .
For each 1 ≤ j ≤ r 1\leq j\leq r 1 ≤ j ≤ r , we denote by gr α j ( E q ) \operatorname{gr}_{\alpha_j}(E_q) gr α j ( E q ) the graded piece of weight α j \alpha_j α j , i.e. gr α j ( E q ) : = E j / E j − 1 \operatorname{gr}_{\alpha_j}(E_q):=E_j/E_{j-1} gr α j ( E q ) := E j / E j − 1 . The α \boldsymbol{\alpha} α -parabolic degree of ( E , E ∙ ) (E,E_{\bullet}) ( E , E ∙ ) is given by
pardeg ( E , E ∙ , α ) : = deg ( E ) + ∑ j = 1 r α j m j . \operatorname{pardeg}(E,E_{\bullet},\boldsymbol{\alpha}):=\deg(E)+\sum_{j=1}^{r}\alpha_jm_j. pardeg ( E , E ∙ , α ) := deg ( E ) + j = 1 ∑ r α j m j .
Given a subbundle E ′ E' E ′ of E E E , we obtain a parabolic subbundle ( E ′ , E ∙ ′ , α ′ ) (E',E'_{\bullet},\boldsymbol{\alpha}') ( E ′ , E ∙ ′ , α ′ ) of ( E , E ∙ , α ) (E,E_{\bullet},\boldsymbol{\alpha}) ( E , E ∙ , α ) as follows. The filtration E ∙ ′ E'_{\bullet} E ∙ ′ is given by intersecting E q ′ E'_q E q ′ with the filtration E ∙ E_{\bullet} E ∙ . The induced parabolic weights α ′ \boldsymbol{\alpha}' α ′ on ( E ′ , E ∙ ′ ) (E',E'_{\bullet}) ( E ′ , E ∙ ′ ) are given by α k ′ = max { α j ∣ E k ′ ⊆ E j } \alpha'_k=\max\{\alpha_j\mid E'_k\subseteq E_j\} α k ′ = max { α j ∣ E k ′ ⊆ E j } . We can then make sense of the α \boldsymbol{\alpha} α -parabolic slope (semi)stability of parabolic bundles, Higgs bundles, and connections, where in the latter two cases we consider subbundles preserved by the Higgs field or connection, respectively. Let B u n α -ss ( C , m ) \mathcal{Bun}^{\alpha\text{-ss}}(C,\boldsymbol{m}) B u n α -ss ( C , m ) , H i g g s α -ss ( C , m ) \mathcal{Higgs}^{\alpha\text{-ss}}(C,\boldsymbol{m}) H i g g s α -ss ( C , m ) , and C o n n α -ss ( C , m ) \mathcal{Conn}^{\alpha\text{-ss}}(C,\boldsymbol{m}) C o nn α -ss ( C , m ) denote the moduli stacks of α \alpha α -semistable parabolic bundles, Higgs bundles, and connections of filtration type m \boldsymbol{m} m , respectively.
Residue assignments and the associated moduli substacks
Let α \boldsymbol{\alpha} α be a set of parabolic weights on a quasi-parabolic bundle of filtration type m \boldsymbol{m} m as in Section 4.1. In order to obtain a non-abelian Hodge correspondence for parabolic Higgs bundles ( E , E ∙ , α , ϕ ) (E,E_{\bullet},\boldsymbol{\alpha},\phi) ( E , E ∙ , α , ϕ ) , we introduce notation and conventions to fix the multiset of eigenvalues for the residues res q ( ϕ ) \operatorname{res}_{q}(\phi) res q ( ϕ ) on each graded piece of the filtration E ∙ E_{\bullet} E ∙ . We fix a ∈ t ( K ) / W \mathbf{a}\in\mathfrak{t}(\mathbb{K})/W a ∈ t ( K ) / W , and fix an (arbitrary) lift of a \mathbf{a} a to a ~ ∈ t ( K ) \widetilde{\mathbf{a}}\in\mathfrak{t}(\mathbb{K}) a ∈ t ( K ) , which is represented by an ordered tuple a ~ = ( a 1 , a 2 , … , a n ) \widetilde{\mathbf{a}}=(a_1,a_2,\ldots,a_n) a = ( a 1 , a 2 , … , a n ) . For each 1 ≤ i ≤ n 1\leq i\leq n 1 ≤ i ≤ n , we assign a parabolic weight α ~ op a ( i ) ∈ α \widetilde{\alpha}_{\operatorname{op}}^{\mathbf{a}}(i)\in\boldsymbol{\alpha} α op a ( i ) ∈ α to each a i a_i a i in a ~ \widetilde{\mathbf{a}} a , subject to the requirement that each α j ∈ α \alpha_j\in\boldsymbol{\alpha} α j ∈ α appears exactly m j m_j m j times. We denote such an assignment of parabolic weights to eigenvalues by a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α . More precisely,
Definition 4.2. Given a filtration type m \boldsymbol{m} m , a set of parabolic weights α ⊆ [ 0 , 1 ) \boldsymbol{\alpha}\subseteq[0,1) α ⊆ [ 0 , 1 ) with size equal to the length of m \boldsymbol{m} m , and an element a ∈ t ( K ) / W \mathbf{a}\in\mathfrak{t}(\mathbb{K})/W a ∈ t ( K ) / W , an assignment a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α (of parabolic weights to eigenvalues) of filtration type m \boldsymbol{m} m is defined as an equivalence class of maps
α ~ op a : { 1 , … , n } ⟶ α , i ⟼ α ~ op a ( i ) , \widetilde{\alpha}_{\operatorname{op}}^{\mathbf{a}}:\{1,\ldots,n\}\longrightarrow\boldsymbol{\alpha},\qquad i\longmapsto\widetilde{\alpha}_{\operatorname{op}}^{\mathbf{a}}(i), α op a : { 1 , … , n } ⟶ α , i ⟼ α op a ( i ) ,
subject to
# { i ∣ α ~ o p ( i ) = α j } = m j \#\{i\mid\tilde{\alpha}_{\mathrm{op}}(i)=\alpha_j\}=m_j # { i ∣ α ~ op ( i ) = α j } = m j
for each j j j . Two such maps are equivalent if they differ by a permutation of indices preserving a ~ \tilde{\mathbf{a}} a ~ . Whenever we write α ~ o p \tilde{\alpha}_{\mathrm{op}} α ~ op , we choose a representative of the equivalence class.
Now we define the following substacks of H i g g s α ( C , m ) \mathit{Higgs}^{\alpha}(\mathcal{C},\mathbf{m}) Higgs α ( C , m ) and C o n n α ( C , m ) \mathit{Conn}^{\alpha}(\mathcal{C},\mathbf{m}) Conn α ( C , m ) characterized by the condition that the multiset of eigenvalues of the residue acting on each graded piece is given by a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α . These substacks are the main players in the non-abelian Hodge correspondences for parabolic Higgs bundles established in this paper.
Definition 4.3. Let a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α be an assignment of filtration type m \mathbf{m} m as in Theorem 4.2.
(1) We denote by H i g g s a ↦ α ( C ) ⊆ H i g g s α ( C , m ) \mathit{Higgs}^{\mathbf{a}\mapsto\boldsymbol{\alpha}}(\mathcal{C})\subseteq\mathit{Higgs}^{\alpha}(\mathcal{C},\mathbf{m}) Higgs a ↦ α ( C ) ⊆ Higgs α ( C , m ) the substack classifying parabolic Higgs bundles ( E , E ∙ , ϕ ) (E,\mathbf{E}_{\bullet},\phi) ( E , E ∙ , ϕ ) that satisfy the following extra condition on the residue res q ( ϕ ) \operatorname{res}_{q}(\phi) res q ( ϕ ) of the Higgs field. For each 1 ≤ j ≤ r 1\leq j\leq r 1 ≤ j ≤ r , we require that the multiset of eigenvalues of res q ( ϕ ) \operatorname{res}_{q}(\phi) res q ( ϕ ) acting on gr α j ( E q ) \operatorname{gr}_{\alpha_j}(E_q) gr α j ( E q ) is given by
{ a i ∣ α ~ o p ( i ) = α j } , \{a_i\mid\tilde{\alpha}_{\mathrm{op}}(i)=\alpha_j\}, { a i ∣ α ~ op ( i ) = α j } , counted with multiplicity. Denote H i g g s a ↦ α , s s ( C ) : = H i g g s a ↦ α ( C ) ∩ H i g g s α - s s ( C , m ) \mathit{Higgs}^{\mathbf{a}\mapsto\boldsymbol{\alpha},\mathrm{ss}}(\mathcal{C}):=\mathit{Higgs}^{\mathbf{a}\mapsto\boldsymbol{\alpha}}(\mathcal{C})\cap\mathit{Higgs}^{\alpha\text{-}\mathrm{ss}}(\mathcal{C},\mathbf{m}) Higgs a ↦ α , ss ( C ) := Higgs a ↦ α ( C ) ∩ Higgs α - ss ( C , m ) .
(2) We denote by C o n n a ↦ α ( C ) ⊆ C o n n α ( C , m ) \mathit{Conn}^{\mathbf{a}\mapsto\boldsymbol{\alpha}}(\mathcal{C})\subseteq\mathit{Conn}^{\alpha}(\mathcal{C},\mathbf{m}) Conn a ↦ α ( C ) ⊆ Conn α ( C , m ) the substack classifying parabolic connections ( E , E ∙ , ∇ ) (E,\mathbf{E}_{\bullet},\nabla) ( E , E ∙ , ∇ ) that satisfy the following extra condition on the residue res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) of the connection. For each 1 ≤ j ≤ r 1\leq j\leq r 1 ≤ j ≤ r , we require that the multiset of eigenvalues of res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) acting on gr α j ( E q ) \operatorname{gr}_{\alpha_j}(E_q) gr α j ( E q ) is given by
{ a i ∣ α ~ o p ( i ) = α j } , \{a_i\mid\tilde{\alpha}_{\mathrm{op}}(i)=\alpha_j\}, { a i ∣ α ~ op ( i ) = α j } , counted with multiplicity. Denote C o n n a ↦ α , s s ( C ) : = C o n n a ↦ α ( C ) ∩ C o n n α - s s ( C , m ) \mathit{Conn}^{\mathbf{a}\mapsto\boldsymbol{\alpha},\mathrm{ss}}(\mathcal{C}):=\mathit{Conn}^{\mathbf{a}\mapsto\boldsymbol{\alpha}}(\mathcal{C})\cap\mathit{Conn}^{\alpha\text{-}\mathrm{ss}}(\mathcal{C},\mathbf{m}) Conn a ↦ α , ss ( C ) := Conn a ↦ α ( C ) ∩ Conn α - ss ( C , m ) .
Non-abelian Hodge correspondence for semistable parabolic connections
Statement of the main theorem
Let a ∈ t ( k ) / W \mathbf{a}\in t(\mathsf{k})/W a ∈ t ( k ) / W and let α \boldsymbol{\alpha} α be a set of parabolic weights. Fix an assignment a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α as in Theorem 4.2. Since the restriction of ν \nu ν to each fiber of the Artin–Schreier map is a bijection onto { − p + 1 , … , − 1 , 0 } \{-p+1,\ldots,-1,0\} { − p + 1 , … , − 1 , 0 } , the following lemma is elementary:
Lemma 5.1. There is a unique ordered tuple b ~ = ( b 1 , b 2 , … , b n ) ∈ k n \tilde{\mathbf{b}}=(b_1,b_2,\ldots,b_n)\in\mathsf{k}^n b ~ = ( b 1 , b 2 , … , b n ) ∈ k n such that
(1) b i p − b i = a i b_i^{p}-b_i=a_i b i p − b i = a i .
(2) 0 ≤ p α o p a ~ ( i ) + ν ( b i ) < 1 0\leq p\alpha_{\mathrm{op}}^{\tilde{\mathbf{a}}}(i)+\nu(b_i)<1 0 ≤ p α op a ~ ( i ) + ν ( b i ) < 1 .
The real numbers p α o p a ~ ( i ) + ν ( b i ) p\alpha_{\mathrm{op}}^{\tilde{\mathbf{a}}}(i)+\nu(b_i) p α op a ~ ( i ) + ν ( b i ) make up a new set of parabolic weights β \boldsymbol{\beta} β . We obtain an assignment b ↦ β \mathbf{b}\mapsto\boldsymbol{\beta} b ↦ β from a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α by setting
β o p b ~ ( i ) = p α o p a ~ ( i ) + ν ( b i ) . (9) \beta_{\mathrm{op}}^{\tilde{\mathbf{b}}}(i)=p\alpha_{\mathrm{op}}^{\tilde{\mathbf{a}}}(i)+\nu(b_i). \tag*{(9)} β op b ~ ( i ) = p α op a ~ ( i ) + ν ( b i ) . (9)
Conversely, given an assignment b ↦ β \mathbf{b}\mapsto\boldsymbol{\beta} b ↦ β , we can obtain a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α by setting
a ~ = AS ( b ~ ) and α o p a ~ ( i ) = β o p b ~ ( i ) − ν ( b i ) p . (10) \tilde{\mathbf{a}}=\operatorname{AS}(\tilde{\mathbf{b}})\qquad\text{and}\qquad
\alpha_{\mathrm{op}}^{\tilde{\mathbf{a}}}(i)=\frac{\beta_{\mathrm{op}}^{\tilde{\mathbf{b}}}(i)-\nu(b_i)}{p}. \tag*{(10)} a ~ = AS ( b ~ ) and α op a ~ ( i ) = p β op b ~ ( i ) − ν ( b i ) . (10)
Note that (9) and (10) are inverse to each other, thus a ↦ α \mathbf{a}\mapsto\boldsymbol{\alpha} a ↦ α and b ↦ β \mathbf{b}\mapsto\boldsymbol{\beta} b ↦ β uniquely determine each other.
For any d ∈ R d\in\mathbb{R} d ∈ R , let H i g g s a ↦ α , s s ( C , d ) ⊂ H i g g s a ↦ α , s s ( C ) \mathit{Higgs}^{\mathbf{a}\mapsto\boldsymbol{\alpha},\mathrm{ss}}(\mathcal{C},d)\subset\mathit{Higgs}^{\mathbf{a}\mapsto\boldsymbol{\alpha},\mathrm{ss}}(\mathcal{C}) Higgs a ↦ α , ss ( C , d ) ⊂ Higgs a ↦ α , ss ( C ) and C o n n b ↦ β , s s ( C , d ) ⊂ C o n n b ↦ β , s s ( C ) \mathit{Conn}^{\mathbf{b}\mapsto\boldsymbol{\beta},\mathrm{ss}}(\mathcal{C},d)\subset\mathit{Conn}^{\mathbf{b}\mapsto\boldsymbol{\beta},\mathrm{ss}}(\mathcal{C}) Conn b ↦ β , ss ( C , d ) ⊂ Conn b ↦ β , ss ( C ) be the substacks characterized by the underlying parabolic bundle having parabolic degree d d d at each geometric point of the test scheme. Now we are ready to state the main theorem of the paper.
Theorem 5.2. Let a ↦ α a \mapsto\alpha a ↦ α and b ↦ β b \mapsto\beta b ↦ β be two assignments related through (10). For any d ∈ R d \in\mathbb{R} d ∈ R , there exists an isomorphism of A a A_a A a -stacks
H α v g × P a ∘ H i g g s a → α , s s ( C ′ , d ) ≅ C o n n b → β , s s ( C , p ) . \mathcal{H}^{\mathrm{vg}}_{\alpha} \times_{\mathcal{P}^{\circ}_{a}} \mathcal{Higgs}^{a\to\alpha,\mathrm{ss}}(C',d) \cong\mathcal{Conn}^{b\to\beta,\mathrm{ss}}(C,p). H α vg × P a ∘ H i g g s a → α , ss ( C ′ , d ) ≅ C o nn b → β , ss ( C , p ) .
:::
If we choose the trivial filtration type m = ( n ) \mathbf{m}=(n) m = ( n ) and trivial parabolic weight α = { 0 } \alpha=\{0\} α = { 0 } , then we have β = { 0 } \beta=\{0\} β = { 0 } and b b b is a regular tuple. In this case, we recover the semistable logarithmic non-abelian Hodge correspondence as in [dCZ25, Theorem 3.24] over A a A_a A a .
Parabolic non-abelian Hodge without semistability
Before proving Theorem 5.2, we state and prove the following non-abelian Hodge correspondence for parabolic connections without imposing semistability.
Theorem 5.3. Let a ↦ α a \mapsto\alpha a ↦ α and b ↦ β b \mapsto\beta b ↦ β be two assignments related through (10). There exists an isomorphism of A a A_a A a -stacks
H α v g × P a ∘ H i g g s a → α ( C ′ ) ≅ C o n n b → β ( C ) . \mathcal{H}^{\mathrm{vg}}_{\alpha} \times_{\mathcal{P}^{\circ}_{a}} \mathcal{Higgs}^{a\to\alpha}(C') \cong\mathcal{Conn}^{b\to\beta}(C). H α vg × P a ∘ H i g g s a → α ( C ′ ) ≅ C o nn b → β ( C ) .
In [LS24, Theorem 5.25], Li and Sun constructed an isomorphism2
(11)
H × A P e x t X β , m ⟶ C o n n β ( C , m ) . \mathcal{H} \times_{A}^{\mathcal{P}^{\mathrm{ext}}} \mathcal{X}^{\beta,\mathbf{m}} \longrightarrow\mathcal{Conn}^{\beta}(C,\mathbf{m}). H × A P ext X β , m ⟶ C o nn β ( C , m ) .
Here the stack X β , m \mathcal{X}^{\beta,\mathbf{m}} X β , m classifies ( E , E ∙ , ∇ , ψ ) (E,E_{\bullet},\nabla,\psi) ( E , E ∙ , ∇ , ψ ) , where ( E , E ∙ , ∇ ) (E,E_{\bullet},\nabla) ( E , E ∙ , ∇ ) is a parabolic connection with parabolic weights β \beta β and filtration type m \mathbf{m} m such that its p p p -curvature is zero, and ψ : E → E ⊗ Fr C ∗ ( ω C ′ ( q ′ ) ) \psi:E\to E\otimes\operatorname{Fr}^{*}_{C}(\omega_{C'}(q')) ψ : E → E ⊗ Fr C ∗ ( ω C ′ ( q ′ )) is an O C \mathcal{O}_{C} O C -linear morphism satisfying
(1) The morphism ψ \psi ψ is horizontal with respect to the canonical connection on H o m ( E , E ⊗ Fr C ∗ ( ω C ′ ( q ′ ) ) ) \mathcal{H}om(E,E\otimes\operatorname{Fr}^{*}_{C}(\omega_{C'}(q'))) H o m ( E , E ⊗ Fr C ∗ ( ω C ′ ( q ′ ))) constructed from ∇ \nabla ∇ on E E E and the canonical connection on Fr C ∗ ( ω C ′ ( q ′ ) ) \operatorname{Fr}^{*}_{C}(\omega_{C'}(q')) Fr C ∗ ( ω C ′ ( q ′ )) .
