§20. Derivations and differentials

The notions introduced in this section will be taken up in geometric form in chapter IV, §16, and will play an important role in the study of preschemes. Their importance in the present chapter rests first of all on their connections with the notion of formally smooth algebra, and notably on theorems (20.4.9), (20.5.7), and (20.5.12), which will be translated into geometric language in the section of chapter IV devoted to smooth morphisms, and are of constant use in applications. On the other hand, the differential notions will serve in §22 to prove important regularity criteria which will play an essential role in the deeper study of Noetherian local rings carried out in §7 of chapter IV.

20.1. Derivations and extensions of algebras

Proposition (20.1.1).

Let be a ring (not necessarily commutative), , , -rings, an -homomorphism whose kernel is of square zero, an -homomorphism. Suppose that there exists an -homomorphism such that factors as . Then the set of -homomorphisms such that is identical to the set of maps , where is a map satisfying the two following conditions:

(i) is a homomorphism of -bimodules.

(ii) For every pair of elements , in , one has

(20.1.1.1)                       D(fg) = f · D(g) + D(f) · g.

To say that for means that belongs to ; writing that , one obtains the relation (20.1.1.1), being of square zero, and condition (i) results from the fact that and are -homomorphisms.

If is the structural homomorphism, one derives from (20.1.1.1) that for every ; but one must have by (i), so, taking , it follows that

                                D(ρ(a)) = 0     for     a ∈ A;

conversely, if is zero on and satisfies (20.1.1.1), it also satisfies (i).

Definition (20.1.2).

Given an -ring and an -bimodule , one calls an -derivation of into a map satisfying conditions (i) and (ii) of (20.1.1).

It follows from (20.1.1.1) that the kernel of an -derivation is a sub--ring of .

One sometimes calls a derivation of into a -derivation, that is, an additive map from into satisfying (20.1.1.1); one has therefore for such a map. If is an algebra over a prime field , every -derivation of is a -derivation: this is clear from what precedes if is of characteristic > 0, and in the opposite case, the relation for gives

so since is of characteristic 0.

It follows at once from definition (20.1.2) that if , are two -derivations of into , the same holds for . In other words, the set of -derivations of into is endowed with a structure of additive group; this group is denoted . If is commutative, a commutative -algebra, and a -module, then, for every -derivation of into and every , aD is again an -derivation of into ; in other words, is then equipped with a structure of -module.

Proposition (20.1.1) is interpreted as follows:

Corollary (20.1.3).

Given two -homomorphisms of -rings , the first of which has a kernel of square zero, the set of -homomorphisms such that is empty or is a principal homogeneous space for the group .

In particular:

Corollary (20.1.4).

Let be a ring, an -ring, a -bimodule, an -extension of by , the augmentation. The map which, to every derivation , associates the map is an isomorphism of the group onto the group of -equivalences of with itself.

Apply (20.1.1) for , : the set of -homomorphisms such that is identical to the set of maps , where . To say that such an -homomorphism is an -equivalence amounts to saying that reduces to the canonical injection on , or again that on ; but this also means that factors as , where is an -derivation. Whence the corollary.

For trivial extensions, (20.1.3) gives:

Corollary (20.1.5).

Let be a ring, , two -rings, a -bimodule, an -homomorphism; the map which, to every derivation , associates the map is a bijection onto the set of -homomorphisms (cf. (18.2.3)) such that factors as .

More particularly:

Corollary (20.1.6).

Let be a ring, an -ring, a -bimodule. If, to every derivation , one associates: 1° the -equivalence of the extension with itself; 2° the -homomorphism of into , a right inverse of the augmentation homomorphism , one defines canonical bijective correspondences between:

(i) the set ;

(ii) the set of -equivalences of with itself;

(iii) the set of -homomorphisms right inverse to the augmentation homomorphism .

It is to be noted that the bijective correspondence thus established between (i) and (ii) respects the group structures, and that the one deduced from it between (ii) and (iii) is none other than the bijective correspondence already defined in (18.3.8).

Corollary (20.1.7).

Let be a ring, an -ring, an -bimodule, a trivial -extension of by , the augmentation. The set of -derivations of into such that on is identical to the set of maps , where ranges over the set of -homomorphisms right inverse to .

It again suffices to apply (20.1.1) for , ; if , the condition for is equivalent to for , that is, to , where is an -homomorphism; in addition the condition is equivalent to , in other words to the fact that is a right inverse of .

20.2. Functorial properties of derivations

(20.2.1) Let be a ring, an -ring, a -bimodule; if is a second -bimodule and a homomorphism of -bimodules, it is clear that the map is a homomorphism of additive groups

(20.2.1.1)                  w_0 : Der_A(B, L) → Der_A(B, L')

and that if is a second homomorphism of -bimodules, one has . When is commutative, a commutative -algebra and a -module, (20.2.1.1) is a homomorphism of -modules.

In the second place, let be an -ring, an -homomorphism which makes into a -bimodule; then the map is a homomorphism of additive groups,

(20.2.1.2)                  v^0 : Der_A(B, L) → Der_A(B', L)

as follows from (20.1.1.1); if is a second -homomorphism, one has . When , , and are commutative and a -module, (20.2.1.2) is a homomorphism of -modules.

