Exposé IX. Algebraic geometry and formal geometry

The goal of this Exposé is to generalize, to the case of a morphism that is not proper, Theorems 3.4.2 and 4.1.5 of EGA III.

1. The comparison theorem

Let be a separated morphism of preschemes of finite type. Suppose that is locally noetherian. Let be a closed subset of and let .

Let and be the formal completions of and along and . Let be the morphism deduced from by passing to the completions.

    X ◀───── Y               X̂ ──j──▶  X
    │                        │           │
    │ f                      │ f̂         │ f
    ▼                        ▼           ▼
    X′ ◀──── Y′      ,       X̂′ ──i──▶ X′.

We denote by (resp. ) the homomorphism from into (resp. from into ). It is known that and are flat.

Let be an ideal of definition of , and let ; this is an ideal of definition of . One therefore has:

X̂′ = (Y′, lim_{k ∈ ℕ} 𝒪_{X′}/I′^{k+1}),    X̂ = (Y, lim_{k ∈ ℕ} 𝒪_X/J^{k+1}).

For every , set:

Y′_k = (Y′, 𝒪_{X′}/I′^{k+1}),    Y_k = (Y, 𝒪_X/J^{k+1}).

Let be a coherent -Module. For every , we set:

F_k = F/J^{k+1}F,    F̂ = j*(F) ≃ lim_{k} F_k.

If we set:

R^i f_*(F)^∧ = lim_{k ∈ ℕ} (R^i f_*(F) ⊗_{𝒪_{X′}} 𝒪_{Y′_k}),    i ∈ ℤ,

one has a natural homomorphism:

r_i: i*(R^i f_*(F)) → R^i f_*(F)^∧,

which is an isomorphism when is coherent.

As is explained in EGA III 4.1.1, one has a commutative diagram:

                       ρ_i
   i*(R^i f_*(F))  ──────▶  R^i f̂_*(F̂)
        │                       │
     r_i│                       │ψ_i
        ▼          ϕ_i          ▼
   R^i f_*(F)^∧  ──────▶  lim_{k ∈ ℕ} R^i f_*(F_k).

In loc. cit. one finds a commutative diagram, since one knows that is coherent, and one identifies the source and target of (1.6). In our case, will be coherent only for certain values of , for which we shall study (1.7).

Consider the graded ring

S = ⨁_{k ∈ ℕ} I′^k,

and the graded -Module:

H^i = ⨁_{k ∈ ℕ} R^i f_*(J^k F),    i ∈ ℤ,

whose -Module structure is defined as follows.

The sheaf is associated with the presheaf which, to every affine open of , associates:

H^i(f⁻¹(U′), J^k F | f⁻¹(U′)).

Let then be an affine open of , set

and let . Let be the image of in . The homothety of ratio on maps into , whence, by functoriality, a morphism:

μ^i_{x′, k}(U′): H^i(U, J^k F | U) → H^i(U, J^{k+m} F | U),

defined for every and every , which gives, by passing to the associated sheaf, the graded -Module structure on .

Theorem.

Let be an integer. Suppose that the graded -Module is of finite type for and . Then:

(0) and are isomorphisms, and is coherent for ;

(1) for , , , and are topological isomorphisms (in particular, the filtration defined on by the kernels of the homomorphisms

R^{n−1} f_*(F) → R^{n−1} f_*(F_k)

is -good);

(2) for , , , and are monomorphisms; furthermore, the filtration on is -good and is an isomorphism;

(3) the projective system of the satisfies, for , the uniform Mittag-Leffler condition, i.e. there exists a fixed integer such that, for every and every , one has:

Im[R^i f_*(F_{p′}) → R^i f_*(F_p)] = Im[R^i f_*(F_{p+k}) → R^i f_*(F_p)].

Proceeding as in EGA III 4.1.8, it is easy to reduce to the case where is the spectrum of a noetherian ring . In this case, one knows that

R^i f_*(F) = H^i(X, F)^~     (cf. 1.10).

