Skip to content

Section 10 · PDF pages 756–758

Geometric inputs and the integral Hodge–Tate comparison

The notation transcription is provisional. Check it against the printed mathematics when citing or reusing a formula.

10 Geometric inputs and the integral Hodge–Tate comparison

The geometric assertions entering the covolume estimate have different mathematical contents. We first resolve the integral annihilator assertion. Its conclusion concerns integral kernels, although the Hodge–Tate exact sequence used to prove it is an exact sequence over a field. The passage between them requires torsion freeness and clearing denominators.

Lemma 10.1 (The valuation-ring hypotheses give the usual complete field). Let C be an algebraically closed field of characteristic zero, let v : C →ℝ_{≥0} be a multiplicative nonarchimedean valuation, and let O be its ring of integers. Suppose 0 < v(p) < 1 and O is separated and complete for the powers of p. Then C = O[1/p] is complete for v, and the given ℤ_{p}-algebra structure on O is the usual one. After rescaling the valuation by a positive real power, C is a complete nonarchimedean extension of ℚ_{p}.

Proof. Regard O as a subring of C. This is legitimate for the source’s abstract ring O: saying that its algebra map identifies it with the integers of v includes injectivity and the condition that its image consists of the elements of valuation at most one. For any z ∈ C, choose m so that v(p)^{m}v(z) ≤ 1. Then p^{m}z ∈ O, proving C = O[1/p]. Moreover

Selectable PDF notation • transcription unchecked
p^{m}O = {z ∈C : v(z) ≤v(p)^{m}}.

Both inclusions follow by dividing by p^{m}. These sets form a neighborhood basis of zero in O, so its valuation topology and its p-adic topology agree.

A Cauchy sequence in C is bounded. Indeed, its sufficiently late terms differ from one fixed term by elements of valuation at most one, and the finitely many earlier terms have a finite maximum valuation. Multiplication by one power of p therefore places the entire sequence in O. It is Cauchy for the powers of p by the preceding equality. Completeness of O gives a limit there; dividing by the fixed power gives a limit in C. This proves completeness.

For a ∈ ℤ_{p} and every m, write a = a_{m} + p^{m}b_{m} with a_{m} an integer and b_{m} ∈ ℤ_{p}. Any unital ring map ℤ_{p} →O sends b_{m} into O, so the image of a differs from a_{m} by an element of p^{m}O. Separation makes

this image uniquely determined. Thus the given algebra structure is the continuous extension of the integer map. It extends to the usual embedding of ℚ_{p} into C. Normalize v by its positive real power for which v(p) = p^{−1}. If an integer is prime to p, a Bezout relation with p shows that its valuation is one: if both terms of that relation had valuation less than one, their sum could not be one. The normalized restriction to ℚ is therefore the usual p-adic absolute value, and completeness gives the assertion about ℚ_{p}.

Lemma 10.2 (The integral factorization used in the source). Let O be a domain, f : M →W an O-linear map and ϕ : M →O a linear functional. Suppose ker f ⊆ ker ϕ and κW ⊆ im f. Then there is an O-linear ℓ: W →O such that

Selectable PDF notation • transcription unchecked
ℓ(f(y)) = κϕ(y) (y ∈M).

Proof. On im f define g(f(y)) = ϕ(y). If f(y) = f(y^{′}), then y −y^{′}∈ ker f, so ϕ(y) = ϕ(y^{′}). This proves that g is well defined; the same formula proves additivity and O-linearity. By the hypothesis on κ, the map w →→κw takes W into im f. Compose it with g. For w = f(y) the result is g(κf(y)) = g(f(κy)) = κϕ(y), which proves the formula.

This is the quotient-by-the-kernel construction of FarguesExactOfPrismatic.lean, lines 18–34. It does not require extending a functional arbitrarily from a submodule.

Lemma 10.3 (Clearing the denominators of a surjective map). Let F be the fraction field of a domain O, let W be finite free over O, and suppose f : M →W becomes surjective over F. Then some 0 ̸= κ ∈O satisfies κW ⊆ im f. If F = O[1/p], one may take κ = p^{a} for an integer a ≥0.

Proof. Choose a finite basis w_{1}, . . . , w_{d} of W. For each i, choose a preimage of w_{i} under f_{F} . An element of F ⊗_{O} M is a finite sum of fractions times elements of M, so multiplying by a common nonzero denominator gives m_{i} ∈M and s_{i} ̸= 0 with f(m_{i}) = s_{i}w_{i}. This equality follows first after localization and then in W, because a free module over a domain injects into its localization. Take∏ κ = ∏_{i}s_{i}. Multiplying m_{i} by _{j̸=i}s_{j} gives a preimage of κw_{i}. Linear combinations give the assertion for all W. When the localization is O[1/p], all the chosen fractions have denominators powers of p; taking the largest required power, and increasing it to account for each chosen preimage, gives the final assertion.

Background fact 10.4 (The precisely used Hodge–Tate theorem). Let C/ℚ_{p} be an algebraically closed complete nonarchimedean field and let G be a p-divisible group over O_{C}. Its Tate module is finite free over ℤ_{p}, and its completed invariant cotangent module ω_{G} is finite locally free over O_{C}. Write Lie G = ω_{G}^{∨}. The freeness of this completed module is verified again for the project’s finite-level model below. The Hodge– Tate sequence is

Selectable PDF notation • transcription unchecked
α^{∗}_{∨α}

0 −→ Lie G ⊗_{OC} C(1)−−−→^{G} T_{p}G ⊗_{ℤp} C−−→^{G} ω_{G}∨⊗_{OC} C −→0.

