betaFeedback

WorksMathematics

Essential dimension of geometric local systems

Abstract

Let f : X → B be a smooth projective family over a smooth complex variety, and let H = (Rmf∗Z)/tors. We prove that there is an integer N such that, whenever pn ≥N, the essential dimension at p of H/pn is at least the dimension of its period image. This proves the Brosnan–Fakhruddin conjecture for these local systems. The proof compares the geometric Sen operator with the Higgs field. Integral Hodge–Tate theory gives a bound uniform in large primes. For each remaining prime, a bounded rational comparison at one geometric point suffices when n is large.

Introduction

Let BB be a smooth irreducible complex algebraic variety, and let H\mathrm{H} be an integral variation of Hodge structure on BB. For a positive integer NN, the finite local system H/N=H⊗Z/NZ\mathrm{H}/N = \mathrm{H} \otimes\mathbb{Z}/N\mathbb{Z} determines a finite étale cover of BB. Its essential dimension at a prime pp is the least dimension of a variety from which this cover can be obtained by rational pullback, after a generically finite base change of degree prime to pp. We recall the field-theoretic definition in § 4.

Motivated by the results for Shimura varieties in [6], Brosnan and Fakhruddin conjectured that, for a polarized integral variation with period map Φ\Phi, there is an integer N0N_0 such that

ed⁡p(H/pn)≥dim⁡Φ(B)if pn≥N0(1)\operatorname{ed}_p(\mathbb{H}/p^n) \ge\dim\Phi(B) \qquad\text{if } p^n \ge N_0 \tag*{(1)}

[3 §1.5, Conjecture 1]. We prove this assertion when H\mathbb{H} is the full cohomology of a smooth projective family, modulo torsion.

Theorem A. Let f:X→Bf : X \to B be a smooth projective morphism, with BB a smooth irreducible complex variety, and fix m≥0m \ge0. Put H=(Rmf∗Z)/tors⁡\mathbb{H} = (R^m f_*\mathbb{Z})/\operatorname{tors}, and let dd be the dimension of its period image. There is an integer N≥2N \ge2 such that, for every prime pp and every integer n≥1n \ge1 with pn≥Np^n \ge N,

ed⁡p(H/pn)≥d.\operatorname{ed}_p(\mathbb{H}/p^n) \ge d.

We first prove the inequality for H/p\mathbb{H}/p at sufficiently large primes. We then treat the finitely many remaining primes, allowing the required exponent to depend on pp. We do not know how to extend the result to an arbitrary variation of Hodge structure and this remains an interesting question.

A sketch of the proof

For simplicity we sketch only the argument for pp sufficiently large. The argument for the remaining primes is given in § 4.3. It is in fact somewhat easier, since the result is only required to hold with nn sufficiently large. A standard argument reduces us to a statement for a proper smooth family over a finite extension of Qp\mathbb{Q}_p, however it will be useful to give a precise formulation. Let KK be a finite extension of Qp\mathbb{Q}_p, with ring of integers OK\mathcal{O}_K and residue field kk. Set C=K‾^C = \widehat{\overline{K}}. Let f:X→Bf : \mathcal{X} \to\mathcal{B} be smooth projective, where B\mathcal{B} is smooth over OK\mathcal{O}_K and its generic and special fibres are geometrically integral. Assume that the Hodge cohomology sheaves Raf∗ΩX/BbR^a f_*\Omega^b_{\mathcal{X}/\mathcal{B}} are locally free and commute with base change. Assume also that the relative de Rham cohomology and its Hodge filtration are locally free and commute with base change, and that T=RmfK,eˊt∗Zp\mathbb{T} = R^m f_{K,\acute{e}t*}\mathbb{Z}_p is finite free.

Write

Ei=Rm−if∗ΩX/Bi,θi:Ei⟶Ei−1⊗ΩB/OK1E^i = R^{m-i}f_*\Omega^i_{\mathcal{X}/\mathcal{B}}, \qquad\theta_i : E^i \longrightarrow E^{i-1} \otimes\Omega^1_{\mathcal{B}/\mathcal{O}_K}

for the Hodge bundles and their Higgs field, induced by the Gauss–Manin connection. We set Ei=0E^i = 0 outside 0≤i≤m0 \le i \le m. The rank of the Higgs field means the rank of the contraction map

θ:TB/OK⟶⨁iHom⁡(Ei,Ei−1),D⟼(θi(D))i.(2)\theta: T_{\mathcal{B}/\mathcal{O}_K} \longrightarrow\bigoplus_i \operatorname{Hom}(E^i,E^{i-1}), \qquad D \longmapsto(\theta_i(D))_i. \tag*{(2)}

Let d0d_0 be its generic rank on the special fibre.

Theorem B. Under these hypotheses,

ed⁡C,p(T/p∣BC)≥d0if p>m+3.\operatorname{ed}_{C,p}(\mathbb{T}/p|_{B_C}) \ge d_0 \qquad\text{if } p > m+3.

To prove Theorem B, let D=Spf⁡OC⟨z⟩\mathcal{D} = \operatorname{Spf}\mathcal{O}_C\langle z\rangle. We again write f:X→Bf : X \to B for the generic fiber of ff, and we denote by X^→B^\widehat{X} \to\widehat{B} the pp-adic completion of ff. Suppose that s:B→Zs : B \to Z is a rational map over CC such that dim⁡Z<d0\dim Z < d_0, and T/p\mathbb{T}/p descends to ZZ. (We ignore here the auxiliary prime to pp covering, which is straightforward to deal with.) The idea is to construct a map of formal schemes γ:D→B^\gamma: \mathcal{D} \to\widehat{B} such that s∘γ∣Cs \circ\gamma|_C is well defined and constant. Then γ∗(T/p)\gamma^*(\mathbb{T}/p) is constant, and the theory of the geometric Sen operator gives a contradiction, provided the pullback of the Kodaira–Spencer map θ\theta is non-zero modulo pε′′p^{\varepsilon''} at the origin, where ε′∈Q\varepsilon' \in\mathbb{Q} satisfies

ε:=1/(p−1)<ε′<1−(m+1)ε=ε′′.\varepsilon:= 1/(p-1) < \varepsilon' < 1-(m+1)\varepsilon= \varepsilon''.

We normalize v(p)=1v(p)=1 and fix a scalar pε′∈Cp^{\varepsilon'} \in C of valuation ε′\varepsilon'. To construct γ\gamma we choose a suitable point x∈B^(OC)x \in\widehat{B}(\mathcal{O}_C) and a derivation DD of OB\mathcal{O}_B (defined on an open containing xx) such that D(OZ)=0D(\mathcal{O}_Z)=0, and x∗(θ(D))x^*(\theta(D)) is non-zero modulo the maximal ideal of OC\mathcal{O}_C, and define γ\gamma by

γ∗(a)=∑j=0∞pε′jj!x∗Dj(a)zj.\gamma^*(a)=\sum_{j=0}^{\infty}\frac{p^{\varepsilon'j}}{j!}x^*D^j(a)z^j.

Crucially, the existence of γ\gamma satisfying these properties does not depend on the integral properties of the map ss. Such a dependence would be fatal, as ss depends on pp.

Let us explain the use of the geometric Sen operator in more detail. Write D\mathbb{D} for the generic fiber of D\mathcal{D}. For a finite free Zp\mathbb{Z}_p-local system L\mathbb{L} on D\mathbb{D}, put V+=L⊗O^D+V^+=\mathbb{L}\otimes\widehat{\mathcal{O}}_{\mathbb{D}}^+ and V=V+[1/p]V=V^+[1/p], where O^D+\widehat{\mathcal{O}}_{\mathbb{D}}^+ is the completed integral structure sheaf on the pro-étale site of D\mathbb{D}. Its geometric Sen operator is a map SL:V→V⊗ΩD/C1(−1)S_{\mathbb{L}}:V\to V\otimes\Omega^1_{\mathbb{D}/C}(-1), where (−1)(-1) denotes the Tate twist. If L/p\mathbb{L}/p is constant and p>2p>2, Proposition 2.1.6 gives

SL(V+)⊂pV+⊗OCOC dz(−1).(3)S_{\mathbb{L}}(V^+)\subset pV^+\otimes_{\mathcal{O}_C}\mathcal{O}_C\,dz(-1). \tag*{(3)}

Apply this to L=γ∗T\mathbb{L}=\gamma^*\mathbb{T} so L/p\mathbb{L}/p is constant. Under the hypotheses of Theorem B, we use Scholze’s results [14] to control the comparison between the étale and Hodge lattices in VV, and Howe’s theorem [9] identifies the graded Sen operator with θγ=γ∗(θ)∘dγ\theta_\gamma=\gamma^*(\theta)\circ d\gamma. These comparisons turn (1.1.2) into

v(pε′γ∗(θ(D))∣0)=v(θγ(∂z)∣0)≥1−(m+1)ε>ε′(4)v(p^{\varepsilon'}\gamma^*(\theta(D))|_0)=v(\theta_\gamma(\partial_z)|_0)\geq1-(m+1)\varepsilon>\varepsilon' \tag*{(4)}

(Proposition 2.4.1). This is a contradiction as v(γ∗(θ(D))∣0)=0v(\gamma^*(\theta(D))|_0)=0.