Let be the ideal of such that , and let

S = ⨁_{k ∈ ℕ} I^k,
H^i = ⨁_{k ∈ ℕ} H^i(X, J^k F),    i ∈ ℤ,

where is equipped with the graded -module structure defined by 1.11, where one has taken .

The proof is modelled on that of EGA III 4.1.5; let us give a summary.

We work on and , which correspond to homomorphisms of modules:

                              H^i(X̂, F̂)
                                  │
                                  │ψ_i
                  ϕ_i             ▼
   H^i(X, F)^∧  ──────▶  lim_{k} H^i(X, F_k).

(a) We assume only that is a graded -module of finite type. We deduce that the filtration defined on by the modules:

R^i_k = ker(H^i(X, F) → H^i(X, F_k))

is -good. For this, we use the cohomology exact sequence:

H^i(X, J^{k+1} F) → H^i(X, F) → H^i(X, F_k),

which shows that the graded -module is a quotient of the graded -submodule

⨁_{k ∈ ℕ} H^i(X, J^{k+1} F)

of , hence is of finite type, since is noetherian. Whence this first point.

Set:

M^i = H^i(X, F),    H^i_k = H^i(X, F_k).

One has a commutative diagram:

                     s_i
   H^i(X, F)^∧  ─────────▶  lim_{k} (M^i / R^i_k)
        \                         │
         \                        │
          \ ϕ_i                   │ t_i
           \                      ▼
            ──────────────▶  lim_{k} H^i_k,

in which is an isomorphism; indeed, the filtration of is -good. Moreover, is a monomorphism; indeed, the functor lim is left exact, and, for every , the natural morphism is a monomorphism, by definition of .

To study the surjectivity of , we introduce:

Q^i_k = coker(H^i(X, F) → H^i(X, F_k)),

whence a projective system of exact sequences:

0 → R^i_k → M^i → H^i_k → Q^i_k → 0.

Using the cohomology exact sequence:

H^i(X, F) → H^i(X, F_k) → H^{i+1}(X, J^{k+1} F),

one sees that the graded -module

Q^i = ⨁_{k ∈ ℕ} Q^i_k

is a graded -submodule of . Moreover, for every , one has:

since is the image of .

(b) We assume only that is of finite type, and we focus on (forgetting ). Since is noetherian, is of finite type; since vanishes, we find that there exist an integer and an integer such that

I^r Q^i_k = 0    for k ⩾ k₀.

It follows that the projective system is essentially zero, and hence the projective system satisfies the uniform Mittag-Leffler condition. From the exact sequence (1.22) one deduces the exact sequence

0 → M^i / R^i_k → H^i_k → Q^i_k → 0,

whence the exact sequence:

0 → lim_{k} M^i / R^i_k  ──t_i──▶  lim_{k} H^i_k → lim_{k} Q^i_k.

Now the projective system is essentially zero, hence is an isomorphism.

(c) Let us prove that, if is of finite type, then is an isomorphism. It suffices to apply EGA 0_III 13.3.1, taking as a basis of opens of the affine opens. This is legitimate;

indeed, by (b), the projective system satisfies the Mittag-Leffler condition.

The theorem follows formally from (a), (b), and (c). One will note that, in fact, the proof uses, at each step, the finiteness of only for a single value of .

Let us give some examples in which the hypothesis of Theorem 1.1 is satisfied.

Corollary.

Suppose that is generated by a section of , and denote by the corresponding section of . Let be a coherent -module and let be an integer. Suppose that:

(i) is -regular (i.e. the homothety of ratio on is a monomorphism).

(ii) is coherent for and .

Then the hypothesis of Theorem 1.1 is satisfied.

Indeed, one observes that multiplication by defines an isomorphism , and one deduces that

H^i ≃ R^i f_*(F) ⊗_{𝒪_{X′}} 𝒪_{X′}[T],

where is an indeterminate. Whence the conclusion.

Corollary.