Finally, let be a ring homomorphism making into an -ring; every -derivation is also an -derivation, whence a canonical injection of commutative groups

(20.2.1.3)                  u^0 : Der_A(B, L) → Der_{A'}(B, L)

and if is a second ring homomorphism, one has ; when , , and are commutative and a -module, (20.2.1.3) is a di-homomorphism of modules (relative to ).

One may further say that

                              (A, B, L) ↦ Der_A(B, L)

is a covariant functor from the category defined in (18.3.5) to the category Ab of commutative groups, by making correspond, to every triple constituting a morphism of , the homomorphism ; the verification of functoriality follows from the commutativity of the diagrams

       Der_A(B, L) ─────→ Der_A(B', L)       Der_A(B, L) ─────→ Der_{A'}(B, L)
            │                  │                  │                   │
         w_0│               w_0│               w_0│                w_0│
            ↓                  ↓                  ↓                   ↓
       Der_A(B, L') ────→ Der_A(B', L')      Der_A(B, L') ────→ Der_{A'}(B, L')
                                                          u^0

for every homomorphism of -bimodules.

Theorem (20.2.2).

Let , be two ring homomorphisms, a -bimodule. One has a canonical exact sequence of commutative groups

(20.2.2.1)   0 → Der_B(C, L) → Der_A(C, L) →^{v^0} Der_A(B, L)
                  →^{∂} Exan_B(C, L) →^{u^1} Exan_A(C, L) →^{v^1} Exan_A(B, L)

where , are the homomorphisms (20.2.1.3) and (20.2.1.2) respectively, , the homomorphisms defined in (18.3.4.1) and (18.3.3.1) respectively, and where is defined as follows: for every -derivation of into , is the class of the -extension of by defined on the ring by the -homomorphism (cf. (20.1.5)). Furthermore, the exact sequence (20.2.2.1) is functorial in (for the homomorphisms defined in (20.2.1.1) and (18.3.1.1) respectively).

Since is an -derivation (and a fortiori a -derivation) of into , the -homomorphism does define on a structure of -extension, hence is well defined (20.1.5). Exactness must be verified at five places:

  1. Exactness at is trivial (cf. (20.2.1)).

  2. By definition (20.2.1), the kernel of is the set of -derivations of into which vanish on , that is, those -derivations which are also -derivations (20.1.1); whence exactness at .

  3. The kernel of is formed by the derivations for which the -extension defined by is -trivial; this means (18.2.3) that there exists a -homomorphism from into (the -ring structure on being defined by ); but such a homomorphism, being a fortiori an -homomorphism, is of the form where (20.1.6); and writing that it is a -homomorphism gives

            D'(v(x) z) + v(x) D'(z) = D'(v(x)) z + v(x) D'(z) = (D(x) + D'(v(x))) z,

for , , which yields ; the kernel of is therefore the image of .

  1. The kernel of is the set of classes of -extensions of by which are -trivial (18.3.7), so (up to equivalence) of the form , where the -ring structure is defined by the homomorphism . Now every -extension structure on is defined by a homomorphism , where is a -derivation of into (20.1.5); to say that the -ring structure of this -extension is deduced from its -ring structure by means of means that for , hence that is an -derivation, or again that the class of the -extension considered is of the form ; whence exactness at .

  2. The kernel of is the set of classes of -extensions of by which become trivial on , that is, those for which there exists an -homomorphism such that factors as ; but such an -homomorphism defines on a structure of -extension whose class has as image under the class of the given -extension; the converse being trivial, exactness at is proved.

Finally, functoriality in follows trivially from the definitions.

Corollary (20.2.3).

Let , , be three commutative rings, , two ring homomorphisms, a -module. One has a canonical exact sequence of -modules

(20.2.3.1)   0 → Der_B(C, L) → Der_A(C, L) →^{v^0} Der_A(B, L) →^{∂}
                  → Exalcom_B(C, L) →^{u^1} Exalcom_A(C, L) →^{v^1} Exalcom_A(B, L)

functorial in .

The reasoning is the same as in (20.2.2) once one has verified that for every derivation , is indeed the class of a -extension of by which is a commutative -algebra; but this follows at once from the commutativity of and the fact that is a -module.

Corollary (20.2.4).

Under the hypotheses of (20.2.2) (resp. (20.2.3)), one has a canonical exact sequence, functorial in ,

(20.2.4.1)   0 → Der_B(C, L) → Der_A(C, L) →^{v^0} Der_A(B, L) →^{∂} Exan_{B/A}(C, L) → 0

(resp.

(20.2.4.2)   0 → Der_B(C, L) → Der_A(C, L) →^{v^0} Der_A(B, L) →^{∂} Exalcom_{B/A}(C, L) → 0).

This follows from the definition of (resp. ) ((18.3.7) and (18.4.2)).

Remark (20.2.5).

Suppose one has a commutative diagram of ring homomorphisms

                                A ──→ B ──→ C
                                │     │     │
                                ↓     ↓     ↓
                                A' ──→ B' ──→ C'

Then one has a commutative diagram

0 → Der_B(C, L) → Der_A(C, L) → Der_A(B, L) → Exan_B(C, L) → Exan_A(C, L) → Exan_A(B, L)
        │              │              │             │              │              │
        ↓              ↓              ↓             ↓              ↓              ↓
0 → Der_{B'}(C', L) → Der_{A'}(C', L) → Der_{A'}(B', L) → Exan_{B'}(C', L) → Exan_{A'}(C', L) → Exan_{A'}(B', L)

and likewise for the exact sequences (20.2.3.1), (20.2.4.1), and (20.2.4.2).