(2) The residue res q ( ψ ) \operatorname{res}_{q}(\psi) res q ( ψ ) respects the flag structure E ∙ E_{\bullet} E ∙ .
The stack P e x t \mathcal{P}^{\mathrm{ext}} P ext classifies ( E , ∇ , ψ ) (E,\nabla,\psi) ( E , ∇ , ψ ) , where ( E , ∇ ) (E,\nabla) ( E , ∇ ) is a logarithmic connection with zero p p p -curvature, and ψ : E → E ⊗ Fr C ∗ ( ω C ′ ( q ′ ) ) \psi:E\to E\otimes\operatorname{Fr}^{*}_{C}(\omega_{C'}(q')) ψ : E → E ⊗ Fr C ∗ ( ω C ′ ( q ′ )) is an O C \mathcal{O}_{C} O C -linear morphism that satisfies the horizontal condition (1) above, as well as the extra requirement that the spectral sheaf of ψ \psi ψ is invertible. The stacks X β , m \mathcal{X}^{\beta,\mathbf{m}} X β , m and P e x t \mathcal{P}^{\mathrm{ext}} P ext map to the Hitchin base A A A by taking the coefficients of the characteristic polynomial of ψ \psi ψ .
The equivalence in [LS24] is described in terms of regular centralizers and parahoric group schemes. We give a more explicit description using spectral curves and parabolic structures as follows. Let E = ( E , ∇ E ) ∈ H \mathcal{E}=(E,\nabla_E)\in\mathcal{H} E = ( E , ∇ E ) ∈ H such that h d R ( E ) = a ∈ A h_{\mathrm{dR}}(\mathcal{E})=a\in A h dR ( E ) = a ∈ A , and let L E \mathcal{L}_{\mathcal{E}} L E be the spectral sheaf of its p p p -curvature ψ ∇ E \psi_{\nabla_E} ψ ∇ E . Let M = ( M , M ∙ , ∇ M , ψ ) ∈ X β , m \mathcal{M}=(M,M_{\bullet},\nabla_M,\psi)\in\mathcal{X}^{\beta,\mathbf{m}} M = ( M , M ∙ , ∇ M , ψ ) ∈ X β , m that maps to the same a ∈ A a\in A a ∈ A , and let F ψ \mathcal{F}_{\psi} F ψ be the spectral sheaf of ψ \psi ψ . We define
(12)
H × A X β , m ⟶ C o n n β ( C , m ) \mathcal{H}\times_{A}\mathcal{X}^{\beta,\mathbf{m}}\longrightarrow\mathcal{Conn}^{\beta}(C,\mathbf{m}) H × A X β , m ⟶ C o nn β ( C , m )
( E , M ) ↦ ( ( π ∗ F r ( L E ⊗ S a F r F ψ ) , ( π ∗ F r ( L E ⊗ S a F r F ψ ) ) ∙ , ∇ ⊗ ) ) , (\mathcal{E},\mathcal{M})\mapsto\left(\left(\pi^{\mathrm{Fr}}_{*}(\mathcal{L}_{\mathcal{E}}\otimes_{\mathcal{S}^{\mathrm{Fr}}_{a}}\mathcal{F}_{\psi}),\left(\pi^{\mathrm{Fr}}_{*}(\mathcal{L}_{\mathcal{E}}\otimes_{\mathcal{S}^{\mathrm{Fr}}_{a}}\mathcal{F}_{\psi})\right)_{\bullet},\nabla_{\otimes}\right)\right), ( E , M ) ↦ ( ( π ∗ Fr ( L E ⊗ S a Fr F ψ ) , ( π ∗ Fr ( L E ⊗ S a Fr F ψ ) ) ∙ , ∇ ⊗ ) ) ,
where ∇ ⊗ \nabla_{\otimes} ∇ ⊗ is the tensor product connection, and the parabolic structure ( π ∗ F r ( L E ⊗ S a F r F ψ ) ) ∙ \left(\pi^{\mathrm{Fr}}_{*}(\mathcal{L}_{\mathcal{E}}\otimes_{\mathcal{S}^{\mathrm{Fr}}_{a}}\mathcal{F}_{\psi})\right)_{\bullet} ( π ∗ Fr ( L E ⊗ S a Fr F ψ ) ) ∙ is induced from the parabolic structure on π ∗ F r F ψ \pi_{*}^{\mathrm{Fr}}\mathcal{F}_{\psi} π ∗ Fr F ψ .
Let N = ( N , N ∙ , V N ) ∈ Conn β ( C , m ) \mathcal{N}=(N,N_{\bullet},V_N)\in\operatorname{Conn}^{\beta}(\mathcal{C},\mathfrak{m}) N = ( N , N ∙ , V N ) ∈ Conn β ( C , m ) with h d R ( N ) = a h_{\mathrm{dR}}(\mathcal{N})=a h dR ( N ) = a , and let F N \mathcal{F}_{\mathcal{N}} F N be the spectral sheaf of its p p p -curvature ψ V N \psi_{V_N} ψ V N . We define a morphism in the opposite direction
H × A Conn β ( C , m ) ⟶ X β , m (11) \mathcal{H}\times_A \operatorname{Conn}^{\beta}(\mathcal{C},\mathfrak{m})\longrightarrow\mathcal{X}^{\beta,\mathfrak{m}} \tag*{(11)} H × A Conn β ( C , m ) ⟶ X β , m (11)
( E , N ) ↦ ( π ∗ F r ( L E − 1 ⊗ S a F r F N ) , ( π ∗ F r ( L E − 1 ⊗ S a F r F N ) ∙ , ∇ ⊗ , ψ ) (\mathcal{E},\mathcal{N})\mapsto(\pi_*^{\mathrm{Fr}}(L_{\mathcal{E}}^{-1}\otimes_{S_a^{\mathrm{Fr}}}\mathcal{F}_{\mathcal{N}}),(\pi_*^{\mathrm{Fr}}(L_{\mathcal{E}}^{-1}\otimes_{S_a^{\mathrm{Fr}}}\mathcal{F}_{\mathcal{N}})_{\bullet},\nabla_{\otimes},\psi) ( E , N ) ↦ ( π ∗ Fr ( L E − 1 ⊗ S a Fr F N ) , ( π ∗ Fr ( L E − 1 ⊗ S a Fr F N ) ∙ , ∇ ⊗ , ψ ) .
Here ∇ ⊗ \nabla_{\otimes} ∇ ⊗ is the tensor product connection, with the connection on L E − 1 L_{\mathcal{E}}^{-1} L E − 1 defined by the natural connection on H o m \mathcal{H}om H o m . The parabolic structure ( π ∗ F r ( L E − 1 ⊗ S a F r F N ) ∙ ) \left(\pi_*^{\mathrm{Fr}}\left(L_{\mathcal{E}}^{-1}\otimes_{S_a^{\mathrm{Fr}}}\mathcal{F}_{\mathcal{N}}\right)_{\bullet}\right) ( π ∗ Fr ( L E − 1 ⊗ S a Fr F N ) ∙ ) is induced from the parabolic structure on π ∗ F r F N \pi_*^{\mathrm{Fr}}\mathcal{F}_{\mathcal{N}} π ∗ Fr F N . The twisted Higgs field ψ \psi ψ is given by the action of the tautological section on S a F r S_a^{\mathrm{Fr}} S a Fr . Similar constructions to those in (12) and (13) endow P e x t \mathcal{P}^{\mathrm{ext}} P ext with the structure of a commutative group stack. They also define actions of P e x t \mathcal{P}^{\mathrm{ext}} P ext on X β , m \mathcal{X}^{\beta,\mathfrak{m}} X β , m and Conn β ( C , m ) \operatorname{Conn}^{\beta}(\mathcal{C},\mathfrak{m}) Conn β ( C , m ) , with respect to which (12) becomes P e x t \mathcal{P}^{\mathrm{ext}} P ext -equivariant, giving rise to the desired morphism (11). Combining [LS24, Lemma 5.18] and Theorem 2.2, we have that H \mathcal{H} H is a P e x t \mathcal{P}^{\mathrm{ext}} P ext -torsor, and it follows immediately that (11) is an isomorphism of stacks over A A A .
Note that the p p p -curvature of ( E , ∇ ) (E,\nabla) ( E , ∇ ) being zero implies that the residue res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) is semisimple and has eigenvalues in F p \mathbb{F}_p F p . An element τ ∈ t ( F p ) / W \tau\in\mathfrak{t}(\mathbb{F}_p)/W τ ∈ t ( F p ) / W corresponds to an unordered n n n -tuple of elements in F p \mathbb{F}_p F p . We denote by X β , m , τ ⊂ X β , m \mathcal{X}^{\beta,\mathfrak{m},\tau}\subset\mathcal{X}^{\beta,\mathfrak{m}} X β , m , τ ⊂ X β , m the substack classifying ( E , E ∙ , ∇ , ψ ) (E,E_{\bullet},\nabla,\psi) ( E , E ∙ , ∇ , ψ ) with res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) conjugate to τ \tau τ . Then we have
X β , m = ∐ τ ∈ t ( F p ) / W X β , m , τ . \mathcal{X}^{\beta,\mathfrak{m}}=\coprod_{\tau\in\mathfrak{t}(\mathbb{F}_p)/W}\mathcal{X}^{\beta,\mathfrak{m},\tau}. X β , m = τ ∈ t ( F p ) / W ∐ X β , m , τ .
Let τ ↦ β \tau\mapsto\beta τ ↦ β be an assignment of filtration type m \mathfrak{m} m . We define X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β to be the fiber product
X τ ↦ β : = X β , m , τ × Conn β ( C , m ) Conn τ ↦ β ( C ) , \mathcal{X}^{\tau\mapsto\beta}:=\mathcal{X}^{\beta,\mathfrak{m},\tau}\mathbin{\times}_{\operatorname{Conn}^{\beta}(\mathcal{C},\mathfrak{m})}\operatorname{Conn}^{\tau\mapsto\beta}(\mathcal{C}), X τ ↦ β := X β , m , τ × Conn β ( C , m ) Conn τ ↦ β ( C ) ,
where the morphism X β , m , τ → Conn β ( C , m ) \mathcal{X}^{\beta,\mathfrak{m},\tau}\to\operatorname{Conn}^{\beta}(\mathcal{C},\mathfrak{m}) X β , m , τ → Conn β ( C , m ) is given by forgetting ψ \psi ψ . Then we have
X β , m , τ = ∐ τ ↦ β X τ ↦ β , \mathcal{X}^{\beta,\mathfrak{m},\tau}=\coprod_{\tau\mapsto\beta}\mathcal{X}^{\tau\mapsto\beta}, X β , m , τ = τ ↦ β ∐ X τ ↦ β ,
where the disjoint union ranges over all assignments τ ↦ β \tau\mapsto\beta τ ↦ β of filtration type m \mathfrak{m} m .
Correspondence between X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β and parabolic Higgs bundles
Now we describe X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β . Let α ↦ α \alpha\mapsto\alpha α ↦ α be the assignment constructed from τ ↦ β \tau\mapsto\beta τ ↦ β through (10)
α ~ = AS ( τ ~ ) = 0 and α o p α ~ ( i ) = β ~ o p τ ~ ( i ) − ν ( τ i ) p , (12) \widetilde{\alpha}=\operatorname{AS}(\widetilde{\tau})=0\qquad\text{and}\qquad\alpha_{\mathrm{op}}^{\widetilde{\alpha}}(i)=\frac{\widetilde{\beta}_{\mathrm{op}}^{\widetilde{\tau}}(i)-\nu(\tau_i)}{p}, \tag*{(12)} α = AS ( τ ) = 0 and α op α ( i ) = p β op τ ( i ) − ν ( τ i ) , (12)
and we denote its filtration type by m ′ \mathfrak{m}' m ′ . A crucial intermediate result towards the proof of Theorem 5.3 is the following
Theorem 5.4. There exists an isomorphism X τ ↦ β ≃ Higgs α ( C ′ , m ′ ) \mathcal{X}^{\tau\mapsto\beta}\simeq\operatorname{Higgs}^{\alpha}(\mathcal{C}',\mathfrak{m}') X τ ↦ β ≃ Higgs α ( C ′ , m ′ ) . This isomorphism is P \mathcal{P} P -equivariant, where P \mathcal{P} P acts by tensoring spectral sheaves with line bundles on the spectral curves, using Frobenius pullback on the X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β side, with the parabolic structures transported accordingly.
Let Conn 0 τ ↦ β ( C ) ⊂ Conn τ ↦ β ( C ) \operatorname{Conn}_{0}^{\tau\mapsto\beta}(\mathcal{C})\subset\operatorname{Conn}^{\tau\mapsto\beta}(\mathcal{C}) Conn 0 τ ↦ β ( C ) ⊂ Conn τ ↦ β ( C ) denote the substack classifying connections with vanishing p p p -curvature. Equivalently, Conn 0 τ ↦ β ( C ) \operatorname{Conn}_{0}^{\tau\mapsto\beta}(\mathcal{C}) Conn 0 τ ↦ β ( C ) is obtained from X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β by forgetting the datum ψ \psi ψ . We fix a model object E 0 = ( E 0 , E ∙ 0 , ∇ 0 ) ∈ Conn 0 τ ↦ β ( C ) ( k ) \mathcal{E}^{0}=(E^{0},E_{\bullet}^{0},\nabla^{0})\in\operatorname{Conn}_{0}^{\tau\mapsto\beta}(\mathcal{C})(k) E 0 = ( E 0 , E ∙ 0 , ∇ 0 ) ∈ Conn 0 τ ↦ β ( C ) ( k ) defined as follows. Let τ ~ = ( τ 1 , τ 2 , … , τ n ) ∈ t ( F p ) \widetilde{\tau}=(\tau_1,\tau_2,\ldots,\tau_n)\in\mathfrak{t}(\mathbb{F}_p) τ = ( τ 1 , τ 2 , … , τ n ) ∈ t ( F p ) be a lift of τ ∈ t ( F p ) / W \tau\in\mathfrak{t}(\mathbb{F}_p)/W τ ∈ t ( F p ) / W . Let E 0 = ⨁ i = 1 n O C ( − ν ( τ i ) q ) E^{0}=\bigoplus_{i=1}^{n}\mathcal{O}_{\mathcal{C}}(-\nu(\tau_i)q) E 0 = ⨁ i = 1 n O C ( − ν ( τ i ) q ) , let ∇ 0 \nabla^{0} ∇ 0 be the logarithmic connection on E 0 E^{0} E 0 induced from the canonical connection on O C \mathcal{O}_{\mathcal{C}} O C , and let E ∙ 0 E_{\bullet}^{0} E ∙ 0 be the partial flag structure on E q 0 E^{0}_{q} E q 0 obtained by assigning the parabolic weight β ~ o p τ ~ ( i ) \widetilde{\beta}_{\mathrm{op}}^{\widetilde{\tau}}(i) β op τ ( i ) to the summand O C ( − ν ( τ i ) q ) q ⊂ E q 0 \mathcal{O}_{\mathcal{C}}(-\nu(\tau_i)q)_{q}\subset E^{0}_{q} O C ( − ν ( τ i ) q ) q ⊂ E q 0 . Let S S S be a k k k -scheme, and let E ∈ C o n n 0 τ ↦ β ( C ) ( S ) \mathcal{E} \in\mathop{\mathrm{Conn}}\nolimits_{0}^{\tau\mapsto\beta}(C)(S) E ∈ Conn 0 τ ↦ β ( C ) ( S ) . We denote by E S 0 \mathcal{E}_{S}^{0} E S 0 the pullback of E 0 \mathcal{E}^{0} E 0 along the projection p C : S × C → C p_{C}: S \times C \to C p C : S × C → C . Let I s o m ( E S 0 , E ) \mathop{\mathrm{Isom}}\nolimits(\mathcal{E}_{S}^{0},\mathcal{E}) Isom ( E S 0 , E ) denote the sheaf of isomorphisms from E S 0 \mathcal{E}_{S}^{0} E S 0 to E \mathcal{E} E over S × C S \times C S × C . The sheaf I s o m ( E S 0 , E ) \mathop{\mathrm{Isom}}\nolimits(\mathcal{E}_{S}^{0},\mathcal{E}) Isom ( E S 0 , E ) carries a natural right action of the sheaf A u t ( E S 0 ) \mathop{\mathrm{Aut}}\nolimits(\mathcal{E}_{S}^{0}) Aut ( E S 0 ) of automorphisms of E S 0 \mathcal{E}_{S}^{0} E S 0 , given by precomposition:
φ ⋅ g : = φ ∘ g , φ ∈ I s o m ( E S 0 , E ) , g ∈ A u t ( E S 0 ) . \varphi\cdot g := \varphi\circ g,\qquad\varphi\in\mathop{\mathrm{Isom}}\nolimits(\mathcal{E}_{S}^{0},\mathcal{E}),\qquad g \in\mathop{\mathrm{Aut}}\nolimits(\mathcal{E}_{S}^{0}). φ ⋅ g := φ ∘ g , φ ∈ Isom ( E S 0 , E ) , g ∈ Aut ( E S 0 ) .
The following lemma is a parabolic analogue of [LS24, Lemma 5.21]; see also [LS24, Lemma 5.26]. We give a direct proof in our setting.
Lemma 5.5. Let E ∈ C o n n 0 τ ↦ β ( C ) ( S ) \mathcal{E} \in\mathop{\mathrm{Conn}}\nolimits_{0}^{\tau\mapsto\beta}(C)(S) E ∈ Conn 0 τ ↦ β ( C ) ( S ) . Then I s o m ( E S 0 , E ) \mathop{\mathrm{Isom}}\nolimits(\mathcal{E}_{S}^{0},\mathcal{E}) Isom ( E S 0 , E ) is an A u t ( E S 0 ) \mathop{\mathrm{Aut}}\nolimits(\mathcal{E}_{S}^{0}) Aut ( E S 0 ) -torsor.
Proof. It is enough to show that there exists a Zariski open cover { U i → S × C } i ∈ I \{U_i \to S \times C\}_{i \in I} { U i → S × C } i ∈ I such that E ∣ U i \mathcal{E}|_{U_i} E ∣ U i and E S 0 ∣ U i \mathcal{E}_{S}^{0}|_{U_i} E S 0 ∣ U i are isomorphic as objects of C o n n 0 τ ↦ β ( C ) \mathop{\mathrm{Conn}}\nolimits_{0}^{\tau\mapsto\beta}(C) Conn 0 τ ↦ β ( C ) for every i ∈ I i \in I i ∈ I . Away from S × q S \times q S × q , the claim follows from Cartier descent: connections with vanishing p p p -curvature are obtained by Frobenius pullback of vector bundles on the Frobenius twist, which are Zariski locally trivial; see [Kat70, Theorem 5.1]. Now we construct a Zariski open neighborhood V ⊂ S × C V \subset S \times C V ⊂ S × C of S × q S \times q S × q , together with an isomorphism
Ξ : E S 0 ∣ V ⟶ E ∣ V . \Xi: \mathcal{E}_{S}^{0}|_{V} \longrightarrow\mathcal{E}|_{V}. Ξ : E S 0 ∣ V ⟶ E ∣ V .