In fact, in the proof, it is more convenient to take xx to be the geometric point associated with the valuation at the generic point of the special fibre of BB. So xx is defined over a complete algebraically closed extension C′/CC'/C. The disc D\mathbb{D} is then defined over OC′\mathcal{O}_{C'}, but the argument above remains unchanged.

We remark that it should be possible to adapt the above proof, at least for sufficiently large pp, to other versions of integral pp-adic Hodge theory, for example Faltings’ integral crystalline theory [5], or the prismatic theory of Bhatt–Scholze [2]. As remarked above, the key observation is that the map γ\gamma above can be chosen without any assumptions on the integral behaviour of the map ss. We leave this as an exercise for the reader.

Relation to earlier work

In the paper [6] with Farb and Wolfson, we used deformation theory at ordinary points to prove incompressibility of congruence covers. Fakhruddin and Saini extended the finite group scheme method to certain nonordinary cases in [4]. At the time of writing [6], it was already clear to us that, morally speaking, there should be an argument with the Kodaira–Spencer map, but we were unable to make this work without assuming good reduction properties of the map ss.

The paper [10] proves pp-incompressibility of principal congruence covers of Shimura varieties at all sufficiently large good primes. For exceptional groups the results of the present paper do not contain those of [10], as the corresponding variations of Hodge structures are not known to be motivic.

In [7], Farb, Kisin, and Wolfson use prismatic cohomology to obtain lower bounds for essential dimension from differential forms. Here we compare the Hodge bundles with the lattices induced by the local system on its Hodge–Tate graded pieces. Proposition 2.3.1 gives the required bound by retaining the integral Kummer complexes in Scholze’s relative Hodge–Tate comparison [14]. We use the local integral calculation and décalage estimates of Bhatt, Morrow, and Scholze [1], in the relative form of Gaisin and Koshikawa [8]. Geometric Sen theory [13] and Howe’s theorem [9] then relate this comparison to the Higgs field. We apply these results directly to the family over the formal disc.

The argument at a fixed prime uses only the rational comparison and geometric Sen theory. It is given in § 4.3; § 4.4 combines the two arguments to prove Theorem A.

AI use

Extensive use of GPT Astra was made both to arrive at the proofs of the results presented here, and to write a first draft of the manuscript. Significant human effort was needed to check and refine the arguments, and make the paper more readable. Part of this work was done at The Overhang, held in San Francisco on 9–10 September 2026. We thank the organizers and other participants for a very pleasant experience.

pp-adic Hodge theory

Constancy and the geometric Sen operator

We first prove an integral estimate on a disc.

Let CC be any complete algebraically closed extension of Qp\mathbb{Q}_p. Write OC\mathcal{O}_C for its valuation ring and mC\mathfrak{m}_C for the maximal ideal, and normalize v(p)=1v(p)=1. Fix compatible pp-power roots of unity and set

ε=1p−1.\varepsilon= \frac{1}{p-1}.

For a rational number aa, the notation paMp^aM denotes the multiple of an OC\mathcal{O}_C-module MM by a scalar of valuation aa; it is independent of the scalar chosen. If MM is finite free, and m∈Mm \in M, we write v(m)v(m) for the minimum of v(λ(m))v(\lambda(m)) over linear maps λ:M→OC\lambda: M \to\mathcal{O}_C (which is well defined).

An almost zero module is a module killed by mC\mathfrak{m}_C. The almost category is the quotient by these modules; a superscript aa denotes passage to this category. When working there we suppress this superscript. If MM is finite free and x∈M[1/p]x \in M[1/p] satisfies mCx⊂paM\mathfrak{m}_C x \subset p^aM, then v(x)≥av(x) \ge a.

For a smooth rigid space B/CB/C, write BproeˊtB_{\mathrm{pro\acute{e}t}} for its pro-étale site and O^B+\widehat{\mathcal{O}}_B^+ for the completed integral structure sheaf. Put O^B=O^B+[1/p]\widehat{\mathcal{O}}_B = \widehat{\mathcal{O}}_B^+[1/p]. For a finite free Zp\mathbb{Z}_p-local system T\mathbb{T} on BB, put

T+=T⊗ZpO^B+,V=V+[1/p].\mathbb{T}^+ = \mathbb{T} \otimes_{\mathbb{Z}_p} \widehat{\mathcal{O}}_B^+,\qquad V = V^+[1/p].

The geometric Sen operator is the O^B\widehat{\mathcal{O}}_B-linear Higgs field

ST:V⟶V⊗OBΩB/C1(−1)(5)S_{\mathbb{T}} : V \longrightarrow V \otimes_{\mathcal{O}_B} \Omega^1_{B/C}(-1) \tag*{(5)}

of [13]. Here (i)(i) is the usual Tate twist.

Let tt be an indeterminate and set

A+=OC⟨t,t−1⟩,A∞+=OC⟨t±1/p∞⟩,A=A+[1/p],A∞=A∞+[1/p].A^+ = \mathcal{O}_C\langle t,t^{-1}\rangle,\qquad A_\infty^+ = \mathcal{O}_C\langle t^{\pm1/p^\infty}\rangle,\qquad A = A^+[1/p],\qquad A_\infty= A_\infty^+[1/p].

The second ring is the pp-adic completion of the union of the finite-level rings. The map Y=Spa⁡(A∞,A∞+)→Spa⁡(A,A+)Y = \operatorname{Spa}(A_\infty, A_\infty^+) \to\operatorname{Spa}(A, A^+) is an affinoid perfectoid pro-étale cover with group Γ=Zp(1)\Gamma= \mathbb{Z}_p(1). Let σ∈Γ\sigma\in\Gamma be the generator satisfying σ(t1/pj)=ζpjt1/pj\sigma(t^{1/p^j}) = \zeta_{p^j}t^{1/p^j} for all jj.

We also write vv for the Gauss valuation on A∞A_\infty and for the minimum valuation of the entries of a matrix over A∞A_\infty.

Lemma 2.1.2. Let T\mathbb{T} be a finite free Zp\mathbb{Z}_p-local system on Spa⁡(A,A+)\operatorname{Spa}(A, A^+). Suppose T/pn\mathbb{T}/p^n is constant. For every rational 0≤β<n0 \leq\beta< n, the module H0(Y,V+)H^0(Y,\mathbb{V}^+) has an A∞+A_\infty^+-basis in which the matrix of σ\sigma is congruent to 11 modulo pβp^\beta.

Reference. Put h=rk⁡Th = \operatorname{rk}\mathbb{T}. A choice of basis of the constant local system T/pn\mathbb{T}/p^n gives an isomorphism of sheaves

V+/pn≃(O^+/pn)h.\mathbb{V}^+/p^n \simeq(\widehat{\mathcal{O}}^+/p^n)^h.

By [16 Lemma 5.9], applied with a=na = n, the module M=H0(Y,V+)M = H^0(Y,\mathbb{V}^+) is finite free of rank hh and, for 0<β<n0 < \beta< n, the trivialization of V+/pn\mathbb{V}^+/p^n induces a Γ\Gamma-equivariant isomorphism

M/pβM≃(A∞+/pβ)h.M/p^\beta M \simeq(A_\infty^+/p^\beta)^h.

and the matrix of σ\sigma on M/pβMM/p^\beta M is 11 mod pβp^\beta in this basis. The proof in loc. cit. uses only almost acyclicity on affinoid perfectoid objects and almost algebra, so applies over the present field CC. □

Lemma 2.1.3. Let MM be a finite free A∞+A_\infty^+-module of rank hh with continuous semilinear Γ\Gamma-action. If its matrix PP for σ\sigma satisfies v(P−1)≥β>εv(P-1) \geq\beta> \varepsilon, an integral change of basis makes PP a matrix P0∈1+pβMh(A+)P_0 \in1 + p^\beta M_h(A^+).

Proof. Projection onto the monomials with integral exponents gives a bounded AA-linear decomposition

A∞=A⊕N.(6)A_\infty= A \oplus N. \tag*{(6)}

The projection to AA has norm one. On a monomial tu/pst^{u/p^s} with s≥1s \geq1 and p∤up \nmid u, the operator σ−1\sigma- 1 acts by multiplication by ζpsu−1\zeta_{p^s}^u - 1. The operator (σ−1)−1(\sigma- 1)^{-1} on the CC-linear span of these monomials is bounded, so extends to its completion NN. Consequently

v((σ−1)−1x)≥v(x)−ε(x∈N).(7)v((\sigma- 1)^{-1}x) \geq v(x) - \varepsilon\qquad(x \in N). \tag*{(7)}

Write P=1+D+EP = 1 + D + E using (2.1.4), with DD and EE in Mh(A)M_h(A) and Mh(N)M_h(N) respectively. Suppose v(E)≥s≥βv(E) \geq s \geq\beta. Put

Q=−(σ−1)−1E,B=1+Q,P′=B−1Pσ(B).Q = -(\sigma- 1)^{-1}E,\qquad B = 1 + Q,\qquad P' = B^{-1}P\sigma(B).