20.3. Continuous derivations in topological rings

(20.3.1) Given two topological rings , (linearly topologized as always), we denote by the set of continuous homomorphisms from to . Given a topological ring , the category of topological -rings is defined as that of -rings (18.1.4) by replacing everywhere "ring" by "topological ring" and "homomorphism" by "continuous homomorphism"; if

and are two topological -rings, we denote by the set of continuous -homomorphisms from to .

Let be a topological ring, a topological -ring, a topological -bimodule; we denote by the set of continuous -derivations from into ; it is clear that this is a subgroup of (and a sub--module when and are commutative and is a -module).

(20.3.2) It is immediate that proposition (20.1.1) remains valid when one replaces in it "ring" by "topological ring" and "homomorphism" by "continuous homomorphism". Likewise, (20.1.3) and (20.1.4) remain valid by replacing Der by , "ring" (resp. "bimodule") by "topological ring" (resp. "topological bimodule"), "homomorphism" by "continuous homomorphism"; it is naturally necessary to assume in (20.1.4) that is continuous, and to replace "-equivalences" by "continuous -equivalences". One has the same results for (20.1.5) and (20.1.6), provided one takes as topology on (resp. ) the product topology (on , resp. ).

Proposition (20.3.3).

Let be a topological ring, a topological -ring, a discrete topological -bimodule annihilated by an open two-sided ideal of . If in the square of every open two-sided ideal is open, then one has Der.cont_A(B, L) = Der_A(B, L).

Indeed, if is an open two-sided ideal of annihilating , and an -derivation of into , one has (20.1.1.1), hence is continuous.

(20.3.4) All the results of (20.2.1) remain valid when one replaces in them "ring" by "topological ring", "bimodule" by "topological bimodule", "homomorphism" by "continuous homomorphism", and Der by .

Proposition (20.3.5).

Let be a topological ring, a topological -ring, a discrete topological -bimodule annihilated by an open two-sided ideal of . One has then a canonical isomorphism

(20.3.5.1)              lim Der_{A/𝔍}(B/𝔎, L) ≅ Der.cont_A(B, L)

where in the left-hand side the inductive limit is taken over the filtered ordered set of pairs of two-sided ideals such that , , .

Since and are discrete, one has canonical homomorphisms w_{𝔎,𝔍} : Der_{A/𝔍}(B/𝔎, L) → Der.cont_A(B, L) forming an inductive system (20.3.4), whence the homomorphism (20.3.5.1) by passage to the inductive limit. Since the homomorphism is surjective for , it follows at once from the definition that the homomorphism (with , , ) is injective, and it is evidently the same for the homomorphism for (20.2.1); one concludes that the homomorphism (20.3.5.1) is injective. On the other hand, if is a continuous -derivation of into , its kernel contains an open two-sided ideal of , and if is an open two-sided ideal of such that and , it is clear that is the canonical image of an -derivation of into , hence (20.3.5.1) is surjective.

Proposition (20.3.6).

Let , be two continuous homomorphisms of topological rings, a discrete -bimodule annihilated by an open two-sided ideal of . One has a canonical exact sequence

(20.3.6.1)   0 → Der.cont_B(C, L) → Der.cont_A(C, L) →^{v^0} Der.cont_A(B, L) →^{∂}
                  → Exantop_B(C, L) →^{u^1} Exantop_A(C, L) →^{v^1} Exantop_A(B, L)

where is defined by passage to the inductive limit from the homomorphism of (20.2.2.1); this exact sequence is functorial in (in the category of -bimodules discrete and annihilated by open two-sided ideals).

This follows from the exactness of the functor lim, starting from (20.2.2).

Corollary (20.3.7).

Let , , be three commutative topological rings, , two continuous homomorphisms, a discrete -module annihilated by an open ideal of . One has a canonical exact sequence of -modules, functorial in ,

(20.3.7.1)   0 → Der.cont_B(C, L) → Der.cont_A(C, L) →^{v^0} Der.cont_A(B, L) →^{∂}
                  → Exalcotop_B(C, L) →^{u^1} Exalcotop_A(C, L) →^{v^1} Exalcotop_A(B, L).

Corollary (20.3.8).

Under the hypotheses of (20.3.5) (resp. (20.3.6)) one has a canonical exact sequence, functorial in ,

(20.3.8.1)   0 → Der.cont_B(C, L) → Der.cont_A(C, L) →^{v^0} Der.cont_A(B, L) →^{∂}
                                                                  → Exantop_{B/A}(C, L) → 0

(resp.

(20.3.8.2)   0 → Der.cont_B(C, L) → Der.cont_A(C, L) →^{v^0} Der.cont_A(B, L) →^{∂}
                                                              → Exalcotop_{B/A}(C, L) → 0).

We leave to the reader the task of writing the diagrams analogous to those of (20.2.5).

20.4. Principal parts and differentials

In the whole sequel of this section and in the three following ones, all rings are assumed to be commutative.