Working Zariski locally on S S S , we may assume that S = Spec ( R ) S = \operatorname{Spec}(R) S = Spec ( R ) for a k k k -algebra R R R . Let x x x be a local parameter around q q q . Let E = ( E , E ∙ , ∇ ) \mathcal{E} = (E,E_{\bullet},\nabla) E = ( E , E ∙ , ∇ ) . We denote Δ : = ∇ x ∂ x \Delta:= \nabla_{x\partial_x} Δ := ∇ x ∂ x , which is an R R R -linear operator acting on E E E . Since x ∂ x x\partial_x x ∂ x preserves the ideal ( x ) (x) ( x ) , Δ \Delta Δ induces an endomorphism Δ q \Delta_q Δ q of E q : = E ∣ S × q E_q := E|_{S \times q} E q := E ∣ S × q , and Δ q = res q ( ∇ ) \Delta_q = \operatorname{res}_{q}(\nabla) Δ q = res q ( ∇ ) . Since E \mathcal{E} E has vanishing p p p -curvature, we have ψ ∇ ( x ∂ x ) = Δ p − Δ = 0 \psi_{\nabla}(x\partial_x) = \Delta^p - \Delta= 0 ψ ∇ ( x ∂ x ) = Δ p − Δ = 0 . For each c ∈ F p c \in\mathbb{F}_p c ∈ F p , let P c ( T ) P_c(T) P c ( T ) be the Lagrange interpolation polynomial
P c ( T ) = ∏ a ∈ F p a ≠ c T − a c − a ∈ F p [ T ] . P_c(T) = \prod_{\substack{a \in\mathbb{F}_p\\a\ne c}} \frac{T-a}{c-a} \in\mathbb{F}_p[T]. P c ( T ) = a ∈ F p a = c ∏ c − a T − a ∈ F p [ T ] .
Modulo T p − T T^p-T T p − T , these polynomials satisfy
( T − c ) P c ( T ) = 0 , P c ( T ) 2 = P c ( T ) , ∑ c ∈ F p P c ( T ) = 1. (T-c)P_c(T)=0,\qquad P_c(T)^2=P_c(T),\qquad\sum_{c\in\mathbb{F}_p}P_c(T)=1. ( T − c ) P c ( T ) = 0 , P c ( T ) 2 = P c ( T ) , c ∈ F p ∑ P c ( T ) = 1.
For each c ∈ F p c \in\mathbb{F}_p c ∈ F p , the idempotent P c ( Δ q ) P_c(\Delta_q) P c ( Δ q ) satisfies
Im P c ( Δ q ) = ker ( Δ q − c ) = ker ( res q ( ∇ ) − c ) . \operatorname{Im}P_c(\Delta_q)=\ker(\Delta_q-c)=\ker(\operatorname{res}_{q}(\nabla)-c). Im P c ( Δ q ) = ker ( Δ q − c ) = ker ( res q ( ∇ ) − c ) .
Hence the eigenspace decomposition of Δ q \Delta_q Δ q acting on E q E_q E q is given by
E q = ⨁ c ∈ F p Im P c ( Δ q ) . E_q=\bigoplus_{c\in\mathbb{F}_p}\operatorname{Im}P_c(\Delta_q). E q = c ∈ F p ⨁ Im P c ( Δ q ) .
Since res q ( ∇ ) = Δ q \operatorname{res}_{q}(\nabla)=\Delta_q res q ( ∇ ) = Δ q respects the parabolic structure E ∙ E_{\bullet} E ∙ , the eigenspace decomposition is compatible with the parabolic filtration. It follows that the parabolic weights and filtration type of the induced parabolic structure on each Im P c ( Δ q ) \operatorname{Im}P_c(\Delta_q) Im P c ( Δ q ) are given by
{ β op τ ( i ) ∣ τ i = c } \{\beta_{\operatorname{op}}^{\tau}(i)\mid\tau_i=c\} { β op τ ( i ) ∣ τ i = c }
counted with multiplicity.
After possibly shrinking S S S to a smaller affine open subset, we choose a basis e = ( e 1 , e 2 , … , e n ) e=(e_1,e_2,\ldots,e_n) e = ( e 1 , e 2 , … , e n ) of E q E_q E q that satisfies the following two properties:
For each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n , e i ∈ Im P τ i ( Δ q ) e_i \in\operatorname{Im}P_{\tau_i}(\Delta_q) e i ∈ Im P τ i ( Δ q ) .
(2) For each c ∈ F p c \in\mathbb{F}_p c ∈ F p and 1 ≤ j ≤ r 1 \le j \le r 1 ≤ j ≤ r , the images of
{ e i ∣ τ i = c , β ~ o p τ ( i ) = β j } \left\{e_i \mid\tau_i=c,\ \widetilde{\beta}_{\mathrm{op}}^{\tau}(i)=\beta_j\right\} { e i ∣ τ i = c , β op τ ( i ) = β j }
form a basis of the graded quotient gr β j ( Im P c ( Δ q ) ) \operatorname{gr}_{\beta_j}(\operatorname{Im} P_c(\Delta_q)) gr β j ( Im P c ( Δ q )) .
Now let U ⊂ C U \subset C U ⊂ C be an affine open neighborhood of q q q on which the chosen local parameter x x x is defined and such that d x x \frac{dx}{x} x d x trivializes ω C ( q ) ∣ U \omega_C(q)|_U ω C ( q ) ∣ U . Since S × U S \times U S × U is affine, for each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n we may lift e i ∈ Im P τ i ( Δ q ) e_i \in\operatorname{Im} P_{\tau_i}(\Delta_q) e i ∈ Im P τ i ( Δ q ) to an element e ~ i ∈ Im P τ i ( Δ ∣ U ) \widetilde{e}_i \in\operatorname{Im} P_{\tau_i}(\Delta|_U) e i ∈ Im P τ i ( Δ ∣ U ) . We denote the resulting lifts by e ~ = ( e ~ 1 , e ~ 2 , … , e ~ n ) \widetilde{e}=(\widetilde{e}_1,\widetilde{e}_2,\ldots,\widetilde{e}_n) e = ( e 1 , e 2 , … , e n ) . Let V ⊂ S × C V \subset S \times C V ⊂ S × C be the Zariski open locus on which ( e ~ 1 , e ~ 2 , … , e ~ n ) (\widetilde{e}_1,\widetilde{e}_2,\ldots,\widetilde{e}_n) ( e 1 , e 2 , … , e n ) forms a basis of E E E . Since ( e 1 , … , e n ) (e_1,\ldots,e_n) ( e 1 , … , e n ) is a basis of E q E_q E q , the open subset V V V contains S × q S \times q S × q . Since ( Δ − c ) P c ( Δ ) = 0 (\Delta-c)P_c(\Delta)=0 ( Δ − c ) P c ( Δ ) = 0 , we have Δ ( e ~ i ) = τ i e ~ i \Delta(\widetilde{e}_i)=\tau_i\widetilde{e}_i Δ ( e i ) = τ i e i for each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n . Now we construct the desired isomorphism Ξ : E S 0 ∣ V ⟶ E ∣ V \Xi:\mathcal{E}^0_S|_V \longrightarrow\mathcal{E}|_V Ξ : E S 0 ∣ V ⟶ E ∣ V . Recall that E 0 = ( E 0 , E ∙ 0 , ∇ 0 ) \mathcal{E}^0=(E^0,E^0_\bullet,\nabla^0) E 0 = ( E 0 , E ∙ 0 , ∇ 0 ) with E 0 = ⨁ i = 1 n O C ( − ν ( τ i ) q ) E^0=\bigoplus_{i=1}^n\mathcal{O}_C(-\nu(\tau_i)q) E 0 = ⨁ i = 1 n O C ( − ν ( τ i ) q ) . Over the open subset U ⊂ C U \subset C U ⊂ C , for each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n , we set
e ~ i 0 : = x ν ( τ i ) ⊗ O C ( − ν ( τ i ) q ) ∣ U . \widetilde{e}_i^0:=x^{\nu(\tau_i)}\otimes\mathcal{O}_C(-\nu(\tau_i)q)|_U. e i 0 := x ν ( τ i ) ⊗ O C ( − ν ( τ i ) q ) ∣ U .
Note that ( e ~ 1 0 , e ~ 2 0 , … , e ~ n 0 ) (\widetilde{e}_1^0,\widetilde{e}_2^0,\ldots,\widetilde{e}_n^0) ( e 1 0 , e 2 0 , … , e n 0 ) forms a basis of E 0 ∣ U E^0|_U E 0 ∣ U , and Δ ( e ~ i 0 ) = τ i e ~ i 0 \Delta(\widetilde{e}_i^0)=\tau_i\widetilde{e}_i^0 Δ ( e i 0 ) = τ i e i 0 for each 1 ≤ i ≤ n 1 \le i \le n 1 ≤ i ≤ n . Now define
Ξ : E S 0 ∣ V ⟶ E ∣ V \Xi:\mathcal{E}^0_S|_V \longrightarrow\mathcal{E}|_V Ξ : E S 0 ∣ V ⟶ E ∣ V
by setting
Ξ ( e ~ i 0 ) = e ~ i , 1 ≤ i ≤ n . \Xi(\widetilde{e}_i^0)=\widetilde{e}_i,\qquad1 \le i \le n. Ξ ( e i 0 ) = e i , 1 ≤ i ≤ n .
It is immediate from the construction that Ξ \Xi Ξ is an isomorphism of logarithmic connections
( E S 0 , ∇ S 0 ) ∣ V ⟶ ( E , ∇ ) ∣ V . (E^0_S,\nabla^0_S)|_V \longrightarrow(E,\nabla)|_V. ( E S 0 , ∇ S 0 ) ∣ V ⟶ ( E , ∇ ) ∣ V .
It also respects the parabolic structures, because the parabolic structure on E 0 \mathcal{E}^0 E 0 (resp. E \mathcal{E} E ) is defined by assigning the parabolic weight β ~ o p τ ( i ) \widetilde{\beta}_{\mathrm{op}}^\tau(i) β op τ ( i ) to e ~ i 0 ∣ q \widetilde{e}_i^0|_q e i 0 ∣ q (resp. e ~ i ∣ q \widetilde{e}_i|_q e i ∣ q ) for each i i i . □
Recall that the parabolic weights α \alpha α and filtration type m ′ \boldsymbol{m}' m ′ arise from the assignment a ↦ α \boldsymbol{a}\mapsto\alpha a ↦ α constructed from τ ↦ β \tau\mapsto\beta τ ↦ β through (14). Let α ~ = ( α ~ o p α ( 1 ) , α ~ o p α ( 2 ) , … , α ~ o p α ( n ) ) \widetilde{\alpha}=(\widetilde{\alpha}_{\mathrm{op}}^\alpha(1),\widetilde{\alpha}_{\mathrm{op}}^\alpha(2),\ldots,\widetilde{\alpha}_{\mathrm{op}}^\alpha(n)) α = ( α op α ( 1 ) , α op α ( 2 ) , … , α op α ( n )) . Then we have
Lemma 5.6. There is a canonical isomorphism of sheaves of groups on C ′ C' C ′
( Fr C ) ∗ Aut ( E 0 ) ⟶ G α ~ . (13) (\operatorname{Fr}_C)_*\operatorname{Aut}(\mathcal{E}^0) \longrightarrow\mathcal{G}_{\widetilde{\alpha}}. \tag*{(13)} ( Fr C ) ∗ Aut ( E 0 ) ⟶ G α . (13)
Proof. Away from q ′ q' q ′ , the sheaf of groups ( Fr C ) ∗ Aut ( E 0 ) (\operatorname{Fr}_C)_*\operatorname{Aut}(\mathcal{E}^0) ( Fr C ) ∗ Aut ( E 0 ) is canonically isomorphic to G L n \mathrm{GL}_n GL n by Cartier descent, see [Kat70, Theorem 5.1]. Therefore we have an isomorphism
( Fr C ) ∗ Aut ( E 0 ) ∣ C ′ ∖ q ′ → ∼ G α ~ ∣ C ′ ∖ q ′ . (\operatorname{Fr}_C)_*\operatorname{Aut}(\mathcal{E}^0)|_{C'\setminus q'} \xrightarrow{\sim} \mathcal{G}_{\widetilde{\alpha}}|_{C'\setminus q'}. ( Fr C ) ∗ Aut ( E 0 ) ∣ C ′ ∖ q ′ ∼ G α ∣ C ′ ∖ q ′ .
The fact that this isomorphism extends over q ′ q' q ′ essentially follows from the computation in [LS24, Lemma 4.14], which we spell out in detail for the reader’s convenience. Let D q = Spec k [ [ x ] ] D_q=\operatorname{Spec} k[[x]] D q = Spec k [[ x ]] be the formal disc around q ∈ C q \in C q ∈ C . An element
g ∈ Aut ( E 0 ) ∣ D q ⊂ G L n ( k ( ( x ) ) ) g \in\operatorname{Aut}(\mathcal{E}^0)|_{D_q} \subset\mathrm{GL}_n(k((x))) g ∈ Aut ( E 0 ) ∣ D q ⊂ GL n ( k (( x )))
is characterized by the following two conditions:
(1) g g g preserves the connection, i.e.
(2) g g g preserves the lattice underlying E 0 ∣ D q E^0|_{D_q} E 0 ∣ D q and the parabolic structure at q q q , i.e.
x − ν ( τ ~ ) g x ν ( τ ~ ) ∈ G β ~ ( k [ [ x ] ] ) , where β ~ = ( β ~ o p τ ( 1 ) , β ~ o p τ ( 2 ) , … , β ~ o p τ ( n ) ) . x^{-\nu(\widetilde{\tau})}gx^{\nu(\widetilde{\tau})}\in\mathcal{G}_{\widetilde{\beta}}(k[[x]]),\qquad\text{where }\widetilde{\beta}=(\widetilde{\beta}_{\mathrm{op}}^\tau(1),\widetilde{\beta}_{\mathrm{op}}^\tau(2),\ldots,\widetilde{\beta}_{\mathrm{op}}^\tau(n)). x − ν ( τ ) g x ν ( τ ) ∈ G β ( k [[ x ]]) , where β = ( β op τ ( 1 ) , β op τ ( 2 ) , … , β op τ ( n )) .
We write g = ( g i j ) g=(g_{ij}) g = ( g ij ) , then condition (1) is equivalent to
g i j = ∑ ℓ a i j ( ℓ ) x ℓ . g_{ij}=\sum_{\ell}a_{ij}^{(\ell)}x^{\ell}. g ij = ℓ ∑ a ij ( ℓ ) x ℓ .
Note that x − ν ( τ ) g x ν ( τ ) = ( g i j ′ ) x^{-\nu(\tau)}gx^{\nu(\tau)}=(g'_{ij}) x − ν ( τ ) g x ν ( τ ) = ( g ij ′ ) , where g i j ′ = ∑ ℓ a i j ( ℓ ) x ℓ p − ν ( τ i ) + ν ( τ j ) g'_{ij}=\sum_{\ell}a_{ij}^{(\ell)}x^{\ell p-\nu(\tau_i)+\nu(\tau_j)} g ij ′ = ∑ ℓ a ij ( ℓ ) x ℓ p − ν ( τ i ) + ν ( τ j ) . Therefore conditions (1) and (2) combine to
ℓ p − ν ( τ i ) + ν ( τ j ) ≥ ∣ β op τ ( j ) − β op τ ( i ) ∣ whenever a i j ( ℓ ) ≠ 0 , det ( g ) ∈ k ⟦ x p ⟧ × . (14) \ell p-\nu(\tau_i)+\nu(\tau_j)\geq\left|\beta_{\operatorname{op}}^{\tau}(j)-\beta_{\operatorname{op}}^{\tau}(i)\right|\qquad\text{whenever }a_{ij}^{(\ell)}\ne0,\qquad\det(g)\in k\llbracket x^p\rrbracket^\times. \tag*{(14)} ℓ p − ν ( τ i ) + ν ( τ j ) ≥ β op τ ( j ) − β op τ ( i ) whenever a ij ( ℓ ) = 0 , det ( g ) ∈ k [ [ x p ] ] × . (14)
It follows that (16) is equivalent to g ∈ G α ( k ⟦ x p ⟧ ) g\in\mathcal{G}_{\alpha}(k\llbracket x^p\rrbracket) g ∈ G α ( k [ [ x p ] ] ) . The same calculation works over an arbitrary k k k -algebra R R R .
Before proving Theorem 5.4, we first establish an isomorphism between the stacks obtained from the two sides of Theorem 5.4 by forgetting the datum ψ \psi ψ and the Higgs field, respectively.
Proposition 5.7. For any τ ∈ t ( F p ) / W \tau\in\mathfrak{t}(\mathbb{F}_p)/W τ ∈ t ( F p ) / W , parabolic weights β \beta β , and assignment τ ↦ β \tau\mapsto\beta τ ↦ β , there exists an isomorphism
C o n n 0 τ ↦ β ( C ) ≅ B u n α ( C ′ , m ′ ) , (15) \mathcal{C}onn_{0}^{\tau\mapsto\beta}(C)\cong\mathcal{B}un^{\alpha}(C',\mathbf{m}'), \tag*{(15)} C o n n 0 τ ↦ β ( C ) ≅ B u n α ( C ′ , m ′ ) , (15) where the parabolic weights α \alpha α and filtration type m ′ \mathbf{m}' m ′ arise from the assignment α ↦ α \alpha\mapsto\alpha α ↦ α constructed from τ ↦ β \tau\mapsto\beta τ ↦ β through (14).
Proof. For a sheaf of groups G \mathcal{G} G on C C C , we denote by B u n G ( C ) \mathcal{B}un_{\mathcal{G}}(C) B u n G ( C ) the moduli stack of G \mathcal{G} G -torsors on C C C . By Lemma 5.6 and Remark 4.1, we have
B u n α ( C ′ , m ′ ) ≅ B u n ( F Γ C ) ∗ Aut ( E 0 ) ( C ′ ) . \mathcal{B}un^{\alpha}(C',\mathbf{m}')\cong\mathcal{B}un_{(\mathcal{F}_{\Gamma_C})_*\operatorname{Aut}(\mathcal{E}^{0})}(C'). B u n α ( C ′ , m ′ ) ≅ B u n ( F Γ C ) ∗ Aut ( E 0 ) ( C ′ ) .
Note that F Γ C : C → C ′ \mathcal{F}_{\Gamma_C}:C\to C' F Γ C : C → C ′ induces an equivalence of étale sites, therefore we have
B u n ( F Γ C ) ∗ Aut ( E 0 ) ( C ′ ) ≅ B u n Aut ( E 0 ) ( C ) . \mathcal{B}un_{(\mathcal{F}_{\Gamma_C})_*\operatorname{Aut}(\mathcal{E}^{0})}(C')\cong\mathcal{B}un_{\operatorname{Aut}(\mathcal{E}^{0})}(C). B u n ( F Γ C ) ∗ Aut ( E 0 ) ( C ′ ) ≅ B u n Aut ( E 0 ) ( C ) .