By (2.1.5), v(Q)≥s−ε>0v(Q) \geq s - \varepsilon> 0; both BB and its inverse are integral. Using σ(B)−B=−E\sigma(B) - B = -E, we first compute

B(P′−1)=(P−1)σ(B)−E=D+(D+E)σ(Q).B(P' - 1) = (P - 1)\sigma(B) - E = D + (D + E)\sigma(Q).

Subtracting BD=D+QDBD = D + QD gives

B(P′−1−D)=(D+E)σ(Q)−QD.B(P' - 1 - D) = (D + E)\sigma(Q) - QD.

Both terms on the right have valuation at least s+β−εs + \beta- \varepsilon. Multiplying by the integral matrix B−1B^{-1} gives the same bound for P′−1−DP' - 1 - D. Now replace PP by P′P' and repeat the above. Each iteration increases v(P−1−D)v(P - 1 - D) by at least β−ε\beta- \varepsilon. The products of the changes of basis converge in GL⁡h(A∞+)\operatorname{GL}_h(A_\infty^+), and taking the limit of the matrices PP gives the required P0P_0. □

Proposition 2.1.6. Let TT be a finite free Zp\mathbb{Z}_p-local system on the generic fibre D\mathbb{D} of D=Spf⁡OC⟨z⟩\mathcal{D} = \operatorname{Spf}\mathcal{O}_C\langle z\rangle. If T/pnT/p^n is constant and n>εn > \varepsilon, then

ST⁡(V+)⊂pnV+⊗OCOC dz(−1).\operatorname{ST}(V^+) \subset p^n V^+ \otimes_{\mathcal{O}_C} \mathcal{O}_C\, dz(-1).

as an inclusion of sheaves on Dproeˊt\mathbb{D}_{\mathrm{pro\acute et}}.

Proof. The formal opens where zz, respectively 1+z1+z, is invertible cover D\mathcal{D}. Each is isomorphic to Spf⁡A+\operatorname{Spf} A^+, with t=zt=z, respectively t=1+zt=1+z. On these opens dlog⁡td\log t is a unit multiple of dzdz, so it suffices to prove the estimate on Spa⁡(A,A+)\operatorname{Spa}(A,A^+).

Fix a rational β\beta with ε<β<n\varepsilon< \beta< n. Lemmas 2.1.2 and 2.1.3 give an integral basis of M=H0(Y,V+)M = H^0(Y,V^+) whose AA-span M0M_0 is Γ\Gamma-stable and satisfies M0⊗AA∞=M[1/p]M_0 \otimes_A A_\infty= M[1/p]. Thus M0M_0 is a decompletion of VV on the Kummer tower. The action of σ\sigma on M0M_0 is AA-linear, with matrix P0∈1+pβMh(A+)P_0 \in1+p^\beta M_h(A^+), where h=rk⁡Th=\operatorname{rk} T. Let σ∨∈Zp(−1)\sigma^\vee\in\mathbb{Z}_p(-1) be dual to σ\sigma. By [13], Proposition 3.1.2 and proof of Proposition 3.2.3, the geometric Sen operator is obtained by extending

M0⟶M0⊗AΩA/C1(−1),m⟼(log⁡P0)m⊗dlog⁡t⊗σ∨\begin{aligned} M_0 &\longrightarrow M_0 \otimes_A \Omega^1_{A/C}(-1),\\ m &\longmapsto(\log P_0)m \otimes d\log t \otimes\sigma^\vee \end{aligned}

O^\widehat{\mathcal{O}}-linearly. The jjth term of the logarithm has valuation at least jβ−vp(j)≥βj\beta-v_p(j) \ge\beta, since β>1/(p−1)\beta> 1/(p-1). This proves that

ST⁡(V+)⊂pβV+⊗OCOC dz(−1).\operatorname{ST}(V^+) \subset p^\beta V^+ \otimes_{\mathcal{O}_C} \mathcal{O}_C\, dz(-1).

The proposition now follows by intersecting the right hand side of this inclusion over all β<n\beta< n, and noting that ⋂δ>0p−δO^+=O^+\bigcap_{\delta>0}p^{-\delta}\widehat{\mathcal{O}}^+ = \widehat{\mathcal{O}}^+.

Lemma 2.1.7. For a finite free Zp\mathbb{Z}_p-local system on BB, the geometric Sen operator commutes with extension of complete algebraically closed constant fields, and with pullback along a morphism of smooth rigid spaces over a field.

Proof. Pullback over a fixed field is [13], Theorem 1.0.3(5). For an extension C′/CC'/C, work locally on a toric chart of BB. The construction in [13], Proposition 3.2.3 computes the operator by differentiating the action of an open subgroup of the Kummer Galois group on a finite projective module at a finite level of the tower. Scalar extension to the infinite tower recovers VV from this module. Completed scalar extension from CC to C′C' preserves this description: the action matrices, their convergent logarithms, Kummer differentials, and Tate twists all commute with it. This proves the assertion.

The Hodge–Tate filtration

We compare the lattices supplied by the local system and the Hodge bundles. The comparison will be applied to the family on the formal disc constructed in §3.

Let K/QpK/\mathbb{Q}_p be finite, and let fK:XK→BKf_K:X_K \to B_K be the analytification of a smooth projective morphism of smooth KK-varieties. Fix m≥0m \ge0, and put

TK=(Rm(fK)eˊt∗Zp)/tors,EKi=Rm−i(fK)∗ΩXK/BKi.T_K = (R^m(f_K)_{\acute et*}\mathbb{Z}_p)/\mathrm{tors}, \qquad E_K^i = R^{m-i}(f_K)_*\Omega^i_{X_K/B_K}.

The relative de Rham comparison [14], Theorem 8.8 and Remark 8.9 identifies the filtered bundle with connection associated to TK[1/p]T_K[1/p] with Rm(fK)∗ΩXK/BK∙R^m(f_K)_*\Omega^\bullet_{X_K/B_K}. Write θi:EKi→EKi−1⊗ΩBK/K1\theta_i:E_K^i \to E_K^{i-1}\otimes\Omega^1_{B_K/K} for its Higgs field.

Fix an embedding of KK into CC, and let g:B→(BK)Cg:B\to(B_K)_C be a morphism of smooth rigid spaces over CC. Write gK:B→BKg_K:B\to B_K for its composite with (BK)C→BK(B_K)_C\to B_K, and put

T=gK∗TK,EBi=gK∗EKi,V+=T⊗ZpO^B+,V=V+[1/p].\mathbb{T}=g_K^*\mathbb{T}_K,\qquad E_B^i=g_K^*E_K^i,\qquad V^+=\mathbb{T}\otimes_{\mathbb{Z}_p}\widehat{\mathcal{O}}_B^+,\qquad V=V^+[1/p].

We also write θi\theta_i for the pullback Higgs field on EBiE_B^i, with values in EBi−1⊗ΩB/C1E_B^{i-1}\otimes\Omega^1_{B/C}.

Let CB\mathcal{C}_B be the pullback of gr⁡0OBdR,BK\operatorname{gr}^0\mathcal{O}\mathbb{B}_{\mathrm{dR},B_K} to BproeˊtB_{\mathrm{pro\acute{e}t}} along gKg_K, with coefficients in O^B\widehat{\mathcal{O}}_B. Taking the degree-zero graded part of the relative de Rham comparison and pulling back gives

ι:V⊗O^BCB⟶∼⨁j=0mEBj⊗OBCB(−j).(8)\iota:V\otimes_{\widehat{\mathcal{O}}_B}\mathcal{C}_B\mathrel{\overset{\sim}{\longrightarrow}}\bigoplus_{j=0}^{m}E_B^j\otimes_{\mathcal{O}_B}\mathcal{C}_B(-j). \tag*{(8)}

Define FiVF_iV to be the inverse image of the sum over j≤ij\le i under v↦ι(v⊗1)v\mapsto\iota(v\otimes1). This is the Hodge–Tate filtration

0=F−1V⊂F0V⊂⋯⊂FmV=V.(9)0=F_{-1}V\subset F_0V\subset\cdots\subset F_mV=V. \tag*{(9)}

Thus Fi=Fil⁡HT−iF_i=\operatorname{Fil}^{-i}_{\mathrm{HT}} in decreasing notation. By [9], proof of Lemma 3.2.1, the graded maps of ι−1\iota^{-1} are the scalar extensions to CB\mathcal{C}_B of isomorphisms

κi:EBi⊗O^B(−i)⟶∼gr⁡iFV.\kappa_i:E_B^i\otimes\widehat{\mathcal{O}}_B(-i)\mathrel{\overset{\sim}{\longrightarrow}}\operatorname{gr}^{F}_iV.