(20.4.1) Let be a topological ring, a topological -algebra; the -algebra will be equipped with the tensor-product topology, which makes it a topological -algebra; we denote by (or ) the canonical surjective -homomorphism

(20.4.1.1)                             p : B ⊗_A B → B

such that ; it is immediate that is continuous. The kernel of will be denoted (or simply if there is no risk of confusion). We denote by and the two canonical -homomorphisms, such that

                              j_1(b) = b ⊗ 1,        j_2(b) = 1 ⊗ b

which are continuous.

Definition (20.4.2).

One calls augmented -algebra of principal parts of order 1 of relative to and denotes by the quotient topological -algebra

(20.4.2.1)                          𝒫^1_{B/A} = (B ⊗_A B) / 𝔍^2

equipped with the structure of -algebra defined by the homomorphism (deduced from by composition with the canonical homomorphism ), and with the augmentation of -algebra (also denoted ) deduced from by passage to the quotient.

Since by definition, it is clear that is indeed an augmentation of -algebra.

Definition (20.4.3).

The kernel of the augmentation ,

equipped with the topology induced by that of , which makes it a topological -module, is called the -module of 1-differentials (or simply of differentials) of relative to .

It is to be noted that the topology of is also the quotient topology of the topology induced on by that of (Bourbaki, Top. gén., chap. III, 3rd ed., §2, n° 7, prop. 20). If is discrete the same holds for . We denote by the separated completion of the topological -module .

Any topological ring may be regarded as a topological -algebra ( being equipped with the discrete topology), so that one can define the topological -module , which is sometimes also called the -module of absolute differentials over and is denoted . If is a topological algebra over a prime field (discrete), one has , hence .

Lemma (20.4.4).

Let be a ring, an -algebra. The ideal of is generated by the elements , where ranges over a generating set of the -algebra .

It is clear that for every , one has ; on the other hand, for any , in , one has . If , one has by definition , so

(20.4.4.1)             ∑ (x_i ⊗ y_i) = ∑ (x_i ⊗ 1)(1 ⊗ y_i − y_i ⊗ 1)
                        i              i

which proves that is the ideal generated by the elements . In addition, if , one has

(20.4.4.2)       x ⊗ 1 − 1 ⊗ x = (s ⊗ 1)(t ⊗ 1 − 1 ⊗ t) + (s ⊗ 1 − 1 ⊗ s)(1 ⊗ t)

which immediately concludes the proof by induction.

Proposition (20.4.5).

Let be a topological ring, a topological -algebra. The topology of is coarser than the topology deduced from that of ((19.0.2)); if in the square of every open ideal is open, these two topologies are identical.

The first assertion is trivial, the tensor-product topology on being coarser than the topology deduced from that of ; a fortiori the topology induced on

by that of is coarser than the topology on deduced from that of . To prove the second assertion, write , by abuse of notation, for the sub-module for two sub--modules , of . Using the relation

(xy) ⊗ z − x ⊗ (yz) = (x ⊗ 1)(1 ⊗ y)(z ⊗ 1 − 1 ⊗ z) = x · (yz ⊗ 1 − 1 ⊗ yz) − x · (y ⊗ 1 − 1 ⊗ y) · z

in the -module (defined by ), one sees at once, taking (20.4.4) into account, that, if is an ideal of , one has

                  ((𝔎 ⊗ B) + (B ⊗ 𝔎)) ∩ 𝔍 ⊂ (𝔎 ⊗ 𝔎) ∩ 𝔍 + 𝔎 · 𝔍 + 𝔍^2

and on the other hand, if , are elements of such that , it follows from (20.4.4.1) that one has , so that finally

(20.4.5.1)                  (𝔎 ⊗ B + B ⊗ 𝔎) ∩ 𝔍 ⊂ 𝔎 · 𝔍 + 𝔍^2.

Now one has a fundamental system of neighbourhoods of 0 in (for the topology induced by that of ) by taking as neighbourhoods of 0 the sets , where ranges over the set of open ideals. Since the topology of deduced from that of is also the quotient by of the topology of deduced from that of , the hypothesis on the open ideals of and the relation (20.4.5.1) complete the proof of the second assertion.

Definition (20.4.6).

Let and be the composite maps , , which are continuous -homomorphisms such that . The continuous -homomorphism of -modules

(20.4.6.1)                          d_{B/A} = p_2 − p_1 : B → Ω_{B/A}

is called the exterior differential of relative to ; for every , (also denoted or dx) is called the differential of (relative to ).

When , one writes instead of ; if is an algebra over a prime field , one has .

Proposition (20.4.7).

The -module is generated by the elements , where ranges over a system of generators of the -algebra .

Since is the canonical image of in , the proposition is an immediate consequence of (20.4.4).

Theorem (20.4.8).

Let be a topological ring, a topological -algebra.

(i) There exists a unique isomorphism of augmented topological -algebras

which reduces to the identity on .

(ii) The homomorphism is an -derivation of into , having the following universal property: for every topological -module , the map is an isomorphism of -modules

(20.4.8.2)              Hom.cont_B(Ω_{B/A}, L) ≅ Der.cont_A(B, L).

(i) It is immediate that (with the notations of (20.4.6)) is necessarily the map , and the inverse isomorphism is the map , these two maps being continuous, which proves the first assertion. It is to be noted that this implies that the topology of is identified (by ) with the quotient of the topology of by .

(ii) The fact that is an -derivation of results from definition (20.4.6) and from (20.1.1).