It follows that the desired (17) is equivalent to
C o n n 0 τ ↦ β ( C ) ≅ B u n Aut ( E 0 ) ( C ) . \mathcal{C}onn_{0}^{\tau\mapsto\beta}(C)\cong\mathcal{B}un_{\operatorname{Aut}(\mathcal{E}^{0})}(C). C o n n 0 τ ↦ β ( C ) ≅ B u n Aut ( E 0 ) ( C ) .
Applying Lemma 5.5, we define
F : C o n n 0 τ ↦ β ( C ) ⟶ B u n Aut ( E 0 ) ( C ) , \mathsf{F}:\mathcal{C}onn_{0}^{\tau\mapsto\beta}(C)\longrightarrow\mathcal{B}un_{\operatorname{Aut}(\mathcal{E}^{0})}(C), F : C o n n 0 τ ↦ β ( C ) ⟶ B u n Aut ( E 0 ) ( C ) ,
E ⟼ Isom ( E 0 , E ) . \mathcal{E}\longmapsto\operatorname{Isom}(\mathcal{E}^{0},\mathcal{E}). E ⟼ Isom ( E 0 , E ) .
In the reverse direction, we define
G : B u n Aut ( E 0 ) ( C ) ⟶ C o n n 0 τ ↦ β ( C ) , \mathsf{G}:\mathcal{B}un_{\operatorname{Aut}(\mathcal{E}^{0})}(C)\longrightarrow\mathcal{C}onn_{0}^{\tau\mapsto\beta}(C), G : B u n Aut ( E 0 ) ( C ) ⟶ C o n n 0 τ ↦ β ( C ) ,
P ⟼ P × Aut ( E 0 ) E 0 , \mathcal{P}\longmapsto\mathcal{P}\mathbin{\times}_{\operatorname{Aut}(\mathcal{E}^{0})}\mathcal{E}^{0}, P ⟼ P × Aut ( E 0 ) E 0 ,
with connection and parabolic structure inherited from E 0 \mathcal{E}^{0} E 0 . Consider the natural morphism
ε E : G ∘ F ( E ) = Isom ( E 0 , E ) × Aut ( E 0 ) E 0 ⟶ E \varepsilon_{\mathcal{E}}:\mathsf{G}\circ\mathsf{F}(\mathcal{E})=\operatorname{Isom}(\mathcal{E}^{0},\mathcal{E})\mathbin{\times}^{\operatorname{Aut}(\mathcal{E}^{0})}\mathcal{E}^{0}\longrightarrow\mathcal{E} ε E : G ∘ F ( E ) = Isom ( E 0 , E ) × Aut ( E 0 ) E 0 ⟶ E
induced by evaluation, sending ( φ , e ) (\varphi,e) ( φ , e ) to φ ( e ) \varphi(e) φ ( e ) . It follows from Lemma 5.5 that ε E \varepsilon_{\mathcal{E}} ε E is an isomorphism of parabolic connections. Moreover, under these identifications, the maps on Hom-sets induced by F \mathsf{F} F and G \mathsf{G} G are inverse to each other. Hence F \mathsf{F} F and G \mathsf{G} G are isomorphisms of stacks that are inverse to each other.
Proof of Theorem 5.4. We now obtain the desired isomorphism
X τ ↦ β ≅ Higgs α ( C ′ , m ′ ) \mathcal{X}^{\tau\mapsto\beta} \cong\operatorname{Higgs}^{\alpha}(C',m') X τ ↦ β ≅ Higgs α ( C ′ , m ′ )
from the isomorphism (17) in Theorem 5.7. Recall that Conn 0 τ ↦ β ( C ) \operatorname{Conn}_{0}^{\tau\mapsto\beta}(C) Conn 0 τ ↦ β ( C ) is obtained from X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β by forgetting the datum ψ \psi ψ , which is an O C \mathcal{O}_{C} O C -linear morphism ψ : E → E ⊗ Fr C ∗ ( ω C ′ ( q ′ ) ) \psi:E\to E\otimes\operatorname{Fr}_{C}^{*}(\omega_{C'}(q')) ψ : E → E ⊗ Fr C ∗ ( ω C ′ ( q ′ )) satisfying the two conditions (1) and (2) described earlier. For any E = ( E , E ∙ , ∇ ) ∈ Conn 0 τ ↦ β ( C ) \mathcal{E}=(E,E_{\bullet},\nabla)\in\operatorname{Conn}_{0}^{\tau\mapsto\beta}(C) E = ( E , E ∙ , ∇ ) ∈ Conn 0 τ ↦ β ( C ) and any line bundle L \mathcal{L} L on C ′ C' C ′ , we have
E ⊗ Fr C ∗ L : = ( E ⊗ Fr C ∗ L , E ∙ ⊗ Fr C ∗ L , ∇ ⊗ 1 + 1 ⊗ ∇ c a n ) ∈ Conn 0 τ ↦ β ( C ) \mathcal{E}\otimes\operatorname{Fr}_{C}^{*}\mathcal{L}:=(E\otimes\operatorname{Fr}_{C}^{*}\mathcal{L},E_{\bullet}\otimes\operatorname{Fr}_{C}^{*}\mathcal{L},\nabla\otimes1+1\otimes\nabla_{\mathrm{can}})\in\operatorname{Conn}_{0}^{\tau\mapsto\beta}(C) E ⊗ Fr C ∗ L := ( E ⊗ Fr C ∗ L , E ∙ ⊗ Fr C ∗ L , ∇ ⊗ 1 + 1 ⊗ ∇ can ) ∈ Conn 0 τ ↦ β ( C )
where ∇ c a n \nabla_{\mathrm{can}} ∇ can denotes the canonical connection on Fr C ∗ L \operatorname{Fr}_{C}^{*}\mathcal{L} Fr C ∗ L . Note that the isomorphism (15)
( Fr C ) ∗ A u t ( E 0 ) ≅ G α ~ (\operatorname{Fr}_{C})_{*}\mathop{\mathrm{Aut}}(\mathcal{E}^{0})\cong\mathcal{G}_{\widetilde{\alpha}} ( Fr C ) ∗ Aut ( E 0 ) ≅ G α
in Theorem 5.6 identifies the central scalar automorphisms on both sides. Therefore under the isomorphism (17)
Conn 0 τ ↦ β ( C ) ≅ Bun α ( C ′ , m ′ ) , \operatorname{Conn}_{0}^{\tau\mapsto\beta}(C)\cong\operatorname{Bun}^{\alpha}(C',m'), Conn 0 τ ↦ β ( C ) ≅ Bun α ( C ′ , m ′ ) ,
if E \mathcal{E} E corresponds to ( F , F ∙ ) (F,F_{\bullet}) ( F , F ∙ ) , then E ⊗ Fr C ∗ L \mathcal{E}\otimes\operatorname{Fr}_{C}^{*}\mathcal{L} E ⊗ Fr C ∗ L corresponds to ( F ⊗ L , F ∙ ⊗ L ) (F\otimes\mathcal{L},F_{\bullet}\otimes\mathcal{L}) ( F ⊗ L , F ∙ ⊗ L ) . Now set L = ω C ′ ( q ′ ) \mathcal{L}=\omega_{C'}(q') L = ω C ′ ( q ′ ) . Then the two conditions on ψ \psi ψ are equivalent to requiring that
ψ : E ⟶ E ⊗ Fr C ∗ ( ω C ′ ( q ′ ) ) \psi:\mathcal{E}\longrightarrow\mathcal{E}\otimes\operatorname{Fr}_{C}^{*}\bigl(\omega_{C'}(q')\bigr) ψ : E ⟶ E ⊗ Fr C ∗ ( ω C ′ ( q ′ ) )
be a morphism of parabolic connections. Under the isomorphism (17), ψ \psi ψ corresponds to a morphism F → F ⊗ ω C ′ ( q ′ ) F\to F\otimes\omega_{C'}(q') F → F ⊗ ω C ′ ( q ′ ) that respects the parabolic structures F ∙ F_{\bullet} F ∙ and F ∙ ⊗ ω C ′ ( q ′ ) F_{\bullet}\otimes\omega_{C'}(q') F ∙ ⊗ ω C ′ ( q ′ ) , which is precisely a parabolic Higgs field. The P \mathcal{P} P -equivariance follows from the isomorphism (15)
( Fr C ) ∗ A u t ( E 0 ) ≅ G α ~ (\operatorname{Fr}_{C})_{*}\mathop{\mathrm{Aut}}(\mathcal{E}^{0})\cong\mathcal{G}_{\widetilde{\alpha}} ( Fr C ) ∗ Aut ( E 0 ) ≅ G α
and the induced identification of their Lie algebras, which matches ψ \psi ψ with the parabolic Higgs field on the other side after locally trivializing ω C ′ ( q ′ ) \omega_{C'}(q') ω C ′ ( q ′ ) . Consequently, tensoring the spectral sheaf of ψ \psi ψ by W ∗ L W^{*}\mathcal{L} W ∗ L corresponds to tensoring the spectral sheaf of the associated parabolic Higgs field by L \mathcal{L} L . □
Proof of Theorem 5.3.
Now we fix a ∈ t ( k ) / W \mathbf{a}\in\mathfrak{t}(\mathbf{k})/W a ∈ t ( k ) / W . Let X τ ↦ β = X × A a τ ↦ β \mathcal{X}^{\tau\mapsto\beta}=\mathcal{X}^{\tau\mapsto\beta}_{\times A_{\mathbf{a}}} X τ ↦ β = X × A a τ ↦ β be the restriction of X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β to A a A_{\mathbf{a}} A a . The stack X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β admits a further decomposition into a disjoint union, which we now describe. Let ( E , E ∙ , ∇ , ψ ) ∈ X τ ↦ β (E,E_{\bullet},\nabla,\psi)\in\mathcal{X}^{\tau\mapsto\beta} ( E , E ∙ , ∇ , ψ ) ∈ X τ ↦ β . Since X τ ↦ β \mathcal{X}^{\tau\mapsto\beta} X τ ↦ β lies over A a A_{\mathbf{a}} A a , the residue res q ( ψ ) \operatorname{res}_{q}(\psi) res q ( ψ ) has eigenvalues a \mathbf{a} a . Since ψ \psi ψ is horizontal with respect to the canonical logarithmic connection constructed from ∇ \nabla ∇ , res q ( ψ ) \operatorname{res}_{q}(\psi) res q ( ψ ) and res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) are commuting matrices. By definition, both res q ( ψ ) \operatorname{res}_{q}(\psi) res q ( ψ ) and res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) respect the flag structure E ∙ E_{\bullet} E ∙ . It follows that the action of res q ( ψ ) \operatorname{res}_{q}(\psi) res q ( ψ ) and res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) induces a decomposition into generalized joint eigenspaces
E q = ⨁ ( a , τ ) ∈ a × τ V a , τ . E_{q}=\bigoplus_{(\mathbf{a},\tau)\in\mathbf{a}\times\tau}V^{\mathbf{a},\tau}. E q = ( a , τ ) ∈ a × τ ⨁ V a , τ .
and the parabolic structure ( E q , E ∙ , B ) (E_{q},E_{\bullet},\mathcal{B}) ( E q , E ∙ , B ) induces a parabolic structure on each V a , τ V^{\mathbf{a},\tau} V a , τ . Let b r e g ∈ t ( k ) / W \mathbf{b}^{\mathrm{reg}}\in\mathfrak{t}(\mathbf{k})/W b reg ∈ t ( k ) / W be the unique regular tuple satisfying AS ( b r e g ) = a \operatorname{AS}(\mathbf{b}^{\mathrm{reg}})=\mathbf{a} AS ( b reg ) = a and ν ( b r e g ) = 0 \nu(\mathbf{b}^{\mathrm{reg}})=0 ν ( b reg ) = 0 . We define b ∈ t ( k ) / W \mathbf{b}\in\mathfrak{t}(\mathbf{k})/W b ∈ t ( k ) / W to be the unordered n n n -tuple in which b + τ b+\tau b + τ occurs with multiplicity dim k V AS ( b ) , τ \dim_{\mathbf{k}}V^{\operatorname{AS}(b),\tau} dim k V AS ( b ) , τ for each ( b , τ ) ∈ b r e g × τ (b,\tau)\in\mathbf{b}^{\mathrm{reg}}\times\tau ( b , τ ) ∈ b reg × τ . The parabolic structures on all V AS ( b ) , τ V^{\operatorname{AS}(b),\tau} V AS ( b ) , τ determine an assignment b ↦ β \mathbf{b}\mapsto\beta b ↦ β as follows. For each parabolic weight β j ∈ B \beta_{j}\in\mathcal{B} β j ∈ B and each ( b , τ ) ∈ b r e g × τ (b,\tau)\in\mathbf{b}^{\mathrm{reg}}\times\tau ( b , τ ) ∈ b reg × τ , we assign the weight β j \beta_{j} β j to dim k V AS ( b ) , τ \dim_{\mathbf{k}}V^{\operatorname{AS}(b),\tau} dim k V AS ( b ) , τ copies of b + τ b+\tau b + τ . In this way, we obtain an assignment b ↦ β \mathbf{b}\mapsto\beta b ↦ β of the same filtration type m \mathbf{m} m as τ ↦ β \tau\mapsto\beta τ ↦ β .
Definition 5.8. We define X a b ↦ β ⊆ X a τ ↦ β \mathcal{X}_{a}^{b\mapsto\beta}\subseteq\mathcal{X}_{a}^{\tau\mapsto\beta} X a b ↦ β ⊆ X a τ ↦ β to be the substack characterized by the actions of res q ( ψ ) \operatorname{res}_{q}(\psi) res q ( ψ ) and res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) on ( E q , ∙ ) (E_{q},\bullet) ( E q , ∙ ) giving rise to the assignment b ↦ β b\mapsto\beta b ↦ β according to the procedure described above.
Then we have
X a τ ↦ β = ∐ X a b ↦ β , \mathcal{X}_{a}^{\tau\mapsto\beta}=\coprod\mathcal{X}_{a}^{b\mapsto\beta}, X a τ ↦ β = ∐ X a b ↦ β ,
where the disjoint union ranges over assignments b ↦ β b\mapsto\beta b ↦ β of filtration type m \boldsymbol{m} m whose b b b -component satisfies AS ( b ) = a \operatorname{AS}(b)=a AS ( b ) = a and b − a b r e g = τ b_{-a}b^{\mathrm{reg}}=\tau b − a b reg = τ . Here ( b − a b r e g ) ∈ t ( k ) / W (b_{-a}b^{\mathrm{reg}})\in\mathfrak{t}(k)/W ( b − a b reg ) ∈ t ( k ) / W is obtained from b b b by replacing each occurrence of b b b in b \boldsymbol{b} b with b − b ′ b-b' b − b ′ , where b ′ b' b ′ is the unique element of b r e g \boldsymbol{b}^{\mathrm{reg}} b reg satisfying AS ( b ) = AS ( b ′ ) \operatorname{AS}(b)=\operatorname{AS}(b') AS ( b ) = AS ( b ′ ) .
Restricting the isomorphism (11)
H × A P e x t X β , m → ∼ Conn β ( C , m ) \mathcal{H}\times_{\mathcal{A}^{\mathrm{Pext}}}\mathcal{X}^{\beta,\boldsymbol{m}}\xrightarrow{\sim}\operatorname{Conn}^{\beta}(\mathcal{C},\boldsymbol{m}) H × A Pext X β , m ∼ Conn β ( C , m )
to smaller substacks, we obtain the following
Proposition 5.9. There exists an isomorphism
H a v g × A a o X a b ↦ β → ∼ Conn b ↦ β ( C ) . \mathcal{H}_{a}^{\mathrm{vg}}\times_{\mathcal{A}_{a}^{o}}\mathcal{X}_{a}^{b\mapsto\beta}\xrightarrow{\sim}\operatorname{Conn}^{b\mapsto\beta}(\mathcal{C}). H a vg × A a o X a b ↦ β ∼ Conn b ↦ β ( C ) .