Lemma 2.2.3 (Howe). The geometric Sen operator satisfies

ST(FiV)⊂Fi−1V⊗ΩB/C1(−1).S_{\mathbb{T}}(F_iV)\subset F_{i-1}V\otimes\Omega^1_{B/C}(-1).

Under κi\kappa_i and κi−1\kappa_{i-1}, its induced graded map is $\theta_i.

Proof. Over BKB_K this is [9], Lemma 3.2.1. For BB the assertion follows by pulling back along gKg_K, using Lemma 2.1.7.

The integral bound

Put X=XK×BKBX=X_K\times_{B_K}B, and suppose that f:X→Bf:X\to B is the rigid generic fibre of a smooth proper morphism f:X→B\mathfrak{f}:\mathfrak{X}\to\mathfrak{B} of smooth pp-adic formal schemes locally of finite type over OC\mathcal{O}_C. Assume that all Raf∗ΩX/BiR^af_*\Omega^i_{\mathfrak{X}/\mathfrak{B}} are locally free and commute with base change, and that Rmfeˊt∗ZpR^mf_{\acute{e}t*}\mathbb{Z}_p is torsion-free. It then identifies with T\mathbb{T}. Put

Ei=Rm−if∗ΩX/Bi,Ei+=Ei⊗O^B+(−i),λ=ζp−1.E^i=R^{m-i}f_*\Omega^i_{\mathfrak{X}/\mathfrak{B}},\qquad E_i^+=E^i\otimes\widehat{\mathcal{O}}_B^+(-i),\qquad\lambda=\zeta_p-1.

Here differentials on formal schemes are continuous, and Ei⊗O^B+E^i\otimes\widehat{\mathcal{O}}_B^+ denotes pullback along the specialization map Bproeˊt→BeˊtB_{\mathrm{pro\acute{e}t}}\to B_{\acute{e}t}. The generic fibre of EiE^i is EBiE_B^i. We regard Ei+E_i^+ as a submodule of gr⁡iFV\operatorname{gr}^F_iV via κi\kappa_i.

Proposition 2.3.1. In the almost category there is an O^B+\widehat{\mathcal{O}}_B^+-submodule H⊂V+\mathcal{H}\subset V^+ such that

λmV+⊂H⊂V+,gr⁡iFH=λiEi+(0≤i≤m).(10)\lambda^mV^+\subset\mathcal{H}\subset V^+,\qquad\operatorname{gr}^F_i\mathcal{H}=\lambda^iE_i^+\qquad(0\le i\le m). \tag*{(10)}

Here FiH=H∩FiVF_i\mathcal{H}=\mathcal{H}\cap F_iV, and the equality of graded pieces is taken inside gr⁡iFV\operatorname{gr}^F_iV.

Proof. We work in the almost category. To take relative pro-étale cohomology before proper push-forward, use the relative topos of [8 §2.1 and Proposition 5.7], with its maps

Xproeˊt→νfD=Bproeˊt×BeˊtXeˊt→πBproeˊt,q:D⟶Xeˊt.X_{\mathrm{pro\acute et}} \xrightarrow{\nu_f} \mathcal{D}=B_{\mathrm{pro\acute et}}\mathop{\times}_{B_{\acute et}}X_{\acute et}\xrightarrow{\pi}B_{\mathrm{pro\acute et}},\qquad q:\mathcal{D}\longrightarrow X_{\acute et}.

Here π∘νf=fproeˊt\pi\circ\nu_f=f_{\mathrm{pro\acute et}} and qq is the other projection. We equip D\mathcal{D} with the sheaf of rings O^D+=νf∗O^X+\widehat{\mathcal O}_{\mathcal D}^{+}=\nu_{f*}\widehat{\mathcal O}_{\mathcal X}^{+}, and take pullbacks to D\mathcal{D} with coefficients in this ring. Put

K+=Rνf∗O^X+,K~+=LηλK+,ΩDj=q∗ΩX/Bj.K^{+}=R\nu_{f*}\widehat{\mathcal O}_{\mathcal X}^{+},\qquad\widetilde K^{+}=L\eta_{\lambda}K^{+},\qquad\Omega_{\mathcal D}^{j}=q^{*}\Omega_{\mathcal X/B}^{j}.

The décalage functor LηλL\eta_{\lambda} is represented, on a nonnegative complex P∙P^{\bullet} of λ\lambda-torsion-free modules, by

(ηλP)j={x∈λjPj:dx∈λj+1Pj+1}.(\eta_{\lambda}P)^{j}=\{x\in\lambda^{j}P^{j}:dx\in\lambda^{j+1}P^{j+1}\}.

Since H0(K+)=νf∗O^X+H^{0}(K^{+})=\nu_{f*}\widehat{\mathcal O}_{\mathcal X}^{+} is λ\lambda-torsion-free, there is a natural map β:K~+→K+\beta:\widetilde K^{+}\to K^{+} [1 Lemma 6.10]. Put H=Hm(Rπ∗K~+)\mathcal H=H^{m}(R\pi_{*}\widetilde K^{+}), and let Hi\mathcal H_{i} be the image of Hm(Rπ∗τ≤iK~+)H^{m}(R\pi_{*}\tau_{\leq i}\widetilde K^{+}) in H\mathcal H. The relative primitive comparison isomorphism [14 Theorem 1.3 and proof of Theorem 8.8(i)] identifies Hm(Rπ∗K+)H^{m}(R\pi_{*}K^{+}) with V+V^{+}. Thus β\beta induces b:H→V+b:\mathcal H\to V^{+}, which is an isomorphism after inverting pp. We will identify the graded pieces of H\mathcal H and bound its image under bb above and below.

The relative integral Hodge–Tate comparison [8 Theorem 5.19] gives

Hj(K~+)≃λjΩDj(−j).(11)H^{j}(\widetilde K^{+})\simeq\lambda^{j}\Omega_{\mathcal D}^{j}(-j). \tag*{(11)}

Now take a formally étale B\mathcal B-morphism U→G^m,Br\mathcal U\to\widehat{\mathbb G}_{m,\mathcal B}^{r}, where U\mathcal U is an open of X\mathcal X, and let uu be the pullback of one of the toric coordinates. Let c(u)c(u) be the image in (R1νf∗O^X+)(1)(R^{1}\nu_{f*}\widehat{\mathcal O}_{\mathcal X}^{+})(1) of the Kummer class of uu on the generic fibre of U\mathcal U. Using (2.3.3), one has

H1(β)(1)(λ dlog⁡u)=λ c(u)H^{1}(\beta)(1)(\lambda\,d\log u)=\lambda\,c(u)

[1 Definition 8.2], [8 (5.34)]. Taking exterior products then determines the map on cohomology induced by β\beta in every degree. By (2.3.3) and coherent base change [8 Corollary 5.27], canonical truncation of K~+\widetilde K^{+} gives the spectral sequence

E2a,j=λjRaf∗ΩX/Bj⊗O^B+(−j)⟹Ha+j(Rπ∗K~+).(12)E_{2}^{a,j}=\lambda^{j}R^{a}f_{*}\Omega_{\mathcal X/\mathcal B}^{j}\otimes\widehat{\mathcal O}_{\mathcal B}^{+}(-j)\Longrightarrow H^{a+j}(R\pi_{*}\widetilde K^{+}). \tag*{(12)}

We next compare the filtration on H[1/p]\mathcal H[1/p] with FF. Let CX\mathcal C_{\mathcal X} be the pullback of gr⁡0OBdR,XK\operatorname{gr}^{0}\mathcal O\mathbb B_{\mathrm{dR},X_{K}} along X→XK\mathcal X\to X_{K}, with coefficients in O^X\widehat{\mathcal O}_{\mathcal X}. The graded relative Poincaré resolution [14 Proposition 8.5] gives a quasi-isomorphism

fproeˊt∗CB→∼(CX→∇‾X/BCX⊗ΩX/B1(−1)⟶⋯ ).(13)f_{\mathrm{pro\acute et}}^{*}\mathcal C_{\mathcal B}\xrightarrow{\sim}\left(\mathcal C_{\mathcal X}\xrightarrow{\overline{\nabla}_{X/B}}\mathcal C_{\mathcal X}\otimes\Omega_{\mathcal X/\mathcal B}^{1}(-1)\longrightarrow\cdots\right). \tag*{(13)}