Suppose that , where is a noetherian ring separated and complete for the -adic topology. Suppose that the -module is of finite type for and (cf. 1.14 and 1.15). Then the hypotheses of Theorem 1.1 are satisfied, and one finds a commutative diagram of isomorphisms:

                    ρ′_i
   H^i(X, F)  ─────────────▶  H^i(X̂, F̂)
         \                      /
          \                    /
           \ ϕ′_i         ψ_i /
            \                /
             ▼              ▼
              lim_{k} H^i(X, F_k)              for i = n − 1.

1

One simply notes that is of finite type, hence isomorphic to its completion. One obtains (1.1) by transcribing the diagram of Modules (1.7) into the category of -modules, and replacing the left vertical by .

Proposition.

Let be a noetherian ring. Let and suppose that is separated and complete for the (tA)-adic topology. Set:

X′ = Spec(A),    Y′ = V(t),    I = (tA).

Let be a closed subset of ; set

X = X′ − T,    Y = Y′ ∩ X = Y′ − (Y′ ∩ T).

Let be a coherent -Module. Finally, let

T′ = {x ∈ X′ | codim({x} ∩ T, {x}) = 1}.

Suppose that:

a) is -regular,

b) ,

c) is a quotient of a regular noetherian ring.

Then, in diagram (1.1), the morphisms , , and are isomorphisms for and monomorphisms for . Moreover is an isomorphism.

By virtue of 1.3 and 1.2, it suffices to prove that is coherent for , which follows from the finiteness theorem 2.1.2

In particular:

Example.

One will apply 1.4 when is a local ring and belongs to the radical of . One will then take . In this case, for , one obtains the following statement:

If is noetherian, separated and complete for the -adic topology, and a quotient of a regular ring (for example, if is complete), if moreover is -regular and if for every such that , then the natural homomorphism

Γ(X, F) → Γ(X̂, F̂)

is an isomorphism.

Indeed, keeping the notation of 1.4, one has , and formula (1.33) says that

T′ = {x ∈ Spec(A) | dim A/x = 1}.

2. The existence theorem

Let us first state EGA III 3.4.2 in a slightly more general form.

Let be an adic morphism3 of formal preschemes, with noetherian. Let be an ideal of definition of ; since is adic, is4 an ideal of definition of .

For every , set

this is an ordinary prescheme having the same underlying topological space as .

Let be a coherent -Module. For every , the -Modules

are coherent. For every , one has a homomorphism

ψ_i: R^i f_*(F) → lim_{k} R^i f_*(F_k),

deduced by functoriality from the natural homomorphism:

Set

S = gr_{I′} 𝒪_{X′} = ⨁_{k ∈ ℕ} I′^k / I′^{k+1},
gr_J(F) = ⨁_{k ∈ ℕ} J^k F / J^{k+1} F,
K^i = R^i f_*(gr_J(F)) = ⨁_{k ∈ ℕ} R^i f_*(J^k F / J^{k+1} F).

It is clear that is equipped with a graded -Module structure.

Theorem.

Suppose that is a graded -Module of finite type for , , . Then:

(i) is coherent.

(ii) The homomorphism (2.3) is a topological isomorphism. The natural filtration of the right-hand side of (2.3) is -good.

(iii) The projective system of the satisfies the uniform Mittag-Leffler condition.

The proof is very easy from EGA 0_III 13.7.7 (cf. EGA III 3.4.2), provided one corrects the text on page 78 as indicated in (EGA III 2, Err_III 24).

Theorem.

5

Let be a noetherian adic ring and let be an ideal of definition of . Let be a closed subset of . Suppose that is generated by a . Take up the notation 1.31, 1.32, and 1.33. Let be a coherent -Module. Set

where is an ideal of definition of . Suppose that is a quotient of a regular noetherian ring and that:

(1) is -regular,

(2) .

Then there exists a coherent -Module such that .

It suffices to prove that is a coherent -Module, where is the morphism of formal preschemes deduced from the injection of into by completion with respect to . Indeed, is separated and complete for the -adic topology, so there will exist an -module whose completion will be isomorphic to . Since is an open of , one will be able to take .