Proof. Let P ′ ⊂ P e x t \mathcal{P}'\subset\mathcal{P}^{\mathrm{ext}} P ′ ⊂ P ext be the substack characterized by res q ( ∇ ) = 0 \operatorname{res}_{q}(\nabla)=0 res q ( ∇ ) = 0 . By definition, P ′ \mathcal{P}' P ′ identifies with the substack of X { 0 } , ( n ) , 0 \mathcal{X}^{\{0\},(n),0} X { 0 } , ( n ) , 0 characterized by the spectral sheaf of ψ \psi ψ being invertible. Here 0 0 0 denotes the zero tuple of residue eigenvalues, { 0 } \{0\} { 0 } denotes the trivial parabolic weight, and ( n ) (n) ( n ) denotes the trivial filtration type. By Theorem 5.4 and the BNR correspondence, we have P ′ ≅ P ( S / A ) = P \mathcal{P}'\cong\mathcal{P}(S/A)=\mathcal{P} P ′ ≅ P ( S / A ) = P , and its actions on H \mathcal{H} H and X β , m \mathcal{X}^{\beta,\boldsymbol{m}} X β , m agree with the ones considered in Section 2.3 and Theorem 5.4, respectively. Let b r e g ∈ t ( k ) / W b^{\mathrm{reg}}\in\mathfrak{t}(k)/W b reg ∈ t ( k ) / W be the unique regular tuple satisfying AS ( b r e g ) = a \operatorname{AS}(b^{\mathrm{reg}})=a AS ( b reg ) = a and ν ( b r e g ) = 0 \nu(b^{\mathrm{reg}})=0 ν ( b reg ) = 0 . Let X a β , m : = X β , m × A A a \mathcal{X}_{a}^{\beta,\boldsymbol{m}}:=\mathcal{X}^{\beta,\boldsymbol{m}}\times_{\mathcal{A}}A_{a} X a β , m := X β , m × A A a . Then the isomorphism (11) restricts to an isomorphism
H b r e g v g × A a o X a β , m → ∼ ∐ b ∈ t ( k ) / W AS ( b ) = a Conn b β ( C ) , (16) \mathcal{H}_{b^{\mathrm{reg}}}^{\mathrm{vg}}\times_{\mathcal{A}_{a}^{o}}\mathcal{X}_{a}^{\beta,\boldsymbol{m}}\xrightarrow{\sim}\coprod_{\substack{b\in\mathfrak{t}(k)/W\\ \operatorname{AS}(b)=a}}\operatorname{Conn}_{b}^{\beta}(\mathcal{C}), \tag*{(16)} H b reg vg × A a o X a β , m ∼ b ∈ t ( k ) / W AS ( b ) = a ∐ Conn b β ( C ) , (16)
where Conn b β ( C , m ) ⊂ Conn β ( C , m ) \operatorname{Conn}_{b}^{\beta}(\mathcal{C},\boldsymbol{m})\subset\operatorname{Conn}^{\beta}(\mathcal{C},\boldsymbol{m}) Conn b β ( C , m ) ⊂ Conn β ( C , m ) is the substack characterized by the multiset of eigenvalues of res q ( ∇ ) \operatorname{res}_{q}(\nabla) res q ( ∇ ) being equal to b \boldsymbol{b} b . Recall that H a v g \mathcal{H}_{a}^{\mathrm{vg}} H a vg is identified with H b r e g v g \mathcal{H}_{b^{\mathrm{reg}}}^{\mathrm{vg}} H b reg vg through Theorem 3.4. Note that
X a β , m = ∐ X a b ↦ β , \mathcal{X}_{a}^{\beta,\boldsymbol{m}}=\coprod\mathcal{X}_{a}^{b\mapsto\beta}, X a β , m = ∐ X a b ↦ β ,
where the disjoint union ranges over all b ∈ t ( k ) / W b\in\mathfrak{t}(k)/W b ∈ t ( k ) / W satisfying AS ( b ) = a \operatorname{AS}(b)=a AS ( b ) = a and all assignments b ↦ β b\mapsto\beta b ↦ β of filtration type m \boldsymbol{m} m . We also have
Conn b β ( C , m ) = ∐ Conn b ↦ β ( C ) \operatorname{Conn}_{b}^{\beta}(\mathcal{C},\boldsymbol{m})=\coprod\operatorname{Conn}^{b\mapsto\beta}(\mathcal{C}) Conn b β ( C , m ) = ∐ Conn b ↦ β ( C )
where the disjoint union ranges over all assignments b ↦ β b\mapsto\beta b ↦ β of filtration type m \boldsymbol{m} m . Now we check that the isomorphism (19) matches X a b ↦ β \mathcal{X}_{a}^{b\mapsto\beta} X a b ↦ β with Conn b ↦ β ( C ) \operatorname{Conn}^{b\mapsto\beta}(\mathcal{C}) Conn b ↦ β ( C ) . Let E = ( E , ∇ E ) ∈ H b r e g v g \mathcal{E}=(E,\nabla_{E})\in\mathcal{H}_{b^{\mathrm{reg}}}^{\mathrm{vg}} E = ( E , ∇ E ) ∈ H b reg vg such that h D R ( E ) = a ∈ A a h_{\mathrm{DR}}(\mathcal{E})=a\in A_{a} h DR ( E ) = a ∈ A a , and let L ψ \mathcal{L}_{\psi} L ψ be the spectral sheaf of its p p p -curvature ψ \psi ψ . Let M = ( M , M , ∇ M , V M ) \mathcal{M}=(M,\mathcal{M},\nabla_{\mathcal{M}},\mathcal{V}_{\mathcal{M}}) M = ( M , M , ∇ M , V M ) be X a b ↦ β \mathcal{X}_{a}^{b\mapsto\beta} X a b ↦ β that maps to the same a ∈ A a a\in A_{a} a ∈ A a , and let F ψ \mathcal{F}_{\psi} F ψ be the spectral sheaf of ψ \psi ψ . Recall that the isomorphism (19) maps ( E , M ) (\mathcal{E},\mathcal{M}) ( E , M ) to
( π ∗ F r ( L E ⊗ O S F r F ψ ) , ( π ∗ F r ( L E ⊗ O S F r F ψ ) ) , ∇ ⊗ ) , \left(\pi_{*}^{\mathrm{Fr}}(L_{\mathcal{E}}\otimes_{\mathcal{O}_{S^{\mathrm{Fr}}}}\mathcal{F}_{\psi}),(\pi_{*}^{\mathrm{Fr}}(L_{\mathcal{E}}\otimes_{\mathcal{O}_{S^{\mathrm{Fr}}}}\mathcal{F}_{\psi})),\nabla_{\otimes}\right), ( π ∗ Fr ( L E ⊗ O S Fr F ψ ) , ( π ∗ Fr ( L E ⊗ O S Fr F ψ )) , ∇ ⊗ ) ,
where ∇ ⊗ \nabla_{\otimes} ∇ ⊗ is the tensor product connection, and the parabolic structure ( π ∗ F r ( L E ⊗ O S F r F ψ ) ) ∙ (\pi_{*}^{\mathrm{Fr}}(L_{\mathcal{E}}\otimes_{\mathcal{O}_{S^{\mathrm{Fr}}}}\mathcal{F}_{\psi}))_{\bullet} ( π ∗ Fr ( L E ⊗ O S Fr F ψ ) ) ∙ is induced from the parabolic structure on π ∗ F r F ψ \pi_{*}^{\mathrm{Fr}}\mathcal{F}_{\psi} π ∗ Fr F ψ . Recall that the spectral curve S a F r S_{a}^{\mathrm{Fr}} S a Fr decomposes as
S a F r × C D q = ∐ j = 1 s S a , a j F r . S_{a}^{\mathrm{Fr}}\times_{C}D_{q}=\coprod_{j=1}^{s}S_{a,a_{j}}^{\mathrm{Fr}}. S a Fr × C D q = j = 1 ∐ s S a , a j Fr .
near q q q , and the spectral sheaf F ψ \mathcal{F}_{\psi} F ψ decomposes accordingly, with summands corresponding to different eigenvalues of res q ( ψ ) \operatorname{res}_{q}(\psi) res q ( ψ ) . Let b j ∈ b r e g b_j \in\mathbf{b}^{\mathrm{reg}} b j ∈ b reg be the unique element such that AS ( b j ) = a j \operatorname{AS}(b_j) = a_j AS ( b j ) = a j . By definition, res q ( ∇ E ) \operatorname{res}_{q}(\nabla_{E}) res q ( ∇ E ) has eigenvalues equal to b j b_j b j when restricted to π ∗ F r ( L ℓ ∣ S a j F r ) \pi_{*}^{\mathrm{Fr}}(\mathcal{L}_{\ell} |_{\mathcal{S}_{a_j}^{\mathrm{Fr}}}) π ∗ Fr ( L ℓ ∣ S a j Fr ) . Tensoring with L ℓ \mathcal{L}_{\ell} L ℓ over S a j F r \mathcal{S}_{a_j}^{\mathrm{Fr}} S a j Fr adds b j b_j b j to the eigenvalues of res q ( ∇ M ) \operatorname{res}_{q}(\nabla_{M}) res q ( ∇ M ) , while transporting the parabolic structure on π ∗ F r F ψ \pi_{*}^{\mathrm{Fr}}\mathcal{F}_{\psi} π ∗ Fr F ψ accordingly. This modification of eigenvalues and parabolic structure agrees exactly with the modification from τ ↦ β \tau\mapsto\beta τ ↦ β to b ↦ β b \mapsto\beta b ↦ β used in Theorem 5.8 of X a b ↦ β \mathcal{X}_{a}^{b \mapsto\beta} X a b ↦ β .
Note that when restricted to A α ⊂ A A_{\alpha} \subset A A α ⊂ A , the right-hand side of the isomorphism in Theorem 5.4 decomposes as
H i g g s α ( C ′ , m ′ ) × A A α = ∐ a → α H i g g s a → α ( C ′ ) , \mathop{\mathrm{Higgs}}\nolimits^{\alpha}(\mathcal{C}',m') \times_{A} A_{\alpha} = \coprod_{a \to\alpha} \mathop{\mathrm{Higgs}}\nolimits^{a \to\alpha}(\mathcal{C}'), Higgs α ( C ′ , m ′ ) × A A α = a → α ∐ Higgs a → α ( C ′ ) ,
where the disjoint union ranges over all assignments a ↦ α a \mapsto\alpha a ↦ α of filtration type m ′ m' m ′ . We have
Proposition 5.10. Let a ↦ α a \mapsto\alpha a ↦ α and b ↦ β b \mapsto\beta b ↦ β be two assignments related through (10). The isomorphism in Theorem 5.4 restricts to an isomorphism
(20)
X a b ↦ β ≅ H i g g s a → α ( C ′ ) . \mathcal{X}_{a}^{b \mapsto\beta} \cong\mathop{\mathrm{Higgs}}\nolimits^{a \to\alpha}(\mathcal{C}'). X a b ↦ β ≅ Higgs a → α ( C ′ ) .
Proof. Let E 0 = ( E 0 , E 0 , ∇ 0 ) ∈ C o n n 0 τ ↦ β ( C ) ( k ) \mathcal{E}^{0} = (E^{0}, \mathbf{E}^{0}, \nabla^{0}) \in\mathop{\mathrm{Conn}}\nolimits_{0}^{\tau\mapsto\beta}(\mathcal{C})(k) E 0 = ( E 0 , E 0 , ∇ 0 ) ∈ Conn 0 τ ↦ β ( C ) ( k ) be the model object defined in Section 5.2.1. Let D q = Spec k [ [ x ] ] D_{q} = \operatorname{Spec} k[[x]] D q = Spec k [[ x ]] be the formal disc around q ∈ C q \in C q ∈ C . The same computation as in the proof of Theorem 5.6 implies that
ϕ ∈ E n d ( E 0 ) ( D q ) ⊂ Lie ( GL n ) ( k ( ( x ) ) ) ⟺ ϕ ∈ Lie ( G α ~ ) ( k [ [ x p ] ] ) ⟹ ψ : = x − ν ( τ ~ ) ϕ x ν ( τ ~ ) ∈ Lie ( G β ~ ) ( k [ [ x ] ] ) , \phi\in\mathop{\mathrm{End}}\nolimits(\mathcal{E}^{0})(D_{q}) \subset\operatorname{Lie}(\operatorname{GL}_{n})(k((x))) \Longleftrightarrow\phi\in\operatorname{Lie}(\mathcal{G}_{\widetilde{\alpha}})(k[[x^{p} ]]) \Longrightarrow\psi:= x^{-\nu(\widetilde{\tau})}\phi x^{\nu(\widetilde{\tau})} \in\operatorname{Lie}(\mathcal{G}_{\widetilde{\beta}})(k[[x]]), ϕ ∈ End ( E 0 ) ( D q ) ⊂ Lie ( GL n ) ( k (( x ))) ⟺ ϕ ∈ Lie ( G α ) ( k [[ x p ]]) ⟹ ψ := x − ν ( τ ) ϕ x ν ( τ ) ∈ Lie ( G β ) ( k [[ x ]]) ,
where α ~ = ( α o p α ~ ( 1 ) , α o p α ~ ( 2 ) , … , α o p α ~ ( n ) ) \widetilde{\alpha} = (\alpha_{\mathrm{op}}^{\widetilde{\alpha}}(1),\alpha_{\mathrm{op}}^{\widetilde{\alpha}}(2),\ldots,\alpha_{\mathrm{op}}^{\widetilde{\alpha}}(n)) α = ( α op α ( 1 ) , α op α ( 2 ) , … , α op α ( n )) as defined through (14), and β ~ = ( β o p τ ~ ( 1 ) , β o p τ ~ ( 2 ) , … , β o p τ ~ ( n ) ) \widetilde{\beta} = (\beta_{\mathrm{op}}^{\widetilde{\tau}}(1),\beta_{\mathrm{op}}^{\widetilde{\tau}}(2),\ldots,\beta_{\mathrm{op}}^{\widetilde{\tau}}(n)) β = ( β op τ ( 1 ) , β op τ ( 2 ) , … , β op τ ( n )) . Let
ℓ α ~ : Lie ( G α ~ ) ( k [ [ x p ] ] ) ⟶ Lie ( L α ) ( k ) \ell_{\widetilde{\alpha}} : \operatorname{Lie}(\mathcal{G}_{\widetilde{\alpha}})(k[[x^{p}]]) \longrightarrow\operatorname{Lie}(\mathcal{L}_{\alpha})(k) ℓ α : Lie ( G α ) ( k [[ x p ]]) ⟶ Lie ( L α ) ( k )
be the morphism on Lie algebras induced by the quotient map from a parahoric group scheme to the corresponding Levi factor. Let ϕ ∈ E n d ( E 0 ) ( D q ) \phi\in\mathop{\mathrm{End}}\nolimits(\mathcal{E}^{0})(D_{q}) ϕ ∈ End ( E 0 ) ( D q ) such that the multiset of eigenvalues of ϕ q \phi_{q} ϕ q acting on E q 0 E_{q}^{0} E q 0 is given by a a a . To obtain the desired isomorphism, it is enough to show that if the actions of ϕ q \phi_{q} ϕ q and res q ( ∇ 0 ) \operatorname{res}_{q}(\nabla^{0}) res q ( ∇ 0 ) on ( E q 0 , E q 0 ) (E_{q}^{0},\mathbf{E}_{q}^{0}) ( E q 0 , E q 0 ) give rise to b ↦ β b \mapsto\beta b ↦ β through the procedure used in Definition 5.8, then the block diagonal matrix ℓ α ~ ( ϕ ) \ell_{\widetilde{\alpha}}(\phi) ℓ α ( ϕ ) has eigenvalues prescribed by a ↦ α a \mapsto\alpha a ↦ α . Note that E 0 E^{0} E 0 is trivialized through the local section x ν ( τ ~ ) x^{\nu(\widetilde{\tau})} x ν ( τ ) , and the action of ϕ q \phi_{q} ϕ q becomes ψ ( 0 ) \psi(0) ψ ( 0 ) . A generalized joint eigenvector of ( ψ ( 0 ) , res q ( ∇ 0 ) ) (\psi(0),\operatorname{res}_{q}(\nabla^{0})) ( ψ ( 0 ) , res q ( ∇ 0 )) with eigenvalues ( a , τ ) ∈ α × τ (a,\tau) \in\alpha\times\tau ( a , τ ) ∈ α × τ and parabolic weight α = ( β − ν ( τ ) ) / p \alpha= (\beta- \nu(\tau))/p α = ( β − ν ( τ )) / p . If b = b ′ + τ b = b' + \tau b = b ′ + τ with b ′ ∈ b r e g b' \in\mathbf{b}^{\mathrm{reg}} b ′ ∈ b reg and τ ∈ F p \tau\in\mathbb{F}_{p} τ ∈ F p , then ν ( b ) = ν ( τ ) \nu(b) = \nu(\tau) ν ( b ) = ν ( τ ) . The desired statement follows from simple unwinding of notations.
Theorem 5.3 follows immediately by combining Theorem 5.9 and Theorem 5.10.
Proof of the main theorem
We write Conn 0 τ ↦ β , s s ( C ) : = Conn 0 τ ↦ β ( C ) ∩ Conn β - s s ( C , m ) \operatorname{Conn}_{0}^{\tau\mapsto\beta,\mathrm{ss}}(\mathcal{C}) := \operatorname{Conn}_{0}^{\tau\mapsto\beta}(\mathcal{C}) \cap\operatorname{Conn}^{\beta\text{-}\mathrm{ss}}(\mathcal{C},m) Conn 0 τ ↦ β , ss ( C ) := Conn 0 τ ↦ β ( C ) ∩ Conn β - ss ( C , m ) . We start by showing
Proposition 5.11. The isomorphism
(21)
Conn 0 τ ↦ β ( C ) ≅ Bun α ( C ′ , m ′ ) \operatorname{Conn}_{0}^{\tau\mapsto\beta}(\mathcal{C}) \cong\operatorname{Bun}^{\alpha}(\mathcal{C}',m') Conn 0 τ ↦ β ( C ) ≅ Bun α ( C ′ , m ′ )
in Proposition 5.7 restricts to an isomorphism
Conn 0 τ ↦ β , s s ( C ) ≅ Bun α - s s ( C ′ , m ′ ) . (17) \operatorname{Conn}_{0}^{\tau\mapsto\beta,\mathrm{ss}}(\mathcal{C}) \cong\operatorname{Bun}^{\alpha\text{-}\mathrm{ss}}(\mathcal{C}',m'). \tag*{(17)} Conn 0 τ ↦ β , ss ( C ) ≅ Bun α - ss ( C ′ , m ′ ) . (17)
Proof. Let E = ( E , E ∙ , ∇ ) ∈ Conn 0 τ ↦ β ( C ) \mathcal{E}=(E,\mathbf{E}_{\bullet},\nabla)\in\operatorname{Conn}_{0}^{\tau\mapsto\beta}(C) E = ( E , E ∙ , ∇ ) ∈ Conn 0 τ ↦ β ( C ) and let F = ( F , F ∙ ) ∈ Bun ( C ′ , m ′ ) \mathcal{F}=(F,\mathbf{F}_{\bullet})\in\operatorname{Bun}(\mathcal{C}',\mathbf{m}') F = ( F , F ∙ ) ∈ Bun ( C ′ , m ′ ) be its image under the isomorphism (21). Note that for any vector bundle E E E over C C C , any vector subbundle E ′ ⊆ E E'\subseteq E E ′ ⊆ E is uniquely determined by its restriction E ′ ∣ U E'|_{U} E ′ ∣ U to U = C \ q U=C\backslash q U = C \ q . Indeed, writing j : U ↪ C j:U\hookrightarrow C j : U ↪ C for the inclusion, we have
E ′ = E ∩ j ∗ ( E ′ ∣ U ) E'=E\cap j_{*}(E'|_{U}) E ′ = E ∩ j ∗ ( E ′ ∣ U )
inside j ∗ ( E ∣ U ) j_{*}(E|_{U}) j ∗ ( E ∣ U ) . Over U U U and U ′ U' U ′ , the correspondence (21) is given by Cartier descent, i.e.
F ∣ U ′ ≅ ( Fr C ) ∗ ker ( ∇ ∣ U : E ∣ U → E ∣ U ⊗ ω U ) , ( E ∣ U , ∇ ∣ U ) ≅ ( Fr C ∗ ( F ∣ U ′ ) , ∇ c a n ) . F|_{U'}\cong(\operatorname{Fr}_{C})_{*}\ker(\nabla|_{U}:E|_{U}\to E|_{U}\otimes\omega_{U}),\qquad(E|_{U},\nabla|_{U})\cong(\operatorname{Fr}_{C}^{*}(F|_{U'}),\nabla_{\mathrm{can}}). F ∣ U ′ ≅ ( Fr C ) ∗ ker ( ∇ ∣ U : E ∣ U → E ∣ U ⊗ ω U ) , ( E ∣ U , ∇ ∣ U ) ≅ ( Fr C ∗ ( F ∣ U ′ ) , ∇ can ) .