The two arrows are dual under the Cartier Tate pairing. The exact sequence and its arrows are [52, Theo-rem 12.1.1 and the description immediately after Remark 12.1.2, pp. 98–99]. The cotangent and Lie modules here are those of the p-adic formal group, not those obtained by taking the tangent sheaf of the generic-fiber torsion sheaf.

Proposition 10.5 (Integral annihilator equality). Let G and C satisfy Fact 10.4. Choose a generator ζ of ℤ_{p}(1) and use it to write the Cartier pairing as a perfect pairing

Selectable PDF notation • transcription unchecked
B : M × N −→O_{C}, M = O_{C} ⊗_{ℤp} T_{p}G, N = O_{C} ⊗_{ℤp} T_{p}G^{∨}.

Let f = α_{G} : M → ω_{G}∨ and g = α_{G}∨ : N → ω_{G} be the integral Hodge–Tate maps. Then, for every

Selectable PDF notation • transcription unchecked
x ∈M,

f(x) = 0 ⇐⇒ B(x, y) = 0 for every y ∈ ker g. (G.1)

Equivalently, f(x) = 0 if and only if there are 0 ̸= c ∈O_{C} and ℓ: ω_{G} →O_{C} such that cB(x, y) = ℓ(g(y)) for every y ∈N.

Proof. We first work over C. A finite-dimensional linear map g_{C} : N_{C} →(ω_{G})_{C} has

im(g_{C}^{∗}) = (ker g_{C})^{⊥}

in N_{C}^{∗} = M_{C}, using B_{C} for the identification. One inclusion follows by evaluation on a kernel vector. For the other, a functional vanishing on ker g_{C} factors through its image: send g_{C}(y) to its value on y. This is well defined by the vanishing. Extend a basis of the image to a basis of the target and extend that functional linearly, giving a preimage under g_{C}^{∗}. By Fact 10.4, its image is ker f_{C}. Hence

ker f_{C} = (ker g_{C})^{⊥}. (G.2)

Every element of ker g_{C} is a fraction of an integral kernel element. To check this explicitly, write it as p^{−a}y with y ∈N. Its image under g vanishes after inverting p. Therefore some power p^{b} kills g(y), and p^{b}y ∈ ker g. Thus the original element equals p^{−a−b}(p^{b}y). In particular the integral kernel spans the rational kernel. This argument is also the usual flat-localization identity C ⊗ ker g = ker g_{C}.

If B(x, ker g) = 0, the preceding spanning assertion gives B_{C}(x, ker g_{C}) = 0. Equation (G.2) then gives f(x) = 0 after localization. The target ω_{G}∨ is free over the local domain O_{C}, so it injects into its localization; hence f(x) = 0 integrally. Conversely, if f(x) = 0, (G.2) implies that B(x, y) is zero in C for each integral y ∈ ker g. The inclusion O_{C} ↪→C makes it zero in O_{C}. This proves (G.1).

For the factorization formulation, g_{C} is surjective by the Hodge–Tate sequence for G^{∨}. Lemma 10.3 supplies 0 ̸= κ ∈O_{C} with κω_{G} ⊆ im g. If the right side of (G.1) holds, apply Lemma 10.2 with ϕ(y) = B(x, y). It gives ℓ with κB(x, y) = ℓ(g(y)). In the reverse direction such a factorization gives cB(x, y) = 0 on ker g; cancellation of c ̸= 0 in the domain proves the right side of (G.1).

The project proves the last equivalence using an integral prismatic sandwich and the stronger bound p^{2}ω_{G} ⊆ im g. The numerical exponent two is absent from the statement FarguesExact: it only asks for (G.1). The proof just given establishes that entire statement with a common nonzero denominator, rather than asserting that rational surjectivity is integral surjectivity.

Lemma 10.6 (The root system and normalization of the pairing). Suppose ζ_{0} = 1, ζ_{n}^{p}_{+1} = ζ_{n} in O_{C}, and ζ_{1} ̸= 1. Then (ζ_{n}) generates the Tate module of µ_{p}∞. Any O_{C}-bilinear form having the prescribed scalar Cartier values on pure Tate tensors is the form B in Proposition 10.5.

Proof. Repeated powering gives ζ_{n}^{pn} = 1 and ζ_{n}^{pn−1} = ζ_{1} ̸= 1 for n > 0. Since p is prime, its order is exactly p^{n}. Every p^{n}-th root of unity lies in O_{C}: its valuation has p^{n}-th power one and so is one. In the algebraically closed characteristic-zero field C, these roots form a cyclic group of order p^{n}, generated by ζ_{n}. Write any compatible root system uniquely as (ζ_{n}^{an}), a_{n}∈ℤ/p^{n}ℤ. Compatibility says exactly that a_{n+1} reduces to a_{n}. Such a system is an element of ℤ_{p}, and conversely every element gives such a system. Thus multiplication by (ζ_{n}) identifies ℤ_{p} with this Tate module.

For t ∈T_{p}G and u ∈T_{p}G^{∨}, the Cartier pairing is consequently aζ for one uniquely determined a ∈ℤ_{p}. Its scalar value in O_{C} is the image of a. The hypothesis on a form specifies exactly this value at (1⊗t, 1⊗u). Multiplication by the two O_{C} coefficients specifies its value at every pair of pure tensors. Tensor products are spanned by pure tensors, so additivity in each argument gives equality of the forms everywhere.