To prove the universal property, recall that is canonically identified with the set of continuous homomorphisms of -algebras such that the composite is the identity (where is the projection; cf. (20.1.6) and (20.3.2)). On the other hand, thanks to the isomorphism , is canonically identified with the set of continuous homomorphisms of -algebras such that the composite is the augmentation . Since by definition, everything reduces to proving that every factors as

                                     B ─────────→ D_B(L)
                                      ↘            ↗
                                    p_1│         v
                                        ↘        ↗
                                          𝒫^1_{B/A}

where is a continuous -homomorphism. Now one already has a continuous homomorphism of -algebras from into , which belongs to ; by the definition of the topological tensor product of topological algebras , there exists therefore a continuous -homomorphism of algebras making commutative the diagram

                                            j_2
                                  B ⊗_A B ──────  B
                                       │           │
                                    w  │           │ u
                                       ↓           ↓
                                  D_B(L) ──────→  B
                                              q

One has therefore by definition , and by virtue of (20.4.4), this entails so that ; consequently factors as

                              B ⊗_A B → 𝒫^1_{B/A} →^{v} D_B(L)

where is a continuous homomorphism of -algebras; moreover, since is a homomorphism of -algebras, so is by the definition of the -algebra structure of ; since one has by definition , this completes the proof.

Theorem (20.4.9).

Suppose is a formally smooth topological -algebra. Then the topological -module is formally projective.

Indeed, the hypothesis entails that (equipped with the structure of topological -algebra defined by ) is a formally smooth topological -algebra (19.3.5, (iii)), and consequently also a formally smooth topological -algebra (19.3.5, (ii)); since is topologically isomorphic to the quotient -algebra and is a formally smooth -algebra, the conclusion follows from (19.5.3).

Corollary (20.4.10).

Suppose in addition that in the square of every open ideal is open; then, for every open ideal of , is a projective -module.

In fact, the topology of is then deduced from that of (20.4.5), and it suffices to apply (19.2.4).

Corollary (20.4.11).

Let , be two Noetherian local rings, a local homomorphism making into a formally smooth topological -algebra (for the preadic topologies). Then, for every ideal of definition of , is a free -module.

In fact, it follows from (20.4.10) that this module is projective, and since is an Artinian ring, every projective -module is free .

Proposition (20.4.12).

Let be a topological ring, a topological -algebra. If the structural homomorphism is surjective, one has .

Indeed, one has for every -module by virtue of (20.1.1), and the proposition therefore follows at once from (20.4.8).

Examples (20.4.13).

(i) Let be a ring, a polynomial algebra over . Then is a free -module, of which the form a basis.

Indeed, the generate this -module (20.4.7). On the other hand, if is a free -module having a basis indexed by , there exists an -homomorphism of into such that for every , hence (20.1.5) an -derivation of into such that for every ; by virtue of (20.4.8.1), this proves that the are linearly independent.

*(ii) Let be a ring, an -module, the -algebra ; then the canonical homomorphism of into is bijective, for it is immediate that the -derivations of into a -module are the maps of the form , where ; one then concludes by (20.4.8, (ii)).

(iii) Suppose that is the product of two topological -algebras B_1, B_2 (identified with ideals of ). Then the images of and of under the homomorphism (20.4.1.1) are zero, whence it follows at once that is identified with the product , and that the -module is the (topological) direct sum of the -modules and (annihilated respectively by B_2 and B_1).

(iv) Let , be two integral rings such that , is integrally closed, is integral over , and the field of fractions of is a separable extension of that of . Then the -module is a torsion module.

Indeed, for every , the minimal polynomial of with respect to the field of fractions of is a polynomial belonging to A[T]; since is separable over , one has , and on the other hand one deduces from the relation that , whence our assertion by virtue of (20.4.7).

Remarks (20.4.14).

(i) It is to be noted that the -module , deprived of its topology, is independent of the topologies of and of .

(ii) We shall introduce later the "algebra of principal parts of order " of relative to , , which is the basis of "differential calculus of order ".

20.5. Fundamental functorial properties of

(20.5.1) In the whole of this number and the following one, unless expressly stated otherwise, the rings and modules considered are assumed to be equipped with the discrete topology.

(20.5.2) Let be a ring, , two -algebras, an -homomorphism; one has a commutative diagram

                                  u ⊗ u
                          B ⊗_A B ────── C ⊗_A C
                              │              │
(20.5.2.1)                p_{B/A}         p_{C/A}
                              ↓              ↓
                              B ───────────→ C
                                     u

whence by passage to the quotients, an -homomorphism of algebras

such that the diagram

                                          u'
                              𝒫^1_{B/A} ────── 𝒫^1_{C/A}
                                  ↑              ↑
                              p_1 │              │ p_1
                                  │              │
                                  B ───────────→ C
                                          u

is commutative; since maps into , one obtains, by restricting to , a map

such that the pair (u'', u) is a di-homomorphism for the -module structure of and the -module structure of ; this last fact allows one to deduce canonically a homomorphism of -modules

(20.5.2.4)                          u_{C/B/A} : Ω_{B/A} ⊗_B C → Ω_{C/A}.