Here ∇‾X/B\overline{\nabla}_{X/B} is the O^X\widehat{\mathcal O}_{\mathcal X}-linear Higgs field induced from the connection on OBdR,XK\mathcal O\mathbb B_{\mathrm{dR},X_{K}} by taking the associated graded. The pullback on the left has coefficients in O^X\widehat{\mathcal O}_{\mathcal X}. The proof of [14 Lemma 8.7(iv)] shows that each term CX⊗ΩX/Bj(−j)\mathcal C_{\mathcal X}\otimes\Omega_{\mathcal X/\mathcal B}^{j}(-j) is νf∗\nu_{f*}-acyclic, with direct image ΩDj[1/p]⊗π∗CB(−j)\Omega_{\mathcal D}^{j}[1/p]\otimes\pi^{*}\mathcal C_{\mathcal B}(-j). The complex obtained from (2.3.5) by applying νf∗\nu_{f*} has zero differentials: the coefficients in O^D+[1/p]\widehat{\mathcal O}_{\mathcal D}^{+}[1/p] and the pulled-back algebra π∗CB\pi^{*}\mathcal C_{\mathcal B} are killed by ∇‾X/B\overline{\nabla}_{X/B}. Applying Rνf∗R\nu_{f*} to the resolution therefore gives

K+[1/p]⊗π∗CB≃⨁j≥0ΩDj[1/p]⊗π∗CB(−j)[−j].(14)K^{+}[1/p]\otimes\pi^{*}\mathcal C_{\mathcal B}\simeq\bigoplus_{j\geq0}\Omega_{\mathcal D}^{j}[1/p]\otimes\pi^{*}\mathcal C_{\mathcal B}(-j)[-j]. \tag*{(14)}

where tensors are over O^B+[1/p]\widehat{\mathcal{O}}_{\mathcal{B}}^{+}[1/p]. The algebra CB\mathcal{C}_{\mathcal{B}} is locally polynomial over O^B\widehat{\mathcal{O}}_{\mathcal{B}}, hence faithfully flat. Thus canonical truncation through degree ii on the left of (2.3.6) corresponds to the sum with j≤ij \le i on the right.

Applying Rπ∗R\pi_* to (2.3.6) and taking HmH^m gives (2.2.1). Indeed, coherent base change identifies the summands on the right, and the relative de Rham comparison is constructed from the same Poincaré resolution [14 proof of Theorem 8.8(ii)]. The definition of FF and faithful flatness of CB\mathcal{C}_{\mathcal{B}} now give b(Hi[1/p])=FiVb(\mathcal{H}^i[1/p]) = F_iV.

The cohomology maps in (2.3.6) are compatible with (2.3.3) via the map β\beta. Indeed, after twisting by (1), the inverse degree-one map sends dlog⁡ud\log u to c(u)⊗1c(u)\otimes1, up to sign [14 Lemma 6.17]. This agrees with H1(β)(1)H^1(\beta)(1) on λΩOB1\lambda\Omega^1_{\mathcal{O}_{\mathcal{B}}} after scalar extension to π∗CB\pi^*\mathcal{C}_{\mathcal{B}}. Exterior products give agreement in every degree. In particular b[1/p]b[1/p] induces the maps κi\kappa_i, up to signs, on associated gradeds. Now (2.3.6) implies rational degeneration of (2.3.4) which then implies integral degeneration, since the E2E_2 terms are pp-torsion-free. Thus H\mathcal{H} and H/Hi\mathcal{H}/\mathcal{H}_i are pp-torsion-free. As b[1/p]b[1/p] is an isomorphism, bb embeds H\mathcal{H} in V+V^+; the equality b(Hi[1/p])=FiVb(\mathcal{H}_i[1/p]) = F_iV then gives Hi=H∩FiV\mathcal{H}_i = \mathcal{H} \cap F_iV. Under this identification, the preceding comparison gives

gr⁡iFH=λiEi+⊂gr⁡iFV.\operatorname{gr}^{F}_{i}\mathcal{H} = \lambda^i E_i^+ \subset\operatorname{gr}^{F}_{i}V.

Finally, by [1 Lemma 6.9], multiplication by λm\lambda^m on τ≤mK+\tau_{\leq m}K^+ factors through τ≤mK~+\tau_{\leq m}\widetilde{K}^+. Apply Rπ∗R\pi_* and take HmH^m; truncation above mm does not affect this cohomology. It follows that λmV+⊂H\lambda^mV^+ \subset\mathcal{H}.

The estimate on a disc

Keep the hypotheses of Theorem B, and let C′/CC'/C be a complete algebraically closed extension. Put D=Spf⁡OC′⟨z⟩\mathbb{D} = \operatorname{Spf}\mathcal{O}_{C'}\langle z\rangle, and write D\mathbb{D} for its rigid generic fibre. For an OK\mathcal{O}_K-morphism γ:D→B^\gamma:\mathbb{D}\to\widehat{B}, let γη:D→BKan\gamma_\eta:\mathbb{D}\to B_K^{\mathrm{an}} be its generic fibre. The pullback family fγ:X×B^D→Df_\gamma:\mathcal{X}\times_{\widehat{B}}\mathbb{D}\to\mathbb{D} has Hodge bundles Eγi=γ∗EiE^i_\gamma=\gamma^*E^i and local system Tγ=γη∗TT_\gamma=\gamma_\eta^*T. Write

θγ,i:Eγi⟶Eγi−1⊗ΩD/OC′1\theta_{\gamma,i}:E^i_\gamma\longrightarrow E^{i-1}_\gamma\otimes\Omega^1_{\mathbb{D}/\mathcal{O}_{C'}}

for its Higgs field, and put θγ=⨁iθγ,i\theta_\gamma=\bigoplus_i\theta_{\gamma,i}.

Proposition 2.4.1. If Tγ/pnT_\gamma/p^n is constant on D\mathbb{D} and n>εn>\varepsilon, then

v(θγ(∂z)∣0)≥n−(m+1)ε.v\bigl(\theta_\gamma(\partial_z)|_0\bigr)\ge n-(m+1)\varepsilon.

Proof. Put Vγ+=Tγ⊗O^D+V^+_\gamma=T_\gamma\otimes\widehat{\mathcal{O}}_{\mathbb{D}}^+. By Proposition 2.1.6, the contraction S=STγ(∂z)S=S_{T_\gamma}(\partial_z) sends Vγ+V^+_\gamma into pnVγ+(−1)p^nV^+_\gamma(-1). Let Hγ⊂Vγ+\mathcal{H}_\gamma\subset V_\gamma^+ denote the lattice H\mathcal{H} of Proposition 2.3.1, applied to the pullback family fγf_\gamma over OC′\mathcal{O}_{C'}. Then

p−nλmS(Hγ)⊂λmVγ+(−1)⊂Hγ(−1).p^{-n}\lambda^mS(\mathcal{H}_\gamma)\subset\lambda^mV^+_\gamma(-1)\subset\mathcal{H}_\gamma(-1).

By Lemma 2.2.3, this operator lowers the filtration on Hγ\mathcal{H}_\gamma. The factors λi\lambda^i and λi−1\lambda^{i-1} in the source and target graded pieces give, after cancelling the Tate twists, an almost integral map

p−nλm+1θγ,i(∂z):Eγi⊗O^D+⟶Eγi−1⊗O^D+(1≤i≤m).p^{-n}\lambda^{m+1}\theta_{\gamma,i}(\partial_z):E^i_\gamma\otimes\widehat{\mathcal{O}}_{\mathbb{D}}^+\longrightarrow E^{i-1}_\gamma\otimes\widehat{\mathcal{O}}_{\mathbb{D}}^+ \qquad(1\le i\le m).

Evaluation at z=0z=0 gives the asserted bound, since v(λ)=εv(\lambda)=\varepsilon. Almost inclusions suffice for this valuation inequality, as observed in § 3.1.

Compressions and bounded derivations

We recall the field-theoretic definition of essential dimension at pp. For a finite étale FF-scheme YY, where F/CF/C is a finitely generated field, define ed⁡C,p(Y)\operatorname{ed}_{C,p}(Y) to be the least trdeg⁡CL\operatorname{trdeg}_C L for which there exist a finite extension F′/FF'/F of degree prime to pp, a finitely generated subfield C⊂L⊂F′C \subset L \subset F', and a finite étale LL-scheme YLY_L such that YF′≃(YL)F′Y_{F'} \simeq(Y_L)_{F'}. For a finite local system on an integral variety, we apply this definition to its étale space at the generic point.

Valuations and bounded derivations

Let CC be a complete algebraically closed extension of Qp\mathbb{Q}_p, and let F/CF/C be a finitely generated extension of transcendence degree bb. Equip FF with a valuation vv extending that of CC, normalized by v(p)=1v(p)=1, with v(F×)=v(C×)=Qv(F^\times)=v(C^\times)=\mathbb{Q}. Write F^\widehat{F} for its completion. Choose a basis ∂1,…,∂b\partial_1,\ldots,\partial_b of Der⁡C(F)\operatorname{Der}_C(F) such that

v(∂j(a))≥v(a)(a∈F).v(\partial_j(a)) \ge v(a) \qquad(a \in F).