It remains to show that 2.1 is applicable to the morphism of formal preschemes and to . Now, by hypothesis (1), for every one has an isomorphism:

J^k F / J^{k+1} F → F/J F,

whence it follows that the hypothesis of 2.1 will be satisfied if one knows that

R^i f_*(F₀) is coherent for i ⩽ 1.

Now this follows from (2) and from the finiteness theorem 2.1.2 Whence the conclusion.

It remains to specialize this statement by supposing that is a local ring.

Corollary.

Let be a noetherian local ring and let be an element of the radical of . Suppose that is separated and complete for the -adic topology and, moreover, a quotient of a regular ring (for example, suppose that is a complete noetherian local ring). Set

X′ = Spec A,    T = {r(A)},

and take up the notation (1.31), (1.32), and (1.33). Let be an -Module. Suppose that:

(1) is -regular,

(2) , with and .

Then there exists a coherent -Module such that .

Note that here is the set of prime ideals of such that .

1

N.D.E. In the same vein, see the article by Chow (Chow W.-L., "Formal functions on homogeneous spaces", Invent. Math. 86 (1986), no. 1, pp. 115–130). The author proves the following result. Let be an algebraic variety over a field, homogeneous under an algebraic group , and let be a complete subvariety of of dimension > 0. Suppose that generates in the following sense: given , let be the set of elements of sending into . One then says that generates if the group generated by the connected component of 1 of is the whole of . In this case, every formal rational function on along is algebraic; compare with the results of Hironaka and Matsumura cited in editor's note (3) p. 138. In the line of the techniques introduced by these authors, let us point out the very pretty algebraization result due to Gieseker (Gieseker D., "On two theorems of Griffiths about embeddings with ample normal bundle", Amer. J. Math. 99 (1977), no. 6, pp. 1137–1150, Theorems 4.1 and 4.2). Let be a connected projective variety of dimension > 0, locally a complete intersection (over an algebraically closed field). Suppose one has two embeddings of into smooth projective varieties Y, W. Then, if the formal completions of in and are equivalent, there exists a scheme containing (as a closed subscheme) which embeds into and as an étale neighborhood of in and . In other words, formally equivalent entails étale-equivalent. See also the article of Faltings (Faltings G., "Formale Geometrie und homogene Räume", Invent. Math. 64 (1981), pp. 123–165).

2

N.D.E. The "finiteness theorem 2.1" referenced here is the finiteness theorem of Exposé VIII (VIII 2.3), not the Theorem 2.1 of the present Exposé; the source's local cross-reference is to the cohomological finiteness statement on which the existence theorem is built.

3

This hypothesis is not essential; cf. XII, p. 118.

4

N.D.E. By definition itself, cf. EGA I 10.12.1.

5

N.D.E. Numerous algebraization statements have been obtained since, not to mention those cited below; cf. the articles of Faltings or of Mme Raynaud cited in editor's notes (22) p. 155 and (7) p. 203 respectively. One has in mind in particular the results of Artin (see notably Artin M., "Algebraization of formal moduli. I", in Global Analysis (Papers in Honor of K. Kodaira), Univ. Tokyo Press, Tokyo, 1969, pp. 21–71), but also the recent algebraicity results for leaves of foliations; see notably Bost J.-B., "Algebraic leaves of algebraic foliations over number fields", Publ. Math. Inst. Hautes Études Sci. 93 (2001), pp. 161–221, and Chambert-Loir A., "Théorèmes d'algébricité en géométrie diophantienne (d'après J.-B. Bost, Y. André, D. & G. Chudnovsky)", in Séminaire Bourbaki, Vol. 2000/2001, Astérisque, vol. 282, Société mathématique de France, Paris, 2002, Exp. 886, pp. 175–209, and the references cited therein. In particular, one will find in these two articles discussions of the link between algebraization questions and the theory of diophantine approximation.