In addition, since the diagram

                                          u'
                              𝒫^1_{B/A} ────── 𝒫^1_{C/A}
                                  ↑              ↑
(20.5.2.5)                    p_2 │              │ p_2
                                  │              │
                                  B ───────────→ C
                                          u

is also commutative, one deduces that the diagram

                                          u''
                              Ω_{B/A} ────── Ω_{C/A}
                                  ↑              ↑
(20.5.2.6)                  d_{B/A}            d_{C/A}
                                  │              │
                                  B ───────────→ C
                                          u

is commutative.

Finally, if is a second homomorphism of -algebras, one has the transitivity property

(20.5.2.7)              (w ∘ u)_{D/B/A} = w_{D/C/A} ∘ (u_{C/B/A} ⊗ 1)

as follows from the definition.

(20.5.3) Let now , be two rings, a ring homomorphism, a -algebra which becomes an -algebra by means of ; then the canonical map is a surjective di-homomorphism of algebras (relative to ) such that the diagram

                                          v_0
                                 C ⊗_A C ────── C ⊗_B C
                                     │              │
(20.5.3.1)                       p_{C/A}        p_{C/B}
                                     ↓              ↓
                                     C ───────────→ C
                                              1_C

is commutative; by passage to the quotients, one deduces a di-homomorphism of algebras

such that the diagram

                                          v'
                              𝒫^1_{C/A} ────── 𝒫^1_{C/B}
                                  ↑              ↑
                              p_1 │              │ p_1
                                  │              │
                                  C ───────────→ C
                                          1_C

is commutative. Since maps into one obtains, by restricting to , a map

which is a homomorphism of -modules.

In addition, since the diagram

                                          v'
                              𝒫^1_{C/A} ────── 𝒫^1_{C/B}
                                  ↑              ↑
(20.5.3.4)                    p_2 │              │ p_2
                                  │              │
                                  C ───────────→ C
                                          1_C

is also commutative, one deduces that the diagram

                                       v_{C/B/A}
                                Ω_{C/A} ─────── Ω_{C/B}
                                  ↑                ↑
(20.5.3.5)                   d_{C/A}            d_{C/B}
                                  │                │
                                  C ─────────────→ C
                                          1_C

is commutative.

Finally, if is a second ring homomorphism, one has the transitivity property

(20.5.3.6)                  (v ∘ s)_{C/B/A'} = v_{C/B/A} ∘ s_{C/A/A'}.

(20.5.4) If one now has a commutative diagram of ring homomorphisms

                                  B ───── B'
                                  ↑         ↑
                                  │         │
                                  A ───── A'
                                       u

one deduces from (20.5.2.4) and (20.5.3.3), by composition, a homomorphism of -modules

(20.5.4.1)                          Ω_{B/A} ⊗_B B' → Ω_{B'/A'}

such that the diagram of -homomorphisms

                                  Ω_{B/A} ⊗ 1
                              Ω_{B/A} ──────────→ Ω_{B'/A'}
                                  ↑                  ↑
(20.5.4.2)                  d_{B/A} ⊗ 1            d_{B'/A'}
                                  │                  │
                                  B ────────────────→ B'
                                            1_{B'}

is commutative.

The homomorphism (20.5.4.1) corresponds moreover to a di-homomorphism of -modules

Proposition (20.5.5).

If , are two -algebras and , the canonical homomorphism (20.5.4.1)

(20.5.5.1)                          Ω_{B/A} ⊗_B B' → Ω_{B'/A'}

is bijective.

The left-hand side of (20.5.5.1) is then nothing other than . One may write up to a canonical isomorphism, and is then identified with ; consequently (since is surjective) and , whence up to a canonical isomorphism, which carries the augmentation ideals into themselves; since the -module is canonically identified with the direct sum of and , one has

(20.5.5.2)                          Ω_{B/A} ⊗_A A' ≅ Ω_{B'/A'}

by the same isomorphism, and one verifies at once that the composite of this isomorphism and of the canonical isomorphism is none other than (20.5.5.1).

(20.5.6) The canonical homomorphisms (20.5.2.4) and (20.5.3.3) give, by functoriality, for every -module , canonical homomorphisms

(20.5.6.1)              Hom_C(Ω_{C/A}, L) → Hom_C(Ω_{B/A} ⊗_B C, L) = Hom_B(Ω_{B/A}, L)
(20.5.6.2)              Hom_C(Ω_{C/B}, L) → Hom_C(Ω_{C/A}, L).

Taking (20.4.8.2) and the commutative diagrams (20.5.2.6) and (20.5.3.5) into account, these homomorphisms are none other (up to canonical identification) than the homomorphisms (20.2.1.2) and (20.2.1.3) respectively.

Theorem (20.5.7).

Let , be two ring homomorphisms.

(i) The sequence of -modules

(20.5.7.1)         Ω_{B/A} ⊗_B C →^{u_{C/B/A}} Ω_{C/A} →^{v_{C/B/A}} Ω_{C/B} → 0

is exact.

(ii) For to be left-invertible, it is necessary and sufficient that be a formally smooth -algebra relative to (for the discrete topologies (cf. (19.9.1))); in particular, it suffices for this that be a formally smooth -algebra (for the discrete topologies).

(i) The exactness of the sequence (20.2.4.2) shows first of all, taking (20.5.6) into account, that the sequence

                  0 → Hom_C(Ω_{C/B}, L) → Hom_C(Ω_{C/A}, L) → Hom_B(Ω_{B/A}, L)

is exact for every -module . One knows that this implies the exactness of the sequence (20.5.7.1) (Bourbaki, Alg., chap. II, 3rd ed., §2, n° 1, th. 1).