These derivations extend continuously to F^\widehat{F} and preserve OF^\mathcal{O}_{\widehat{F}}. Put

D+=⨁j=1bOF^∂j,D=D+[1/p].\mathcal{D}^{+}=\bigoplus_{j=1}^{b}\mathcal{O}_{\widehat{F}}\partial_j,\qquad\mathcal{D}=\mathcal{D}^{+}[1/p].

For a finite extension F′/FF'/F, write F′^\widehat{F'} for its completion at a chosen extension of the valuation.

Lemma 3.1.1. If F′^/F^\widehat{F'}/\widehat{F} is finite of degree prime to pp, then OF′^/OF^\mathcal{O}_{\widehat{F'}}/\mathcal{O}_{\widehat{F}} is finite étale. Each ∂j\partial_j extends uniquely to a derivation of F′^\widehat{F'} preserving OF′^\mathcal{O}_{\widehat{F'}}.

Proof. The value group of F^\widehat{F} is divisible, so the ramification index is one. By Ostrowski’s lemma [11 Chapter 11, Lemma 11.17], the defect is a power of pp, as is the inseparable residue degree. Since [F′^:F^][\widehat{F'}:\widehat{F}] is prime to pp, the residue extension is separable of degree [F′^:F^][\widehat{F'}:\widehat{F}].

By [15 Tag 04GK], the residue extension lifts to a finite étale OF^\mathcal{O}_{\widehat{F}}-algebra O′\mathcal{O}'. Henselian lifting embeds O′\mathcal{O}' in OF′^\mathcal{O}_{\widehat{F'}}, and equality of degrees gives Frac⁡O′=F′^\operatorname{Frac}\mathcal{O}'=\widehat{F'}. Since O′\mathcal{O}' is normal, O′=OF′^\mathcal{O}'=\mathcal{O}_{\widehat{F'}}. The assertion about derivations follows from formal étaleness; see [15 Tag 00UP].

For a finite free OF^\mathcal{O}_{\widehat{F}}-module Λ\Lambda, let vΛv_\Lambda denote the minimum of the coordinate valuations in any OF^\mathcal{O}_{\widehat{F}}-basis. We use this notation on Λ[1/p]\Lambda[1/p] and after any valued extension of F^\widehat{F}. Let θ:D+→Λ\theta:\mathcal{D}^{+}\to\Lambda be an OF^\mathcal{O}_{\widehat{F}}-linear map whose reduction has rank d0d_0. Fix F′/FF'/F and an extension of the valuation for which [F′^:F^][\widehat{F'}:\widehat{F}] is prime to pp.

Lemma 3.1.2. Let C⊂L⊂F′^C\subset L\subset\widehat{F'} be a finitely generated field with trdeg⁡CL=r<d0\operatorname{trdeg}_C L=r<d_0. There is D∈D+⊗OF^OF′^D\in\mathcal{D}^{+}\otimes_{\mathcal{O}_{\widehat{F}}}\mathcal{O}_{\widehat{F'}} such that D∣L=0D|_L=0 and vΛ(θ(D))=0v_\Lambda(\theta(D))=0.

Proof. Extend scalars to OF′^\mathcal{O}_{\widehat{F'}}, retaining the notation D\mathcal{D}, D+\mathcal{D}^{+}, Λ\Lambda, and θ\theta. Choose a transcendence basis x1,…,xrx_1,\ldots,x_r of L/CL/C. The equations D(xj)=0D(x_j)=0 impose at most rr linear conditions on DD. In characteristic zero they also imply D∣L=0D|_L=0. Their solution space WW has dimension at least b−rb-r.

The quotient D+/(W∩D+)\mathcal{D}^{+}/(W\cap\mathcal{D}^{+}) is finite and torsion-free over the valuation ring OF′^\mathcal{O}_{\widehat{F'}}, hence free by [15 Tags 0539 and 00NZ]. Thus W∩D+W\cap\mathcal{D}^{+} is a direct summand of D+\mathcal{D}^{+} of rank dim⁡W\dim W.

The reduction of W∩D+W\cap\mathcal{D}^{+} has dimension at least b−r>b−d0b-r>b-d_0. It cannot be contained in the kernel of the reduction of θ\theta, which has dimension b−d0b-d_0. Choose a vector in this reduction outside the kernel, and lift it to D∈W∩D+D\in W\cap\mathcal{D}^{+}. This derivation has the required properties.

The exponential disc

Lemma 3.2.1. Let DD be a CC-linear derivation of F^′\widehat{F}^{\prime} preserving OF^′\mathcal{O}_{\widehat{F}^{\prime}}, and let c∈Cc \in C satisfy v(c)>εv(c)>\varepsilon. For an indeterminate zz, the map

a⟼exp⁡(czD)(a)=∑j≥0cjDj(a)j!zja \longmapsto\exp(czD)(a)=\sum_{j\geq0}\frac{c^{j}D^{j}(a)}{j!}z^{j}

is a CC-algebra embedding F^′↪F^′⟨z⟩\widehat{F}^{\prime}\hookrightarrow\widehat{F}^{\prime}\langle z\rangle. It maps OF^′\mathcal{O}_{\widehat{F}^{\prime}} into OF^′⟨z⟩\mathcal{O}_{\widehat{F}^{\prime}\langle z\rangle}, and it fixes every element annihilated by DD.

Proof. Since v(F^′×)=v(C×)v(\widehat{F}^{\prime\times})=v(C^{\times}), dividing a≠0a\ne0 by a constant of the same valuation shows that v(D(a))≥v(a)v(D(a))\geq v(a). This also holds for DjD^{j}. Hence

v(cjDj(a)j!a)≥j(v(c)−ε).(15)v\left(\frac{c^{j}D^{j}(a)}{j!a}\right)\geq j\bigl(v(c)-\varepsilon\bigr). \tag*{(15)}

using vp(j!)≤j/(p−1)v_{p}(j!)\leq j/(p-1). The series converges in F^′⟨z⟩\widehat{F}^{\prime}\langle z\rangle, and the Leibniz rule shows that exp⁡(czD)\exp(czD) is a CC-algebra homomorphism. Evaluation at z=0z=0 is a left inverse to exp⁡(czD)\exp(czD), so exp⁡(czD)\exp(czD) is injective. By (3.2.2), exp⁡(czD)\exp(czD) maps OF^′\mathcal{O}_{\widehat{F}^{\prime}} into OF^′⟨z⟩\mathcal{O}_{\widehat{F}^{\prime}\langle z\rangle}. If D(a)=0D(a)=0, all positive-degree terms in the defining series vanish, so exp⁡(czD)(a)=a\exp(czD)(a)=a.

Proof of Theorem B. After replacing B\mathcal{B} by an affine open meeting the generic point of its special fibre, write B=Spec⁡A\mathcal{B}=\operatorname{Spec} A and choose étale coordinates t1,…,tbt_{1},\ldots,t_{b} over OK\mathcal{O}_{K}, where b=dim⁡BKb=\dim B_{K}. We may also trivialize the Hodge bundles and arrange that the reduction of (1.1.1) has rank d0d_{0} throughout this open.

Take F=κ(BC)F=\kappa(\mathcal{B}_{C}) with the Gauss valuation defined by the generic point η\eta of Bk\mathcal{B}_{k}. It extends the valuation of CC, has value group Q\mathbb{Q}, and has residue field κ(Bk‾)\kappa(\mathcal{B}_{\overline{k}}). The coordinate derivations ∂j=∂/∂tj\partial_{j}=\partial/\partial t_{j} preserve A⊗OKOCA\otimes_{\mathcal{O}_{K}}\mathcal{O}_{C} and its localization at η\eta. For a∈Fa\in F, normalizing its numerator and denominator by constants of CC and applying the quotient rule gives v(∂j(a))≥v(a)v(\partial_{j}(a))\geq v(a). The reductions of the ∂j\partial_{j} form a basis of TBk‾/k,ηT_{\mathcal{B}_{\overline{k}}/k,\eta}. Put

Λ=⨁iHom⁡OF^(Ei⊗AOF^,Ei−1⊗AOF^).\Lambda=\bigoplus_{i}\operatorname{Hom}_{\mathcal{O}_{\widehat{F}}}\left(E^{i}\otimes_{A}\mathcal{O}_{\widehat{F}},E^{i-1}\otimes_{A}\mathcal{O}_{\widehat{F}}\right).

The Higgs contraction induces θ:D+→Λ\theta:D^{+}\to\Lambda whose reduction has rank d0d_{0}.

Suppose that T/pT/p admits a compression of dimension r<d0r<d_{0}. Thus there is an extension F′/FF^{\prime}/F of degree prime to pp and a subfield C⊂L⊂F′C\subset L\subset F^{\prime}, of transcendence degree rr, over which the finite étale space of T/pT/p descends to a finite étale LL-scheme YLY_{L}. The algebra F′⊗FF^F^{\prime}\otimes_{F}\widehat{F} is a product of finite extensions of F^\widehat{F}. Since the sum of their degrees is prime to pp, at least one factor has degree prime to pp. Choose the corresponding valuation on F′F^{\prime}; its completion F^′\widehat{F}^{\prime} is this factor.