Cartier descent also induces a correspondence between ∇ ∣ U \nabla|_{U} ∇ ∣ U -invariant subbundles of E ∣ U E|_{U} E ∣ U and subbundles of F ∣ U ′ F|_{U'} F ∣ U ′ . Let E U ′ ⊆ E ∣ U E'_{U}\subseteq E|_{U} E U ′ ⊆ E ∣ U be the ∇ ∣ U \nabla|_{U} ∇ ∣ U -invariant subbundle corresponding to F U ′ ′ ⊆ F ∣ U ′ F'_{U'}\subseteq F|_{U'} F U ′ ′ ⊆ F ∣ U ′ . Let E ′ = E ∩ j ∗ ( E U ′ ) E'=E\cap j_{*}(E'_{U}) E ′ = E ∩ j ∗ ( E U ′ ) and F ′ = F ∩ j ∗ ′ ( F U ′ ′ ) F'=F\cap j'_{*}(F'_{U'}) F ′ = F ∩ j ∗ ′ ( F U ′ ′ ) . These are the unique saturated extensions of E U ′ E'_{U} E U ′ and F U ′ ′ F'_{U'} F U ′ ′ , respectively. Let ( E ′ , E ∙ ′ , β ′ ) (E',\mathbf{E}'_{\bullet},\beta') ( E ′ , E ∙ ′ , β ′ ) (resp. ( F ′ , F ∙ ′ , α ′ ) (F',\mathbf{F}'_{\bullet},\alpha') ( F ′ , F ∙ ′ , α ′ ) ) be the parabolic subbundle of ( E , E ∙ , β ) (E,\mathbf{E}_{\bullet},\beta) ( E , E ∙ , β ) (resp. ( F , F ∙ , α ) (F,\mathbf{F}_{\bullet},\alpha) ( F , F ∙ , α ) ) with the induced parabolic structure. For the desired matching of semistable substacks through (21), it is enough to show that for any choice of E U ′ E'_{U} E U ′ and F U ′ ′ F'_{U'} F U ′ ′ , we have
pardeg ( E ′ , E ∙ ′ , β ′ ) = p pardeg ( F ′ , F ∙ ′ , α ′ ) . (18) \operatorname{pardeg}(E',\mathbf{E}'_{\bullet},\beta')=p\operatorname{pardeg}(F',\mathbf{F}'_{\bullet},\alpha'). \tag*{(18)} pardeg ( E ′ , E ∙ ′ , β ′ ) = p pardeg ( F ′ , F ∙ ′ , α ′ ) . (18)
Recall that the degree of a vector bundle E E E on C C C is equal to the degree of its determinant line bundle L = det ( E ) L=\det(E) L = det ( E ) , which can be computed as the sum of the orders of vanishing of any nonzero rational section of L L L at all points of C C C , with poles counted negatively. Since E U ′ ≅ Fr C ∗ F U ′ ′ E'_{U}\cong\operatorname{Fr}_{C}^{*}F'_{U'} E U ′ ≅ Fr C ∗ F U ′ ′ , the contributions over U U U and U ′ U' U ′ to the respective parabolic degrees satisfy the multiplication-by-p p p formula (22).
Now we compute the contributions at q q q and q ′ q' q ′ , which can be done by working with the model object E ∘ = ( E ∘ , E ∙ ∘ , ∇ ∘ , β ) \mathcal{E}^{\circ}=(E^{\circ},\mathbf{E}^{\circ}_{\bullet},\nabla^{\circ},\beta) E ∘ = ( E ∘ , E ∙ ∘ , ∇ ∘ , β ) defined in Section 5.2.1, as well as the model object F ∘ = ( F ∘ , F ∙ ∘ , α ) \mathcal{F}^{\circ}=(F^{\circ},\mathbf{F}^{\circ}_{\bullet},\alpha) F ∘ = ( F ∘ , F ∙ ∘ , α ) associated to α ~ \widetilde{\alpha} α as in Theorem 4.1. Let D = Spec ( k [ [ x ] ] ) D=\operatorname{Spec}(k[[x]]) D = Spec ( k [[ x ]]) be the formal disc around q ∈ C q\in C q ∈ C . Let e 1 , e 2 , … , e n e_{1},e_{2},\ldots,e_{n} e 1 , e 2 , … , e n be the standard basis of F ∘ ∣ D ′ ≅ ⨁ i = 1 n O D ′ F^{\circ}|_{D'}\cong\bigoplus_{i=1}^{n}\mathcal{O}_{D'} F ∘ ∣ D ′ ≅ ⨁ i = 1 n O D ′ . We equip { 1 , 2 , … , n } \{1,2,\ldots,n\} { 1 , 2 , … , n } with a total order by setting
i ≺ j if α ~ o p α ( i ) < α ~ o p α ( j ) , i\prec j\quad\text{if}\quad\widetilde{\alpha}_{\mathrm{op}}^{\alpha}(i)<\widetilde{\alpha}_{\mathrm{op}}^{\alpha}(j), i ≺ j if α op α ( i ) < α op α ( j ) ,
and when α ~ o p α ( i ) = α ~ o p α ( j ) \widetilde{\alpha}_{\mathrm{op}}^{\alpha}(i)=\widetilde{\alpha}_{\mathrm{op}}^{\alpha}(j) α op α ( i ) = α op α ( j ) , we break ties using the usual ordering of the indices. By choosing an arbitrary basis of F q ′ ′ ⊆ k n F'_{q'}\subseteq k^{n} F q ′ ′ ⊆ k n and applying Gaussian elimination, we obtain a basis v 1 , v 2 , … , v n ′ v_{1},v_{2},\ldots,v_{n'} v 1 , v 2 , … , v n ′ of the form
v j = e i j + ∑ i j < i c j i e i , i 1 ≺ i 2 ≺ ⋯ ≺ i n ′ . v_{j}=e_{i_{j}}+\sum_{i_{j}<i}c_{ji}e_{i},\qquad i_{1}\prec i_{2}\prec\cdots\prec i_{n'}. v j = e i j + i j < i ∑ c j i e i , i 1 ≺ i 2 ≺ ⋯ ≺ i n ′ .
Consequently, the parabolic weights (counted with multiplicity) of F ′ F' F ′ at q ′ q' q ′ are α ~ o p α ( i 1 ) , α ~ o p α ( i 2 ) , … , α ~ o p α ( i n ′ ) \widetilde{\alpha}_{\mathrm{op}}^{\alpha}(i_{1}),\widetilde{\alpha}_{\mathrm{op}}^{\alpha}(i_{2}),\ldots,\widetilde{\alpha}_{\mathrm{op}}^{\alpha}(i_{n'}) α op α ( i 1 ) , α op α ( i 2 ) , … , α op α ( i n ′ ) , and their contribution to the parabolic degree is ∑ j = 1 n ′ α ~ o p α ( i j ) \sum_{j=1}^{n'}\widetilde{\alpha}_{\mathrm{op}}^{\alpha}(i_{j}) ∑ j = 1 n ′ α op α ( i j ) . Each v j v_{j} v j lifts to a section v ~ j \widetilde{v}_{j} v j of F ′ ∣ D ′ F'|_{D'} F ′ ∣ D ′ , which is of the form
v ~ j = e i j + ∑ i j < i c j i e i + x p ∑ i a j i ( x p ) e i . \widetilde{v}_{j}=e_{i_{j}}+\sum_{i_{j}<i}c_{ji}e_{i}+x^{p}\sum_{i}a_{ji}(x^{p})e_{i}. v j = e i j + i j < i ∑ c j i e i + x p i ∑ a j i ( x p ) e i .
v ~ 1 , v ~ 2 , … , v ~ n ′ \widetilde{v}_{1},\widetilde{v}_{2},\ldots,\widetilde{v}_{n'} v 1 , v 2 , … , v n ′ form a basis of F ′ ∣ D ′ F'|_{D'} F ′ ∣ D ′ .
By abuse of notation, we also denote by e 1 , e 2 , … , e n e_{1},e_{2},\ldots,e_{n} e 1 , e 2 , … , e n the standard basis of
Fr C ∗ ( F ∘ ∣ D ′ ) = ⨁ i = 1 n O D ⊆ E ∘ ∣ D = ⨁ i = 1 n O D ( − ν ( τ i ) q ) . \operatorname{Fr}_{C}^{*}(F^{\circ}|_{D'})=\bigoplus_{i=1}^{n}\mathcal{O}_{D}\subseteq E^{\circ}|_{D}=\bigoplus_{i=1}^{n}\mathcal{O}_{D}(-\nu(\tau_{i})q). Fr C ∗ ( F ∘ ∣ D ′ ) = i = 1 ⨁ n O D ⊆ E ∘ ∣ D = i = 1 ⨁ n O D ( − ν ( τ i ) q ) .
Set f i = x ν ( τ i ) e i f_{i}=x^{\nu(\tau_{i})}e_{i} f i = x ν ( τ i ) e i , then E ∘ ∣ D E^{\circ}|_{D} E ∘ ∣ D is trivialized as E ∘ ∣ D = ⨁ i = 1 n O D f i E^{\circ}|_{D}=\bigoplus_{i=1}^{n}\mathcal{O}_{D}f_{i} E ∘ ∣ D = ⨁ i = 1 n O D f i . Now set u ~ j : = x ν ( τ i j ) v ~ j \widetilde{u}_{j}:=x^{\nu(\tau_{i_{j}})}\widetilde{v}_{j} u j := x ν ( τ i j ) v j . Then
u ~ j = f i j + ∑ i j < i c j i x ν ( τ i j ) − ν ( τ i ) f i + ∑ i x p + ν ( τ i j ) − ν ( τ i ) a j i ( x p ) f i . \widetilde{u}_{j}=f_{i_{j}}+\sum_{i_{j}<i}c_{ji}x^{\nu(\tau_{i_{j}})-\nu(\tau_{i})}f_{i}+\sum_{i}x^{p+\nu(\tau_{i_{j}})-\nu(\tau_{i})}a_{ji}(x^{p})f_{i}. u j = f i j + i j < i ∑ c j i x ν ( τ i j ) − ν ( τ i ) f i + i ∑ x p + ν ( τ i j ) − ν ( τ i ) a j i ( x p ) f i .
Note that
i < j ⇒ α ~ o p a ~ ( i ) ≤ α ~ o p a ~ ( j ) ⇒ { ν ( τ i ) > ν ( τ j ) , or ν ( τ i ) = ν ( τ j ) and β ~ o p τ ~ ( i ) ≤ β ~ o p τ ~ ( j ) . (19) i<j \Rightarrow\tilde{\alpha}_{\mathrm{op}}^{\tilde{a}}(i) \leq\tilde{\alpha}_{\mathrm{op}}^{\tilde{a}}(j) \Rightarrow\left\{\begin{array}{ll}
\nu(\tau_i)>\nu(\tau_j), & \text{or}\\
\nu(\tau_i)=\nu(\tau_j)\text{ and }\tilde{\beta}_{\mathrm{op}}^{\tilde{\tau}}(i)\leq\tilde{\beta}_{\mathrm{op}}^{\tilde{\tau}}(j).
\end{array}\right. \tag*{(19)} i < j ⇒ α ~ op a ~ ( i ) ≤ α ~ op a ~ ( j ) ⇒ { ν ( τ i ) > ν ( τ j ) , ν ( τ i ) = ν ( τ j ) and β ~ op τ ~ ( i ) ≤ β ~ op τ ~ ( j ) . or (19)
Also note that since the image of ν \nu ν lies in { − p + 1 , … , − 1 , 0 } \{-p+1,\ldots,-1,0\} { − p + 1 , … , − 1 , 0 } , we have p + ν ( τ j ) − ν ( τ i ) > 0 p+\nu(\tau_j)-\nu(\tau_i)>0 p + ν ( τ j ) − ν ( τ i ) > 0 for any indices i , j i,j i , j . Therefore u ~ j \tilde{u}_j u ~ j is a section of E 0 ∣ D E^0|_D E 0 ∣ D for each 1 ≤ j ≤ n ′ 1\leq j\leq n' 1 ≤ j ≤ n ′ . Furthermore, u ~ 1 ( 0 ) , u ~ 2 ( 0 ) , … , u ~ n ′ ( 0 ) \tilde{u}_1(0),\tilde{u}_2(0),\ldots,\tilde{u}_{n'}(0) u ~ 1 ( 0 ) , u ~ 2 ( 0 ) , … , u ~ n ′ ( 0 ) are linearly independent, and u ~ 1 ∣ D × , u ~ 2 ∣ D × , … , u ~ n ′ ∣ D × \tilde{u}_1|_{D\times},\tilde{u}_2|_{D\times},\ldots,\tilde{u}_{n'}|_{D\times} u ~ 1 ∣ D × , u ~ 2 ∣ D × , … , u ~ n ′ ∣ D × form a basis of E ′ ∣ D × E'|_{D\times} E ′ ∣ D × . By uniqueness of saturated extension, u ~ 1 , u ~ 2 , … , u ~ n ′ \tilde{u}_1,\tilde{u}_2,\ldots,\tilde{u}_{n'} u ~ 1 , u ~ 2 , … , u ~ n ′ form a basis of E ′ ∣ D E'|_D E ′ ∣ D . It follows that the contribution to deg ( E ′ ) \deg(E') deg ( E ′ ) at q q q is ∑ j = 1 n ′ ( − ν ( τ i j ) ) \sum_{j=1}^{n'}(-\nu(\tau_{i_j})) ∑ j = 1 n ′ ( − ν ( τ i j )) . Note that
u ~ j ( 0 ) = f i j + ∑ i j < i ν ( τ i ) = ν ( τ i j ) c j i f i for each 1 ≤ j ≤ n ′ . \tilde{u}_j(0)=f_{i_j}+\sum_{\substack{i_j<i\\ \nu(\tau_i)=\nu(\tau_{i_j})}}c_{ji}f_i\qquad\text{for each }1\leq j\leq n'. u ~ j ( 0 ) = f i j + i j < i ν ( τ i ) = ν ( τ i j ) ∑ c j i f i for each 1 ≤ j ≤ n ′ .
By (23), the parabolic weights (counted with multiplicity) of E ′ E' E ′ at q q q are β ~ o p τ ~ ( i 1 ) , β ~ o p τ ~ ( i 2 ) , … , β ~ o p τ ~ ( i n ′ ) \tilde{\beta}_{\mathrm{op}}^{\tilde{\tau}}(i_1),\tilde{\beta}_{\mathrm{op}}^{\tilde{\tau}}(i_2),\ldots,\tilde{\beta}_{\mathrm{op}}^{\tilde{\tau}}(i_{n'}) β ~ op τ ~ ( i 1 ) , β ~ op τ ~ ( i 2 ) , … , β ~ op τ ~ ( i n ′ ) . The desired equality
∑ j = 1 n ′ β ~ o p τ ~ ( i j ) + ∑ j = 1 n ′ ( − ν ( τ i j ) ) = p ∑ j = 1 n ′ α ~ o p a ~ ( i j ) \sum_{j=1}^{n'}\tilde{\beta}_{\mathrm{op}}^{\tilde{\tau}}(i_j)+\sum_{j=1}^{n'}(-\nu(\tau_{i_j}))=p\sum_{j=1}^{n'}\tilde{\alpha}_{\mathrm{op}}^{\tilde{a}}(i_j) j = 1 ∑ n ′ β ~ op τ ~ ( i j ) + j = 1 ∑ n ′ ( − ν ( τ i j )) = p j = 1 ∑ n ′ α ~ op a ~ ( i j )
follows directly from the definition (14).
Proof of Theorem 5.2. Let X α b , s s ⊂ X α b X_{\alpha}^{\mathbf{b},\mathrm{ss}}\subset X_{\alpha}^{\mathbf{b}} X α b , ss ⊂ X α b be the substack characterized by the semistability inequality for any subbundle preserved by both the connection ∇ \nabla ∇ and the twisted Higgs field ψ \psi ψ . We start by showing that the isomorphism
H a v g × A a P a ∘ X α b ≃ Conn b → β ( C ) (20) \mathcal{H}_{a}^{\mathrm{vg}}\times_{A_a}^{P_a^\circ}X_{\alpha}^{\mathbf{b}} \simeq\operatorname{Conn}^{\mathbf{b}\to\beta}(\mathcal{C}) \tag*{(20)} H a vg × A a P a ∘ X α b ≃ Conn b → β ( C ) (20)
in Theorem 5.9 restricts to an isomorphism
H a v g × A a P a ∘ X α b , s s ≃ Conn b → β , s s ( C ) . (21) \mathcal{H}_{a}^{\mathrm{vg}}\times_{A_a}^{P_a^\circ}X_{\alpha}^{\mathbf{b},\mathrm{ss}}\simeq\operatorname{Conn}^{\mathbf{b}\to\beta,\mathrm{ss}}(\mathcal{C}). \tag*{(21)} H a vg × A a P a ∘ X α b , ss ≃ Conn b → β , ss ( C ) . (21)
Let b r e g ∈ t ( k ) / W \mathbf{b}^{\mathrm{reg}}\in t(k)/W b reg ∈ t ( k ) / W be the unique regular tuple satisfying AS ( b r e g ) = a \operatorname{AS}(\mathbf{b}^{\mathrm{reg}})=\mathbf{a} AS ( b reg ) = a and ν ( b r e g ) = 0 \nu(\mathbf{b}^{\mathrm{reg}})=0 ν ( b reg ) = 0 . We identify H a v g \mathcal{H}_{a}^{\mathrm{vg}} H a vg with H b r e g v g \mathcal{H}_{\mathbf{b}^{\mathrm{reg}}}^{\mathrm{vg}} H b reg vg through Theorem 3.4. Let E = ( E , ∇ ) ∈ H b r e g v g \mathcal{E}=(E,\nabla)\in\mathcal{H}_{\mathbf{b}^{\mathrm{reg}}}^{\mathrm{vg}} E = ( E , ∇ ) ∈ H b reg vg such that h D R ( E ) = a = A a h_{\mathrm{DR}}(\mathcal{E})=\mathbf{a}=A_a h DR ( E ) = a = A a , and let L E \mathcal{L}_{\mathcal{E}} L E be the spectral sheaf of its p p p -curvature ψ E \psi_{\mathcal{E}} ψ E . Let M = ( M , ∇ M , ψ ) ∈ X α b \mathcal{M}=(M,\nabla_M,\psi)\in X_{\alpha}^{\mathbf{b}} M = ( M , ∇ M , ψ ) ∈ X α b that maps to the same a ∈ A a \mathbf{a}\in A_a a ∈ A a , and let F ψ \mathcal{F}_{\psi} F ψ be the spectral sheaf of ψ \psi ψ . Recall that the isomorphism (24) maps ( E , M ) (\mathcal{E},\mathcal{M}) ( E , M ) to
N = ( N , N ∙ , ∇ N ) : = ( π ∗ F r ( L E ⊗ O S a F r F ψ ) , ( π ∗ F r ( L E ⊗ O S a F r F ψ ) ∙ , ∇ ⊗ ) , \mathcal{N}=(N,\mathbf{N}_{\bullet},\nabla_N):=(\pi_*^{\mathrm{Fr}}(\mathcal{L}_{\mathcal{E}}\otimes_{\mathcal{O}_{S_a^{\mathrm{Fr}}}}\mathcal{F}_{\psi}),(\pi_*^{\mathrm{Fr}}(\mathcal{L}_{\mathcal{E}}\otimes_{\mathcal{O}_{S_a^{\mathrm{Fr}}}}\mathcal{F}_{\psi})_{\bullet},\nabla_{\otimes}), N = ( N , N ∙ , ∇ N ) := ( π ∗ Fr ( L E ⊗ O S a Fr F ψ ) , ( π ∗ Fr ( L E ⊗ O S a Fr F ψ ) ∙ , ∇ ⊗ ) ,
see (12). Let M ′ ⊆ M M'\subseteq M M ′ ⊆ M be a subbundle preserved by ∇ M \nabla_M ∇ M and ψ \psi ψ . Let Y a F r ⊆ S a F r Y_a^{\mathrm{Fr}}\subseteq S_a^{\mathrm{Fr}} Y a Fr ⊆ S a Fr be the spectral subcurve defined as the vanishing scheme of the characteristic polynomial of ψ ∣ M ′ \psi|_{M'} ψ ∣ M ′ , and let F ψ ′ \mathcal{F}_{\psi}' F ψ ′ be the spectral sheaf on Y a F r Y_a^{\mathrm{Fr}} Y a Fr corresponding to M ′ M' M ′ . Set N ′ : = π ∗ F r ( L E ∣ Y a F r ⊗ Y a F r F ψ ′ ) N':=\pi_*^{\mathrm{Fr}}(\mathcal{L}_{\mathcal{E}}|_{Y_a^{\mathrm{Fr}}}\otimes_{Y_a^{\mathrm{Fr}}}\mathcal{F}_{\psi}') N ′ := π ∗ Fr ( L E ∣ Y a Fr ⊗ Y a Fr F ψ ′ ) . Then N ′ ⊆ N N'\subseteq N N ′ ⊆ N is a subbundle preserved by ∇ N \nabla_N ∇ N , and every subbundle of N N N preserved by ∇ N \nabla_N ∇ N arises this way from some subbundle M ′ ⊆ M M'\subseteq M M ′ ⊆ M preserved by ∇ M \nabla_M ∇ M and ψ \psi ψ . Note that if S a F r S_a^{\mathrm{Fr}} S a Fr decomposes into integral components as
S a F r = ∑ t = 1 N n t S a , t F r , S_a^{\mathrm{Fr}}=\sum_{t=1}^{N}n_tS_{a,t}^{\mathrm{Fr}}, S a Fr = t = 1 ∑ N n t S a , t Fr ,
then
Y a F r = ∑ t = 1 N n t ′ S a , t F r , 0 ≤ n t ′ ≤ n t . Y_a^{\mathrm{Fr}}=\sum_{t=1}^{N}n'_tS_{a,t}^{\mathrm{Fr}},\qquad0\leq n'_t\leq n_t. Y a Fr = t = 1 ∑ N n t ′ S a , t Fr , 0 ≤ n t ′ ≤ n t .