(ii) By virtue of the exactness of (20.5.7.1), to say that is left-invertible means that the sequence

(20.5.7.2)              0 → Ω_{B/A} ⊗_B C → Ω_{C/A} → Ω_{C/B} → 0

is exact and split; one knows (Bourbaki, loc. cit., n° 1, prop. 1) that this is equivalent to saying that for every -module , the sequence

              0 → Hom_C(Ω_{C/B}, L) → Hom_C(Ω_{C/A}, L) → Hom_B(Ω_{B/A}, L) → 0

is exact; taking (20.5.6) and (20.2.4.2) into account, this condition is equivalent to for every -module , and the conclusion therefore follows from (19.9.8.1).

Let us note moreover that if one has a commutative diagram of ring homomorphisms

                              A' ───→ B' ───→ C'
                                ↑        ↑       ↑
                                │        │       │
                                A ────→ B ────→ C

one deduces a commutative diagram

(20.5.7.3)
        Ω_{B/A} ⊗_B C ────────→ Ω_{C/A} ─────→ Ω_{C/B} ────→ 0
              │                      │              │
              ↓                      ↓              ↓
        Ω_{B'/A'} ⊗_{B'} C' ──→ Ω_{C'/A'} ───→ Ω_{C'/B'} ──→ 0

where the vertical arrows come from the di-homomorphisms (20.5.4.3).

Corollary (20.5.8).

Suppose that the homomorphism makes a formally étale -algebra (for the discrete topologies (19.10.2)); then the homomorphism (20.5.3.3)

                                  u_{C/B/A} : Ω_{B/A} ⊗_B C → Ω_{C/A}

is bijective.

Indeed, if is a formally unramified -algebra for the discrete topologies, it follows from (19.10.4), (20.4.8), and (20.1.1) that one has for every -module , hence (cf. (20.7.4)); on the other hand, if is a formally smooth -algebra for the discrete topologies, the sequence (20.5.7.2) is exact; whence the corollary.

Corollary (20.5.9).

Let be a ring, an -algebra, a multiplicative subset of ; then the canonical homomorphism

(20.5.9.1)                          S^{−1} Ω_{B/A} → Ω_{S^{−1} B / A}

is bijective.

It suffices to apply (20.5.8) to , which is a formally étale -algebra for the discrete topologies (19.10.3, (ii)).

Taking (20.5.5) into account, one may therefore write

(20.5.9.2)                  Ω_{S^{−1} B / S^{−1} A} = S^{−1} Ω_{B/A} = Ω_{S^{−1} B / A},

up to canonical isomorphisms.

Corollary (20.5.10).

If is a field and a purely transcendental extension of , the form a basis of the -vector space .

Since is the field of fractions of the polynomial ring , this follows from (20.4.13, (i)) and from (20.5.9).

(20.5.11) Let be a ring, an -algebra, an ideal of , the quotient -algebra , and consider the composite homomorphism of -modules

(20.5.11.1)                         𝔎 → B →^{d} Ω_{B/A}

where the first arrow is the canonical injection; since , one sees that , whence, by passage to the quotients, a homomorphism of -modules

(20.5.11.2)                 δ_{C/B/A} : 𝔎/𝔎^2 → Ω_{B/A} ⊗_B C = Ω_{B/A} / 𝔎 · Ω_{B/A}.

But in fact, is a homomorphism of -modules, for and , one has , so (mod ), which first proves that (20.5.11.2) is a homomorphism of -modules, and since annihilates both sides, this establishes our assertion.

If is a second -algebra, an -homomorphism, an ideal of such that , and the quotient algebra, one has a commutative diagram

                                         δ
                                 𝔎/𝔎^2 ─────→ Ω_{B/A} ⊗_B C
                                    │                │
(20.5.11.3)                         ↓                ↓
                                 𝔎'/𝔎'^2 ────→ Ω_{B'/A'} ⊗_{B'} C'
                                         δ_{C'/B'/A}

where the vertical arrows come from (20.5.2.4).

Theorem (20.5.12).

Let be a ring, an -algebra, the quotient -algebra , the canonical homomorphism.

(i) One has an exact sequence of -modules

(20.5.12.1)              𝔎/𝔎^2 →^{δ_{C/B/A}} Ω_{B/A} ⊗_B C →^{u_C} Ω_{C/A} → 0

where and are defined by (20.5.2.4) and (20.5.11.2) respectively.

(ii) If one sets , the canonical homomorphism (20.5.2.4)

                                  Ω_{B/A} ⊗_B C → Ω_{E/A} ⊗_E C

is bijective.

(iii) The three following conditions are equivalent:

a) is left-invertible.

b) Every -extension of by a -module , whose inverse image under is -trivial, is itself -trivial.

c) The -algebra is an -trivial extension of by .

(iv) There is a canonical bijective correspondence between the left inverses of and the right inverses of the canonical homomorphism .