Choose the bounded derivation DD of Lemma 3.1.2. Since p>m+3p>m+3, we may choose c∈Cc\in C with

ε<v(c)<1−(m+1)ε.\varepsilon<v(c)<1-(m+1)\varepsilon.

Let C′C^{\prime} be a completed algebraic closure of F^′\widehat{F}^{\prime} and put γ∗=exp⁡(czD):F^′→F^′⟨z⟩\gamma^{*}=\exp(czD):\widehat{F}^{\prime}\to\widehat{F}^{\prime}\langle z\rangle. The restriction γ∗∣A\gamma^{*}|_{A} takes values in OC′⟨z⟩\mathcal{O}_{C^{\prime}\langle z\rangle}, and hence gives a formal map

γ:Spf⁡OC′⟨z⟩⟶B^OC′.\gamma:\operatorname{Spf}\mathcal{O}_{C^{\prime}\langle z\rangle}\longrightarrow\widehat{\mathcal{B}}_{\mathcal{O}_{C^{\prime}}}.

Here B^C′\widehat{\mathcal{B}}_{C'} is the pp-adic completion of B⊗OKOC′\mathcal{B}\otimes_{\mathcal{O}_K}\mathcal{O}_{C'}, and (dγ)∣0(∂z)=cD(d\gamma)|_0(\partial_z)=cD. Write γη:D→BC′\gamma_\eta:\mathbb{D}\to\mathcal{B}_{C'} for the generic fibre, where D=Spa⁡(C′⟨z⟩,OC′⟨z⟩)\mathbb{D}=\operatorname{Spa}(C'\langle z\rangle,\mathcal{O}_{C'}\langle z\rangle).

By the definition of the compression, the étale space of γη∗(T/p)\gamma_\eta^*(\mathbb{T}/p) is identified with YL×LDY_L\times_L\mathbb{D}. Since D∣L=0D|_L=0, the map γ∗∣L:L→C′⟨z⟩\gamma^*|_L:L\to C'\langle z\rangle factors through the algebraically closed field C′C', so this étale space is constant. Its module operations are constant as well, since D\mathbb{D} is connected.

Proposition 2.4.1 now gives

v(θγ(∂z)∣0)≥1−(m+1)ε.v(\theta_\gamma(\partial_z)|_0)\ge1-(m+1)\varepsilon.

But this matrix is cθ(D)c\theta(D) and θ(D)\theta(D) has a unit entry. Its valuation is v(c)<1−(m+1)εv(c)<1-(m+1)\varepsilon, a contradiction.

The estimate for an arbitrary prime

We now remove the restriction on pp and the hypothesis of good reduction by passing to sufficiently high pp-power levels. The rational Hodge–Tate comparison at a fixed analytic point gives a valuation estimate with a constant independent of the level and the compression; increasing the exponent absorbs this constant.

Let K/QpK/\mathbb{Q}_p be finite, put C=K^C=\widehat{K}, and let f:X→Bf:X\to B be smooth projective, with B/KB/K smooth and geometrically integral. Put

T=(Rmfeˊt∗Zp)/tors⁡,EKi=Rm−if∗ΩX/Bi.\mathbb{T}=(R^m f_{\acute{e}t*}\mathbb{Z}_p)/\operatorname{tors},\qquad E_K^i=R^{m-i}f_*\Omega^i_{X/B}.

The Gauss–Manin connection on Rmf∗ΩX/B∙R^m f_*\Omega^\bullet_{X/B} induces a Higgs contraction

θ:TB/K⟶⨁iHom⁡(EKi,EKi−1).\theta:T_{B/K}\longrightarrow\bigoplus_i\operatorname{Hom}(E_K^i,E_K^{i-1}).

Let dd be its generic rank.

Theorem 3.3.1. There is an integer n0≥1n_0\ge1 such that

ed⁡C,p(T/pn∣BC)≥d(n≥n0).\operatorname{ed}_{C,p}(\mathbb{T}/p^n|_{B_C})\ge d\qquad(n\ge n_0).

Proof. We may assume d>0d>0. After a finite extension of KK, choose a closed analytic polydisc UK⊂BanU_K\subset B^{\mathrm{an}} in the rank-dd locus of θ\theta, with coordinates t1,…,tbt_1,\ldots,t_b obtained by rescaling algebraic étale coordinates, where b=dim⁡Bb=\dim B. Put U=(UK)CU=(U_K)_C and A=C⟨t1,…,tb⟩A=C\langle t_1,\ldots,t_b\rangle. Specialize § 4.1 to F=κ(BC)F=\kappa(B_C) with the Gauss valuation induced by AA and ∂j=∂/∂tj\partial_j=\partial/\partial t_j. Since the subfield C(t1,…,tb)⊂FC(t_1,\ldots,t_b)\subset F is dense in Frac⁡A\operatorname{Frac}A, the inclusion A↪F^A\hookrightarrow\widehat{F} defines the Gauss point η:Spa⁡(F^,OF^)→U\eta:\operatorname{Spa}(\widehat{F},\mathcal{O}_{\widehat{F}})\to U.

Fix a completed algebraic closure C′C' of F^\widehat{F}, and write ηˉ∈UC′(C′)\bar{\eta}\in U_{C'}(C') for the extension of η\eta to C′C'. Put (EKi)F^=η∗EKi(E_K^i)_{\widehat{F}}=\eta^*E_K^i. The Higgs contraction at η\eta is

θ:D⟶GF^:=⨁iHom⁡F^((EKi)F^,(EKi−1)F^).\theta:\mathcal{D}\longrightarrow G_{\widehat{F}}:=\bigoplus_i\operatorname{Hom}_{\widehat{F}}\bigl((E_K^i)_{\widehat{F}},(E_K^{i-1})_{\widehat{F}}\bigr).

The image θ(D+)\theta(\mathcal{D}^+) is a finitely generated torsion-free module over the valuation ring OF^\mathcal{O}_{\widehat{F}}, hence free of rank dd by [15 Tags 0539 and 00NZ]. To preserve this rank on reduction, extend an OF^\mathcal{O}_{\widehat{F}}-basis of θ(D+)\theta(\mathcal{D}^+) to an F^\widehat{F}-basis of GF^G_{\widehat{F}}, and let Λ⊂GF^\Lambda\subset G_{\widehat{F}} be its OF^\mathcal{O}_{\widehat{F}}-span. Then θ(D+)\theta(\mathcal{D}^+) is a direct summand of Λ\Lambda; in particular, the reduction of θ:D+→Λ\theta:\mathcal{D}^+\to\Lambda has rank dd.

We now replace Proposition 2.4.1 by an estimate with a constant depending on ηˉ\bar{\eta} and Λ\Lambda. Put

V=Tηˉ⊗ZpC′,V+=Tηˉ⊗ZpOC′.V=\mathbb{T}_{\bar{\eta}}\otimes_{\mathbb{Z}_p}C',\qquad V^+=\mathbb{T}_{\bar{\eta}}\otimes_{\mathbb{Z}_p}\mathcal{O}_{C'}.

Pulling (2.2.2) back to ηˉ\bar{\eta} gives a filtration FF on VV. The comparison maps κi\kappa_i of § 3.2 identify gr⁡iFV\operatorname{gr}^{F}_{i}V with (EKi)ηˉ(−i)(E^{i}_{K})_{\bar{\eta}}(-i). Let M⊂Hom⁡C′(V,V(−1))M \subset\operatorname{Hom}_{C'}(V,V(-1)) consist of the maps uu with u(FiV)⊂Fi−1V(−1)u(F_iV) \subset F_{i-1}V(-1) for every ii. Using these identifications, taking associated graded and cancelling the Tate twists defines

ρ:M⟶GF^⊗F^C′,u⟼(gr⁡iu)i.\rho:M \longrightarrow G_{\widehat{F}}\otimes_{\widehat{F}}C',\qquad u\longmapsto(\operatorname{gr}_{i}u)_{i}.

Since ρ\rho extends to a C′C'-linear map on Hom⁡C′(V,V(−1))\operatorname{Hom}_{C'}(V,V(-1)), there is an integer a≥0a\geq0 such that

paρ(M∩Hom⁡OC′(V+,V+(−1)))⊂Λ⊗OF^OC′.p^{a}\rho\left(M\cap\operatorname{Hom}_{\mathcal{O}_{C'}}(V^{+},V^{+}(-1))\right)\subset\Lambda\otimes_{\mathcal{O}_{\widehat{F}}}\mathcal{O}_{C'}.