Since ν ( b r e g ) = 0 \nu(b^{\mathrm{reg}})=0 ν ( b reg ) = 0 , Theorem 3.1 implies that L E \mathcal{L}_{\mathcal{E}} L E has degree zero on each component S a , t F r S^{\mathrm{Fr}}_{a,t} S a , t Fr . Hence, by [dCGZ24, Proof of Theorem 4.7, (22)], we have
χ ( F ψ ′ ) = χ ( L E ∣ Y a F r ⊗ Y a F r F ψ ′ ) , \chi(\mathcal{F}'_{\psi})=\chi\left(\mathcal{L}_{\mathcal{E}}\big|_{Y^{\mathrm{Fr}}_{a}}\otimes_{Y^{\mathrm{Fr}}_{a}}\mathcal{F}'_{\psi}\right), χ ( F ψ ′ ) = χ ( L E Y a Fr ⊗ Y a Fr F ψ ′ ) ,
thus
χ ( π ∗ F r F ψ ′ ) = χ ( π ∗ F r ( L E ∣ Y a F r ⊗ Y a F r F ψ ′ ) ) , \chi\left(\pi^{\mathrm{Fr}}_{*}\mathcal{F}'_{\psi}\right)=\chi\left(\pi^{\mathrm{Fr}}_{*}\left(\mathcal{L}_{\mathcal{E}}\big|_{Y^{\mathrm{Fr}}_{a}}\otimes_{Y^{\mathrm{Fr}}_{a}}\mathcal{F}'_{\psi}\right)\right), χ ( π ∗ Fr F ψ ′ ) = χ ( π ∗ Fr ( L E Y a Fr ⊗ Y a Fr F ψ ′ ) ) ,
which implies
deg ( M ′ ) = deg ( N ′ ) . \deg(M')=\deg(N'). deg ( M ′ ) = deg ( N ′ ) .
Recall that the flag structure N_{\bullet}=\left(\pi^{\mathrm{Fr}}_{*}\left(\mathcal{L}_{\mathcal{E}}\otimes_{\mathcal{O}_{S^{\mathrm{Fr}}_{a}}\mathcal{F}_{\psi}\right)\right)_{\bullet}} is induced from the flag structure M ∙ = ( π ∗ F r F ψ ) ∙ M_{\bullet}=\left(\pi^{\mathrm{Fr}}_{*}\mathcal{F}_{\psi}\right)_{\bullet} M ∙ = ( π ∗ Fr F ψ ) ∙ . It follows that the induced flag structure on M q ′ M'_{q} M q ′ arises in the same fashion from the induced flag structure on N q ′ N'_{q} N q ′ . Thus they have the same multiset of parabolic weights counted with multiplicity. It follows that
pardeg ( M ′ , M ∙ ′ ) = pardeg ( N ′ , N ∙ ′ ) . (22) \operatorname{pardeg}(M',M'_{\bullet})=\operatorname{pardeg}(N',N'_{\bullet}). \tag*{(22)} pardeg ( M ′ , M ∙ ′ ) = pardeg ( N ′ , N ∙ ′ ) . (22)
which implies the desired isomorphism (25).
By Theorem 5.11, the isomorphism
X a b → β ≃ H i g g s a → α ( C ′ ) \mathsf{X}^{b\to\beta}_{a}\simeq\mathit{Higgs}^{a\to\alpha}(C') X a b → β ≃ Higgs a → α ( C ′ )
in Theorem 5.10 restricts to an isomorphism
X a b → β , s s ≃ H i g g s a → α , s s ( C ′ ) . \mathsf{X}^{b\to\beta,\mathrm{ss}}_{a}\simeq\mathit{Higgs}^{a\to\alpha,\mathrm{ss}}(C'). X a b → β , ss ≃ Higgs a → α , ss ( C ′ ) .
Combining with the isomorphism (25) and the equalities of parabolic degrees (22) and (26), we obtain the desired isomorphism in Theorem 5.2. □
Cartier descent for filtered sheaves
The goal of this section is to give a more explicit description of the isomorphism (17)
Conn 0 τ → β ( C ) ≃ Bun α ( C ′ , m ′ ) \operatorname{Conn}^{\tau\to\beta}_{0}(C)\simeq\operatorname{Bun}^{\alpha}(C',m') Conn 0 τ → β ( C ) ≃ Bun α ( C ′ , m ′ )
in Proposition 5.7, showing that it may be regarded as a generalization of Cartier descent to filtered sheaves. We start by recalling the correspondence between parabolic bundles and locally free filtered sheaves as described in [BY96].
Definition 6.1. A locally free filtered sheaf E ∘ \mathcal{E}^{\circ} E ∘ on ( C , q ) (C,q) ( C , q ) consists of a locally free sheaf E 0 \mathcal{E}_{0} E 0 on C C C , together with an R \mathbb{R} R -indexed collection of locally free subsheaves { E x } x ∈ R ⊂ E 0 ( ∗ q ) \{\mathcal{E}_{x}\}_{x\in\mathbb{R}}\subset\mathcal{E}_{0}(*q) { E x } x ∈ R ⊂ E 0 ( ∗ q ) satisfying the following four conditions:
(1) If x ≤ y x\leq y x ≤ y , then E y ⊆ E x \mathcal{E}_{y}\subseteq\mathcal{E}_{x} E y ⊆ E x ,
(2) For x ≡ y x\equiv y x ≡ y , the inclusion E y ⊆ E x \mathcal{E}_{y}\subseteq\mathcal{E}_{x} E y ⊆ E x is an isomorphism over C \ q C\backslash q C \ q ,
(3) E x + 1 = E x ( − q ) \mathcal{E}_{x+1}=\mathcal{E}_{x}(-q) E x + 1 = E x ( − q ) ,
(4) For sufficiently small ϵ > 0 \epsilon>0 ϵ > 0 , we have E x − ϵ = E x \mathcal{E}_{x-\epsilon}=\mathcal{E}_{x} E x − ϵ = E x .
Given a parabolic bundle ( E , E ∙ , α ) (E,E_{\bullet},\alpha) ( E , E ∙ , α ) with filtration
0 = E 0 ⊂ E 1 ⊂ ⋯ ⊂ E r − 1 ⊂ E r = E q 0=E_{0}\subset E_{1}\subset\cdots\subset E_{r-1}\subset E_{r}=E_{q} 0 = E 0 ⊂ E 1 ⊂ ⋯ ⊂ E r − 1 ⊂ E r = E q
and parabolic weights
α = ( α 1 , α 2 , … , α r ) , 1 > α 1 > α 2 > ⋯ > α r ≥ 0 , \alpha=(\alpha_1,\alpha_2,\ldots,\alpha_r),\qquad1>\alpha_1>\alpha_2>\cdots>\alpha_r\geq0, α = ( α 1 , α 2 , … , α r ) , 1 > α 1 > α 2 > ⋯ > α r ≥ 0 ,
we set α 0 = 1 \alpha_0=1 α 0 = 1 and define
E x = { E , 0 ≤ x ≤ α r , ker ( E → E q / E j ) , α j + 1 < x ≤ α j , 0 ≤ j ≤ r − 1. \mathcal{E}_x=
\begin{cases}
E, & 0\leq x\leq\alpha_r,\\
\ker(E\to E_q/E_j), & \alpha_{j+1}<x\leq\alpha_j,\qquad0\leq j\leq r-1.
\end{cases} E x = { E , ker ( E → E q / E j ) , 0 ≤ x ≤ α r , α j + 1 < x ≤ α j , 0 ≤ j ≤ r − 1.
which extends to a unique collection { E x } x ∈ R \{\mathcal{E}_x\}_{x\in\mathbb{R}} { E x } x ∈ R satisfying the four conditions above. Conversely, given a locally free filtered sheaf { E x } x ∈ R \{\mathcal{E}_x\}_{x\in\mathbb{R}} { E x } x ∈ R , we associate with it a parabolic bundle ( E , E ∙ , α ) (E,E_\bullet,\alpha) ( E , E ∙ , α ) as follows. We define gr x ( E ) : = E x / E x + ϵ \operatorname{gr}_x(\mathcal{E}):=\mathcal{E}_x/\mathcal{E}_{x+\epsilon} gr x ( E ) := E x / E x + ϵ . We set E = E 0 E=\mathcal{E}_0 E = E 0 , set
α = { x ∈ [ 0 , 1 ) ∣ gr x ( E ) ≠ 0 } = { α 1 , α 2 , … , α r } , 1 > α 1 > α 2 > ⋯ > α r ≥ 0 , \alpha=\{x\in[0,1)\mid\operatorname{gr}_x(\mathcal{E})\neq0\}=\{\alpha_1,\alpha_2,\ldots,\alpha_r\},\qquad1>\alpha_1>\alpha_2>\cdots>\alpha_r\geq0, α = { x ∈ [ 0 , 1 ) ∣ gr x ( E ) = 0 } = { α 1 , α 2 , … , α r } , 1 > α 1 > α 2 > ⋯ > α r ≥ 0 ,
and set
E j = E α j / E 1 ⊆ E 0 / E 0 ( − q ) ≅ E q for j = 1 , 2 , … , r . E_j=\mathcal{E}_{\alpha_j}/\mathcal{E}_1\subseteq\mathcal{E}_0/\mathcal{E}_0(-q)\cong E_q\qquad\text{for }j=1,2,\ldots,r. E j = E α j / E 1 ⊆ E 0 / E 0 ( − q ) ≅ E q for j = 1 , 2 , … , r .
It is straightforward to check that these constructions are mutually quasi-inverse functors, thus induce an equivalence between the groupoid of locally free filtered sheaves and the groupoid of parabolic bundles.
Definition 6.2. A filtered logarithmic connection ( { E x } x ∈ R , ∇ ) (\{\mathcal{E}_x\}_{x\in\mathbb{R}},\nabla) ({ E x } x ∈ R , ∇ ) on ( C , q ) (C,q) ( C , q ) consists of a locally free filtered sheaf { E x } x ∈ R \{\mathcal{E}_x\}_{x\in\mathbb{R}} { E x } x ∈ R together with a logarithmic connection
∇ : E 0 ⟶ E 0 ⊗ ω C ( q ) \nabla:\mathcal{E}_0\longrightarrow\mathcal{E}_0\otimes\omega_C(q) ∇ : E 0 ⟶ E 0 ⊗ ω C ( q ) such that ∇ ( E x ) ⊆ E x ⊗ ω C ( q ) \nabla(\mathcal{E}_x)\subseteq\mathcal{E}_x\otimes\omega_C(q) ∇ ( E x ) ⊆ E x ⊗ ω C ( q ) for each x ∈ R x\in\mathbb{R} x ∈ R , where ∇ \nabla ∇ also denotes its natural extension to E 0 ( ∗ q ) \mathcal{E}_0(*q) E 0 ( ∗ q ) .
It is straightforward to check that the equivalence between locally free filtered sheaves and parabolic bundles described above upgrades to an equivalence between filtered logarithmic connections and parabolic connections.
Denote by C o n n 0 p a r ( C ) \mathop{\rm Conn}\nolimits^{\mathrm{par}}_0(C) Conn 0 par ( C ) the moduli stack of parabolic connections on ( C , q ) (C,q) ( C , q ) with zero p p p -curvature, and by B u n p a r ( C ′ ) \mathop{\rm Bun}\nolimits^{\mathrm{par}}(C') Bun par ( C ′ ) the moduli stack of parabolic bundles on ( C ′ , q ′ ) (C',q') ( C ′ , q ′ ) . More precisely, we set
C o n n 0 p a r ( C ) : = ∐ τ , β , τ ↦ β C o n n 0 τ ↦ β ( C ) , \mathop{\rm Conn}\nolimits^{\mathrm{par}}_0(C):=\coprod_{\tau,\beta,\tau\mapsto\beta}\mathop{\rm Conn}\nolimits^{\tau\mapsto\beta}_0(C), Conn 0 par ( C ) := τ , β , τ ↦ β ∐ Conn 0 τ ↦ β ( C ) ,
where the disjoint union ranges over all τ ∈ t ( F p ) / W \tau\in t(\mathbb{F}_p)/W τ ∈ t ( F p ) / W , parabolic weights β \beta β and assignments τ ↦ β \tau\mapsto\beta τ ↦ β ; and we set
B u n p a r ( C ′ ) : = ∐ α , m ′ B u n α ( C ′ , m ′ ) , \mathop{\rm Bun}\nolimits^{\mathrm{par}}(C'):=\coprod_{\alpha,\mathbf{m}'}\mathop{\rm Bun}\nolimits^\alpha(C',\mathbf{m}'), Bun par ( C ′ ) := α , m ′ ∐ Bun α ( C ′ , m ′ ) ,
where the disjoint union ranges over all ordered tuples m ′ \mathbf{m}' m ′ that sum to n n n , and sets of parabolic weights α \alpha α such that # α \#\alpha # α is equal to the length of m ′ \mathbf{m}' m ′ .
Let E = ( E , E ∙ , ∇ E ) ∈ C o n n 0 p a r ( C ) \mathcal{E}=(E,E_\bullet,\nabla_E)\in\mathop{\rm Conn}\nolimits^{\mathrm{par}}_0(C) E = ( E , E ∙ , ∇ E ) ∈ Conn 0 par ( C ) , and let ( { E x } x ∈ R , ∇ E ) (\{\mathcal{E}_x\}_{x\in\mathbb{R}},\nabla_{\mathcal{E}}) ({ E x } x ∈ R , ∇ E ) be the filtered connection corresponding to E \mathcal{E} E . For any x ∈ R x\in\mathbb{R} x ∈ R , we define
F x : = ( F r C ) ∗ ker ( ∇ E : E p x ⟶ E p x ⊗ ω C ( q ) ) . (23) \mathcal{F}_x:=(\mathrm{Fr}_C)_*\ker(\nabla_{\mathcal{E}}:\mathcal{E}_{px}\longrightarrow\mathcal{E}_{px}\otimes\omega_C(q)). \tag*{(23)} F x := ( Fr C ) ∗ ker ( ∇ E : E p x ⟶ E p x ⊗ ω C ( q )) . (23)
Lemma 6.3. The R \mathbb{R} R -collection { F x } x ∈ R \{\mathcal{F}_x\}_{x\in\mathbb{R}} { F x } x ∈ R is a locally free filtered sheaf over ( C ′ , q ′ ) (C',q') ( C ′ , q ′ ) .
Proof. We start by showing F x \mathcal{F}_x F x is locally free for each x x x . Recall that by Theorem 5.5, E \mathcal{E} E is locally isomorphic to the model object E 0 \mathcal{E}^{0} E 0 defined in Section 5.2.1. The fact that each F x \mathcal{F}_x F x is locally free follows from the simple computation that
( Fr C ) ∗ ker ( ∇ 0 : O C ( γ q ) ⟶ O C ( γ q ) ⊗ ω C ( q ) ) = O C ′ ( ⌊ γ p ⌋ q ′ ) . (\operatorname{Fr}_{C})_*\ker\left(\nabla^{0}:\mathcal{O}_{C}(\gamma q)\longrightarrow\mathcal{O}_{C}(\gamma q)\otimes\omega_{C}(q)\right)=\mathcal{O}_{C'}\left(\left\lfloor\frac{\gamma}{p}\right\rfloor q'\right). ( Fr C ) ∗ ker ( ∇ 0 : O C ( γ q ) ⟶ O C ( γ q ) ⊗ ω C ( q ) ) = O C ′ ( ⌊ p γ ⌋ q ′ ) .
for any integer γ \gamma γ . This argument works for E \mathcal{E} E over an arbitrary test k k k -scheme.
The conditions (1), (2), (4) follow from the fact that { E x } x ∈ R \{\mathcal{E}_x\}_{x\in\mathbb{R}} { E x } x ∈ R satisfies those conditions. Let x x x be a local coordinate at q q q . Condition (3) follows from the observation that for any section s s s of E 0 ( q ∗ ) \mathcal{E}_{0}(q*) E 0 ( q ∗ ) around q q q , we have
∇ E ( s ) = 0 ⟺ ∇ E ( x p s ) = 0. \nabla_{\mathcal{E}}(s)=0\Longleftrightarrow\nabla_{\mathcal{E}}(x^{p}s)=0. ∇ E ( s ) = 0 ⟺ ∇ E ( x p s ) = 0.
□ \square □
By Lemma 6.3 , the R \mathbb{R} R -collection of sheaves { F x } x ∈ R \{\mathcal{F}_x\}_{x\in\mathbb{R}} { F x } x ∈ R defined in (27) forms a locally free filtered sheaf on ( C ′ , q ′ ) (C',q') ( C ′ , q ′ ) . Via the correspondence with parabolic bundles, this construction defines a morphism of stacks
Φ : Conn 0 p a r ( C ) ⟶ B u n p a r ( C ′ ) . \Phi:\operatorname{Conn}_{0}^{\mathrm{par}}(C)\longrightarrow\mathcal{Bun}^{\mathrm{par}}(C'). Φ : Conn 0 par ( C ) ⟶ B u n par ( C ′ ) .
Theorem 6.4. For any τ ∈ t ( F p ) / W \tau\in\mathfrak{t}(\mathbb{F}_{p})/W τ ∈ t ( F p ) / W , parabolic weights β \beta β and assignment τ ↦ β \tau\mapsto\beta τ ↦ β , the morphism Φ \Phi Φ restricts to an isomorphism
Conn 0 τ ↦ β ( C ) ≅ B u n α ( C ′ , m ′ ) \operatorname{Conn}_{0}^{\tau\mapsto\beta}(C)\cong\mathcal{Bun}^{\alpha}(C',m') Conn 0 τ ↦ β ( C ) ≅ B u n α ( C ′ , m ′ ) that agrees with the one constructed in Proposition 5.7 . Since (14) gives a bijection between assignments τ ↦ β \tau\mapsto\beta τ ↦ β and pairs ( α , m ′ ) (\alpha,m') ( α , m ′ ) , Φ \Phi Φ induces an isomorphism
Φ : Conn 0 p a r ( C ) → ∼ B u n p a r ( C ′ ) . \Phi:\operatorname{Conn}_{0}^{\mathrm{par}}(C)\xrightarrow{\sim}\mathcal{Bun}^{\mathrm{par}}(C'). Φ : Conn 0 par ( C ) ∼ B u n par ( C ′ ) .
Proof. We denote the isomorphism constructed in Proposition 5.7 by
Φ τ ↦ β ′ : Conn 0 τ ↦ β ( C ) ≅ B u n α ( C ′ , m ′ ) \Phi'_{\tau\mapsto\beta}:\operatorname{Conn}_{0}^{\tau\mapsto\beta}(C)\cong\mathcal{Bun}^{\alpha}(C',m') Φ τ ↦ β ′ : Conn 0 τ ↦ β ( C ) ≅ B u n α ( C ′ , m ′ )
and the restriction of Φ \Phi Φ to Conn 0 τ ↦ β ( C ) \operatorname{Conn}_{0}^{\tau\mapsto\beta}(C) Conn 0 τ ↦ β ( C ) by Φ τ ↦ β \Phi_{\tau\mapsto\beta} Φ τ ↦ β . Let E 0 \mathcal{E}^{0} E 0 be the model object defined in Section 5.2.1. We have the following 2-commutative diagram
B u n A u t ( E 0 ) ( C ) → Lemma 5.6 B u n G α ( C ′ ) G ↓ ↓ Remark 4.1 Conn 0 τ ↦ β ( C ) → Φ τ ↦ β ′ B u n α ( C ′ , m ′ ) \begin{CD}
\mathcal{Bun}_{\mathit{Aut}}(\mathcal{E}^{0})(C) @>{\text{Lemma 5.6}}>> \mathcal{Bun}_{\mathfrak{G}_{\alpha}}(C') \\
@V{\mathrm{G}}VV @VV{\text{Remark 4.1}}V \\
\operatorname{Conn}_{0}^{\tau\mapsto\beta}(C) @>{\Phi'_{\tau\mapsto\beta}}>> \mathcal{Bun}^{\alpha}(C',m')
\end{CD} B u n Aut ( E 0 ) ( C ) G ↓ ⏐ Conn 0 τ ↦ β ( C ) Lemma 5.6 Φ τ ↦ β ′ B u n G α ( C ′ ) ↓ ⏐ Remark 4.1 B u n α ( C ′ , m ′ )
where G \mathrm{G} G is constructed in the proof of Proposition 5.7 . Recall that for any P ∈ B u n A u t ( E 0 ) ( C ) \mathcal{P}\in\mathcal{Bun}_{\mathit{Aut}}(\mathcal{E}^{0})(C) P ∈ B u n Aut ( E 0 ) ( C ) , we have
G ( P ) = P × A u t ( E 0 ) E 0 , \mathrm{G}(\mathcal{P})=\mathcal{P}\times^{\mathit{Aut}(\mathcal{E}^{0})}\mathcal{E}^{0}, G ( P ) = P × Aut ( E 0 ) E 0 ,
with connection and parabolic structure inherited from E 0 \mathcal{E}^{0} E 0 . Let ( { E x 0 } x ∈ R , ∇ E 0 ) \left(\{\mathcal{E}^{0}_{x}\}_{x\in\mathbb{R}},\nabla^{0}_{\mathcal{E}}\right) ( { E x 0 } x ∈ R , ∇ E 0 ) be the filtered connection corresponding to E 0 \mathcal{E}^{0} E 0 . Then the filtered sheaf corresponding to G ( P ) \mathrm{G}(\mathcal{P}) G ( P ) is given by
\mathcal{P}\times^{\mathit{Aut}\left(\{\mathcal{E}^{0}_{x}\}_{x\in\mathbb{R},\nabla^{0}_{\mathcal{E}}\right)}\{\mathcal{E}^{0}_{x}\}_{x\in\mathbb{R}}.}
By the definition of Φ \Phi Φ through (27), we have
Φ τ ↦ β ( G ( P ) ) ≅ ( Fr C ) ∗ P × ( Fr C ) ∗ A u t ( E 0 ) Φ ( E 0 ) . \Phi_{\tau\mapsto\beta}(\mathrm{G}(\mathcal{P}))\cong(\operatorname{Fr}_{C})_*\mathcal{P}\times^{(\operatorname{Fr}_{C})_*\mathit{Aut}(\mathcal{E}^{0})}\Phi(\mathcal{E}^{0}). Φ τ ↦ β ( G ( P )) ≅ ( Fr C ) ∗ P × ( Fr C ) ∗ Aut ( E 0 ) Φ ( E 0 ) .
By the 2-commutative diagram above, we have
Φ τ ↦ β ′ ( G ( P ) ) ≅ ( F r C ) ∗ P × G α ~ F 0 , \Phi'_{\tau\mapsto\beta}(\mathcal{G}(\mathcal{P})) \cong(\mathrm{Fr}_{C})_*\mathcal{P}\times^{\mathcal{G}_{\tilde{\alpha}}}\mathcal{F}^{0}, Φ τ ↦ β ′ ( G ( P )) ≅ ( Fr C ) ∗ P × G α ~ F 0 ,
where F 0 = ( F 0 , F ∙ 0 ) \mathcal{F}^{0}=(F^{0},F^{0}_{\bullet}) F 0 = ( F 0 , F ∙ 0 ) is the model object constructed in Remark 4.1. Therefore in order to show Φ τ ↦ β ≅ Φ τ ↦ β ′ \Phi_{\tau\mapsto\beta}\cong\Phi'_{\tau\mapsto\beta} Φ τ ↦ β ≅ Φ τ ↦ β ′ , it is enough to identify Φ ( E 0 ) ≅ F 0 \Phi(\mathcal{E}^{0})\cong\mathcal{F}^{0} Φ ( E 0 ) ≅ F 0 , such that under this identification, the action of ( F r C ) ∗ A u t ( E 0 ) (\mathrm{Fr}_{C})_*\mathit{Aut}(\mathcal{E}^{0}) ( Fr C ) ∗ Aut ( E 0 ) on Φ ( E 0 ) \Phi(\mathcal{E}^{0}) Φ ( E 0 ) agrees with the action of G α ~ \mathcal{G}_{\tilde{\alpha}} G α ~ on F 0 \mathcal{F}^{0} F 0 through the canonical isomorphism
( F r C ) ∗ A u t ( E 0 ) ≅ G α ~ (\mathrm{Fr}_{C})_*\mathit{Aut}(\mathcal{E}^{0})\cong\mathcal{G}_{\tilde{\alpha}} ( Fr C ) ∗ Aut ( E 0 ) ≅ G α ~
in Lemma 5.6. Let E i 0 = ( O C ( − ν ( τ i ) q ) , ∇ 0 , β ~ o p τ ( i ) ) \mathcal{E}^{0}_{i}=(\mathcal{O}_{C}(-\nu(\tau_i)q),\nabla^{0},\tilde{\beta}^{\tau}_{\mathrm{op}}(i)) E i 0 = ( O C ( − ν ( τ i ) q ) , ∇ 0 , β ~ op τ ( i )) be the rank one parabolic connection with parabolic weight β ~ o p τ ( i ) \tilde{\beta}^{\tau}_{\mathrm{op}}(i) β ~ op τ ( i ) , with the logarithmic connection ∇ 0 \nabla^{0} ∇ 0 induced from the canonical connection on O C \mathcal{O}_{C} O C . Then we have E 0 = ⨁ i = 1 n E i 0 \mathcal{E}^{0}=\bigoplus_{i=1}^{n}\mathcal{E}^{0}_{i} E 0 = ⨁ i = 1 n E i 0 . Note that Φ \Phi Φ respects direct sums, therefore the desired identification Φ ( E 0 ) ≅ F 0 \Phi(\mathcal{E}^{0})\cong\mathcal{F}^{0} Φ ( E 0 ) ≅ F 0 follows from the rank one computation that
Φ ( E i 0 ) ≅ ( O C ′ , α o p α ~ ( i ) ) , where α o p α ~ ( i ) = β ~ o p τ ( i ) − ν ( τ i ) p , \Phi(\mathcal{E}^{0}_{i})\cong(\mathcal{O}_{C'},\alpha^{\tilde{\alpha}}_{\mathrm{op}}(i)),\qquad\text{where }\alpha^{\tilde{\alpha}}_{\mathrm{op}}(i)=\frac{\tilde{\beta}^{\tau}_{\mathrm{op}}(i)-\nu(\tau_i)}{p}, Φ ( E i 0 ) ≅ ( O C ′ , α op α ~ ( i )) , where α op α ~ ( i ) = p β ~ op τ ( i ) − ν ( τ i ) ,
see Lemma 6.5. The agreement of actions follows from the computation in Lemma 5.6. □
Lemma 6.5. For any β ∈ [ 0 , 1 ) \beta\in[0,1) β ∈ [ 0 , 1 ) and γ ∈ { 0 , 1 , 2 , … , p − 1 } \gamma\in\{0,1,2,\ldots,p-1\} γ ∈ { 0 , 1 , 2 , … , p − 1 } , let α = β + γ p \alpha=\frac{\beta+\gamma}{p} α = p β + γ , then we have
Φ ( ( O C ( γ q ) , ∇ 0 , β ) ) ≅ ( O C ′ , α ) . \Phi\bigl((\mathcal{O}_{C}(\gamma q),\nabla^{0},\beta)\bigr)\cong(\mathcal{O}_{C'},\alpha). Φ ( ( O C ( γ q ) , ∇ 0 , β ) ) ≅ ( O C ′ , α ) .
Proof. Let { E x } x ∈ R , ∇ ∙ 0 \{ \mathcal{E}_{x}\}_{x\in\mathbb{R}},\nabla^{0}_{\bullet} { E x } x ∈ R , ∇ ∙ 0 be the filtered connection corresponding to ( O C ( γ q ) , ∇ 0 , β ) (\mathcal{O}_{C}(\gamma q),\nabla^{0},\beta) ( O C ( γ q ) , ∇ 0 , β ) and let { F x } x ∈ R \{\mathcal{F}_{x}\}_{x\in\mathbb{R}} { F x } x ∈ R be the filtered sheaf corresponding to Φ ( ( O C ( γ q ) , ∇ 0 , β ) ) \Phi((\mathcal{O}_{C}(\gamma q),\nabla^{0},\beta)) Φ (( O C ( γ q ) , ∇ 0 , β )) . By definition, we have
F x = ( F r C ) ∗ ker ( ∇ E 0 : E p x ⟶ E p x ⊗ ω C ( q ) ) \mathcal{F}_{x}=(\mathrm{Fr}_{C})_*\ker\left(\nabla^{0}_{\mathcal{E}}:\mathcal{E}_{\mathrm{p}x}\longrightarrow\mathcal{E}_{\mathrm{p}x}\otimes\omega_{C}(q)\right) F x = ( Fr C ) ∗ ker ( ∇ E 0 : E p x ⟶ E p x ⊗ ω C ( q ) )
as in (27). Setting x = 0 x=0 x = 0 , we have
F 0 = ( F r C ) ∗ ker ( ∇ 0 : O C ( γ q ) ⟶ O C ( γ q ) ⊗ ω C ( q ) ) = O C ′ . \mathcal{F}_{0}=(\mathrm{Fr}_{C})_*\ker\left(\nabla^{0}:\mathcal{O}_{C}(\gamma q)\longrightarrow\mathcal{O}_{C}(\gamma q)\otimes\omega_{C}(q)\right)=\mathcal{O}_{C'}. F 0 = ( Fr C ) ∗ ker ( ∇ 0 : O C ( γ q ) ⟶ O C ( γ q ) ⊗ ω C ( q ) ) = O C ′ .
Note that the only x ∈ [ 0 , 1 ) x\in[0,1) x ∈ [ 0 , 1 ) where F \mathcal{F} F jumps is the one where E p x \mathcal{E}_{\mathrm{p}x} E p x jumps from O C \mathcal{O}_{C} O C to O C ( − q ) \mathcal{O}_{C}(-q) O C ( − q ) , which happens when p x = β + γ px=\beta+\gamma p x = β + γ . □
Remark 6.6. The inverse of
Φ : C o n n 0 p a r ( C ) ⟶ B u n p a r ( C ′ ) \Phi:\mathop{\mathrm{Conn}}\nolimits_{0}^{\mathrm{par}}(C)\longrightarrow\mathop{\mathrm{Bun}}\nolimits^{\mathrm{par}}(C') Φ : Conn 0 par ( C ) ⟶ Bun par ( C ′ )
can be described as follows. Let { F x } x ∈ R \{\mathcal{F}_{x}\}_{x\in\mathbb{R}} { F x } x ∈ R be a locally free filtered sheaf on ( C ′ , q ′ ) (C',q') ( C ′ , q ′ ) . We define
E x : = ∑ γ = 0 p − 1 ( F r C ∗ F ( x + γ ) / p ) ( γ q ) = ( F r C ∗ F x ) ( ∗ q ) . \mathcal{E}_{x}:=\sum_{\gamma=0}^{p-1}(\mathrm{Fr}_{C}^{*}\mathcal{F}_{(x+\gamma)/p})(\gamma q)=(\mathrm{Fr}_{C}^{*}\mathcal{F}_{x})(*q). E x := γ = 0 ∑ p − 1 ( Fr C ∗ F ( x + γ ) / p ) ( γ q ) = ( Fr C ∗ F x ) ( ∗ q ) .
It is clear that { E x } x ∈ R \{\mathcal{E}_{x}\}_{x\in\mathbb{R}} { E x } x ∈ R is a locally free filtered sheaf on ( C , q ) (C,q) ( C , q ) . The restriction of the canonical connection on ( F r C ∗ F x ) ( ∗ q ) (\mathrm{Fr}_{C}^{*}\mathcal{F}_{x})(*q) ( Fr C ∗ F x ) ( ∗ q ) gives rise to a connection ∇ 0 \nabla^{0} ∇ 0 on each E x \mathcal{E}_{x} E x , and we have ( { E x } x ∈ R , ∇ 0 ) ∈ C o n n 0 p a r ( C ) (\{\mathcal{E}_{x}\}_{x\in\mathbb{R}},\nabla^{0})\in\mathop{\mathrm{Conn}}\nolimits_{0}^{\mathrm{par}}(C) ({ E x } x ∈ R , ∇ 0 ) ∈ Conn 0 par ( C ) . Let
Ψ : B u n p a r ( C ′ ) ⟶ C o n n 0 p a r ( C ) \Psi:\mathop{\mathrm{Bun}}\nolimits^{\mathrm{par}}(C')\longrightarrow\mathop{\mathrm{Conn}}\nolimits_{0}^{\mathrm{par}}(C) Ψ : Bun par ( C ′ ) ⟶ Conn 0 par ( C )
be the morphism of stacks mapping { F x } x ∈ R \{\mathcal{F}_{x}\}_{x\in\mathbb{R}} { F x } x ∈ R to ( { E x } x ∈ R , ∇ 0 ) (\{\mathcal{E}_{x}\}_{x\in\mathbb{R}},\nabla^{0}) ({ E x } x ∈ R , ∇ 0 ) . Note that
F r C ∗ F x = ∑ γ = 0 p − 1 ( F r C ∗ F ( p x + γ ) / p ) ( γ q ) ⊆ ( F r C ∗ F x ) ( ( p − 1 ) q ) , \mathrm{Fr}_{C}^{*}\mathcal{F}_{x}=\sum_{\gamma=0}^{p-1}(\mathrm{Fr}_{C}^{*}\mathcal{F}_{(px+\gamma)/p})(\gamma q)\subseteq(\mathrm{Fr}_{C}^{*}\mathcal{F}_{x})((p-1)q), Fr C ∗ F x = γ = 0 ∑ p − 1 ( Fr C ∗ F ( p x + γ ) / p ) ( γ q ) ⊆ ( Fr C ∗ F x ) (( p − 1 ) q ) ,
therefore
( F r C ) ∗ ker ( ∇ 0 ∣ F r C ∗ F x ) ⊆ ( Φ ∘ Ψ ( { F x } x ∈ R ) ) x ⊆ ( F r C ) ∗ ker ( ∇ 0 ∣ ( F r C ∗ F x ) ( ( p − 1 ) q ) ) . (\mathrm{Fr}_{C})_* \ker\left(\nabla^0\big|_{\mathrm{Fr}_{C}^{*}\mathcal{F}_x}\right) \subseteq\left(\Phi\circ\Psi\left(\{\mathcal{F}_x\}_{x\in\mathbb{R}}\right)\right)_x \subseteq(\mathrm{Fr}_{C})_* \ker\left(\nabla^0\big|_{(\mathrm{Fr}_{C}^{*}\mathcal{F}_x)((p-1)q)}\right). ( Fr C ) ∗ ker ( ∇ 0 Fr C ∗ F x ) ⊆ ( Φ ∘ Ψ ( { F x } x ∈ R ) ) x ⊆ ( Fr C ) ∗ ker ( ∇ 0 ( Fr C ∗ F x ) (( p − 1 ) q ) ) .
We have
( F r C ) ∗ ker ( ∇ 0 ∣ F r C ∗ F x ) = ( F r C ) ∗ ker ( ∇ 0 ∣ ( F r C ∗ F x ) ( ( p − 1 ) q ) ) = F x , (\mathrm{Fr}_{C})_* \ker\left(\nabla^0\big|_{\mathrm{Fr}_{C}^{*}\mathcal{F}_x}\right) = (\mathrm{Fr}_{C})_* \ker\left(\nabla^0\big|_{(\mathrm{Fr}_{C}^{*}\mathcal{F}_x)((p-1)q)}\right) = \mathcal{F}_x, ( Fr C ) ∗ ker ( ∇ 0 Fr C ∗ F x ) = ( Fr C ) ∗ ker ( ∇ 0 ( Fr C ∗ F x ) (( p − 1 ) q ) ) = F x ,
therefore
( Φ ∘ Ψ ( { F x } x ∈ R ) ) x = F x . \left(\Phi\circ\Psi\left(\{\mathcal{F}_x\}_{x\in\mathbb{R}}\right)\right)_x = \mathcal{F}_x. ( Φ ∘ Ψ ( { F x } x ∈ R ) ) x = F x .
Remark 6.7. The formulation of Cartier descent for filtered sheaves through (23) potentially extends to higher dimensional varieties. We plan to investigate this generalization in future work.
References
Institute of Science and Technology Austria (ISTA), Klosterneuburg, Austria
Email address: shiyu.shen.math@gmail.com