Put D=Spa⁡(C′⟨z⟩,OC′⟨z⟩)\mathbb{D}=\operatorname{Spa}(C'\langle z\rangle,\mathcal{O}_{C'}\langle z\rangle). For a map γ:D→UC′\gamma:\mathbb{D}\to U_{C'} with γ(0)=ηˉ\gamma(0)=\bar{\eta}, write θγ=γ∗θ∘dγ\theta_{\gamma}=\gamma^{*}\theta\circ\mathrm{d}\gamma. Lemmas 2.2.3 and 2.1.7 give

ρ(Sγ∗π(∂z)∣0)=θγ(∂z)∣0.\rho(S_{\gamma}^{*}\pi(\partial_z)|_0)=\theta_{\gamma}(\partial_z)|_0.

Proposition 2.1.6 therefore gives

vΛ(θγ(∂z)∣0)≥n−a(16)v_{\Lambda}\left(\theta_{\gamma}(\partial_z)|_0\right)\geq n-a \tag*{(16)}

whenever γ∗(T/pn)\gamma^{*}(\mathbb{T}/p^{n}) is constant and n>εn>\varepsilon. The constant aa is independent of γ\gamma and of the compression.

Now a compression of dimension r<dr<d gives a finite extension F′/FF'/F of degree prime to pp and a subfield C⊂L⊂F′C\subset L\subset F' with trdeg⁡CL=r\operatorname{trdeg}_{C}L=r. As in the proof of Theorem B, choose a completion F^′\widehat{F}' with [F^′:F^][\widehat{F}':\widehat{F}] prime to pp, and embed F^′\widehat{F}' in the fixed C′C'. Lemma 3.1.2, with d0=dd_{0}=d, gives D∈D+⊗OF^OF^′D\in D^{+}\otimes_{\mathcal{O}_{\widehat{F}}}\mathcal{O}_{\widehat{F}'} such that D∣L=0D|_{L}=0 and vΛ(θ(D))=0v_{\Lambda}(\theta(D))=0. For n>a+εn>a+\varepsilon, choose c∈Cc\in C with ε<v(c)<n−a\varepsilon<v(c)<n-a. By Lemma 3.2.1, the restriction of γ∗=exp⁡(czD)\gamma^{*}=\exp(czD) to AA defines γ:D→UC′\gamma:\mathbb{D}\to U_{C'} with γ(0)=ηˉ\gamma(0)=\bar{\eta} and (dγ)∣0(∂z)=cD(\mathrm{d}\gamma)|_{0}(\partial_z)=cD. Since D∣L=0D|_{L}=0, the argument in the proof of Theorem B shows that γ∗(T/pn)\gamma^{*}(\mathbb{T}/p^{n}) is constant. But (3.3.2) would give

n−a≤vΛ(θγ(∂z)∣0)=v(c)<n−a.n-a\leq v_{\Lambda}\left(\theta_{\gamma}(\partial_z)|_0\right)=v(c)<n-a.

This contradiction proves the assertion for any integer n0>a+εn_{0}>a+\varepsilon.

Spreading out

Proof of Theorem A. The Higgs contraction (1.1.1) has generic rank d=dim⁡Φ(B)d=\dim\Phi(B). After spreading out in characteristic zero, choose a specialization fE:XE→BEf_{E}:X_{E}\to B_{E} over a number field E⊂CE\subset\mathbb{C} whose Higgs contraction still has generic rank dd. By [12 Theorem 10.2], applied to the frame torsors of H/pn\mathbb{H}/p^{n}, and [6 Lemma 2.2.4], essential dimension at pp cannot increase under this specialization. We may therefore replace ff by fE×ECf_{E}\times_{E}\mathbb{C}, retaining the notation ff and H\mathbb{H}.

For each prime pp, choose a place v∣pv\mid p of EE and put K=EvK=E_{v}, C=K^C=\widehat{K}, fK=fE×EKf_{K}=f_{E}\times_{E}K, and T=(RmfK,eˊt∗Zp)/tors⁡\mathbb{T}=(R^{m}f_{K,\acute{e}t*}\mathbb{Z}_{p})/\operatorname{tors}. By the étale–Betti comparison isomorphism and invariance of essential dimension under extension of algebraically closed base fields [6 Lemma 2.2.5 and §2.2.6],

ed⁡p(H/pn)=ed⁡C,p(T/pn),n≥1.(17)\operatorname{ed}_{p}(\mathbb{H}/p^{n})=\operatorname{ed}_{C,p}(\mathbb{T}/p^{n}),\qquad n\geq1. \tag*{(17)}

After shrinking BEB_{E} and inverting an integer M≥1M\geq1, spread out fEf_{E} to a smooth projective morphism X→BX\to B, where BB is smooth over OE[1/M]\mathcal{O}_{E}[1/M] with geometrically integral fibres. We may arrange that the Hodge and de Rham cohomology sheaves of X/BX/B and the Hodge filtration are locally free and commute with base change [15 Tags of FM10 and OB8W], and that the Higgs contraction has rank dd on every fibre. Enlarge MM to exclude the primes dividing the torsion in Rmf∗ZR^{m}f_{*}\mathbb{Z}. For p∤Mp\nmid M, the base change to OK\mathcal{O}_{K} then satisfies the hypotheses of Theorem B, with special-fibre Higgs rank dd. Together with (3.4.1), this gives

ed⁡p(H/p)≥d(p>p0)\operatorname{ed}_{p}(\mathbb{H}/p)\geq d\qquad(p>p_{0})

for some p0≥2p_{0} \ge2. By [6 Lemma 2.1.6], ed⁡p(H/pn)≥ed⁡p(H/p)\operatorname{ed}_{p}(\mathbb{H}/p^{n}) \ge\operatorname{ed}_{p}(\mathbb{H}/p), so the same lower bound holds for every n≥1n \ge1 when p>p0p > p_{0}.

For each prime p≤p0p \le p_{0}, Theorem 3.3.1 and (3.4.1) give the same bound for n≥n0(p)n \ge n_{0}(p), for some integer n0(p)≥1n_{0}(p) \ge1. This proves the theorem.

References

  1. [1]B. Bhatt, M. Morrow, and P. Scholze, Integral p-adic Hodge theory, Publ. Math. Inst. Hautes Études Sci. 128 (2018), 219–397. arXiv:1602.03148.arxiv.org/abs/1602.03148
  2. [2]B. Bhatt and P. Scholze, Prisms and prismatic cohomology, Ann. of Math. (2) 196 (2022), no. 3, 1135–1275. doi:10.4007/annals.2022.196.3.5.
  3. [3]P. Brosnan and N. Fakhruddin, Fixed points, local monodromy, and incompressibility of congruence covers, arXiv:2007.01752v2 (2020).arxiv.org/abs/2007.01752
  4. [4]N. Fakhruddin and R. Saini, Finite group scheme actions and incompressibility of Galois covers: beyond the ordinary case, Doc. Math. 27 (2022), 151–182. doi:10.4171/DM/868.DOI
  5. [5]G. Faltings, Integral crystalline cohomology over very ramified valuation rings, J. Amer. Math. Soc. 12 (1999), no. 1, 117–144. doi:10.1090/S0894-0347-99-00273-8.DOI
  6. [6]B. Farb, M. Kisin, and J. Wolfson, The essential dimension of congruence covers, Compos. Math. 157 (2021), 2407–2432. arXiv:1901.09013v2.DOI
  7. [7]B. Farb, M. Kisin, and J. Wolfson, Essential dimension via prismatic cohomology, Duke Math. J. 173 (2024), 3059–3106. arXiv:2110.05534.DOI
  8. [8]I. Gaisin and T. Koshikawa, Relative Ainf-cohomology, arXiv:2206.07983v1 (2022).
  9. [9]S. Howe, The geometric Sen morphism is the unique lift of the Kodaira–Spencer morphism, arXiv:2512.09669v1 (2025).arxiv.org/abs/2512.09669
  10. [10]M. Kisin, Essential dimension of principal congruence covers of Shimura varieties, manuscript, 2026.
  11. [11]F.-V. Kuhlmann, Valuation theory, book manuscript, Chapter 11, Defect. https://www.fvk" + "uhlmann.de/bookch11.pdf.
  12. [12]Z. Reichstein and F. Scavia, The behavior of essential dimension under specialization, II, Algebra Number Theory 17 (2023), no. 11, 1925–1958. doi:10.2140/ant.2023.17.1925.DOI
  13. [13]J. E. Rodríguez Camargo, Geometric Sen theory over rigid analytic spaces, arXiv:2205.02016v4.arxiv.org/abs/2205.02016
  14. [14]P. Scholze, p-adic Hodge theory for rigid-analytic varieties, Forum Math. Pi 1 (2013), e1. arXiv:1205.3463v2.arxiv.org/abs/1205.3463
  15. [15]The Stacks Project Authors, The Stacks Project. https://stacks.math.columbia.edu.
  16. [16]Y. Wang, A p-adic Simpson correspondence for rigid analytic varieties, Algebra Number Theory 17 (2023), 1453–1499. arXiv:2101.00263v4.DOI

Paper details

vlnc:W_3hevpsa4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

GPT Astra

View statement
Looking for related papers…
Contents