(i) Since is surjective, one has for every -module by virtue of (20.1.1). The exact sequence (20.2.3.1) therefore becomes

(20.5.12.2)         0 → Der_A(B, L) →^{∂} Exalcom_B(C, L) →^{u^1} Exalcom_A(C, L) →^{v^1} Exalcom_A(B, L)

where is the homomorphism . Recall on the other hand (18.3.8) that is canonically identified with ; one therefore deduces from (20.5.12.2) and (20.4.8) the exact sequence

(20.5.12.3)         0 → Hom_C(Ω_{C/A}, L) → Hom_C(Ω_{B/A} ⊗_B C, L) →^{φ} Hom_C(𝔎/𝔎^2, L) → Ker(ψ) → 0

with and . Going back to the definitions of (20.2.2) and of (18.3.8), one sees at once that is precisely the homomorphism . The existence of the exact sequence formed by the first four terms of (20.5.12.3) shows therefore that the sequence (20.5.12.1) is exact (Bourbaki, Alg., chap. II, 3rd ed., §2, n° 1, th. 1).

(ii) Apply to and to the ideal the exact sequence (20.5.12.1), which gives

(20.5.12.4)                         𝔎^2/𝔎^4 → Ω_{B/A} ⊗_B E → Ω_{E/A} → 0

whence, tensoring with (considered as -algebra), the exact sequence

                              𝔎^2/𝔎^4 ⊗_E C → Ω_{B/A} ⊗_B C → Ω_{E/A} ⊗_E C → 0.

Now, if , are two elements of , and the class of xy mod , the image is by definition , but since the images of and of in are zero, one also has in , which proves our assertion.

(iii) To say that is left-invertible amounts, taking the exactness of (20.5.12.1) into account, to saying that the sequence

(20.5.12.5)             0 → 𝔎/𝔎^2 →^{δ_{C/B/A}} Ω_{B/A} ⊗_B C →^{u_{C/B/A}} Ω_{C/A} → 0

is exact and split, and it amounts to the same (Bourbaki, loc. cit.) to say that in the exact sequence (20.5.12.3) for every , which shows the equivalence of conditions a) and b) (cf. (18.3.6.2)).

The fact that b) implies c) comes from the fact that the inverse image under of the -extension of by is -trivial, being composed of the canonical homomorphisms (18.1.6). Conversely, c) implies b), for every -extension of by is -equivalent to (18.3.8), in other words its class is the image of the class of (considered as -extension) under the homomorphism

corresponding to a -homomorphism . The fact that c) implies b) then follows from the commutativity of the diagram (18.3.6.5)

                              Exan_B(C, 𝔎/𝔎^2) ─^{w_*}→ Exan_B(C, L)
                                    │                          │
                                    ↓                          ↓
                              Exan_A(C, 𝔎/𝔎^2) ─────────→ Exan_A(C, L)

(iv) One saw (20.1.7) that the right inverses of correspond canonically and bijectively to the set of elements such that on , hence also, by (20.4.8), to the set of -homomorphisms such that the composite is the identity. By tensorization with , one deduces (since is a -module) that the composite

                              𝔎/𝔎^2 →^{d ⊗ 1} Ω_{E/A} ⊗_E C →^{h ⊗ 1} 𝔎/𝔎^2

is the identity; now, since is a -module, is an isomorphism from the set onto ; and on the other hand (ii) proves that one can canonically identify and , being then identified with . Q.E.D.

Example (20.5.13).

Let be a polynomial algebra over , an ideal of , a system of generators of and ; one knows that is a free -module of which the form a basis (20.4.13, (i)), hence the also form a basis of the free -module . On the other hand, it follows at once from the definition that the image of under is the sub--module generated by the

                              dP_λ = ∑ (∂P_λ / ∂X_α) dX_α.
                                     α

One concludes that is isomorphic to the quotient of the free -module having the as basis, by the sub--module generated by the , which gives a description of a module of differentials of an arbitrary algebra, every -algebra being obtainable in the preceding way.

Corollary (20.5.14).

If is a formally smooth -algebra (for the discrete topologies), the sequence

(20.5.14.1)         0 → 𝔎/𝔎^2 →^{δ_{C/B/A}} Ω_{B/A} ⊗_B C →^{u_{C/B/A}} Ω_{C/A} → 0

is exact and split.

In fact, every -extension of by a -module is then trivial (19.4.4.1).

Remark (20.5.15).

Let be a surjective homomorphism of rings; then, for every ring homomorphism , the canonical homomorphism

is bijective; this follows in fact from the exact sequence (20.5.7.1), since (20.4.12).

20.6. Imperfection modules and characteristic homomorphisms

Definition (20.6.1).

Given two ring homomorphisms , , the imperfection module of the -algebra relative to , denoted , is the -module kernel of the homomorphism .

One thus has by definition (cf. (20.5.7)) the exact sequence

  (20.6.1.1)   0 → Υ_{C/B/A} → Ω^1_{B/A} ⊗_B C  ──v_{C/B/A}──▸  Ω^1_{C/A}  ──u_{C/B/A}──▸  Ω^1_{C/B}  → 0.

When (so that the modules and are the "absolute" differential modules and ), we write in place of . When and are algebras over a prime field , one has .

Let , be multiplicative subsets of and respectively, such that the image of is contained in . It then follows from the exact sequence (20.6.1.1) and from (20.5.9) that

  (20.6.1.2)   Υ_{S⁻¹C / R⁻¹B / A}  =  S⁻¹ Υ_{C/B/A}.

Proposition (20.6.2).

If is a -algebra formally smooth relative to (and in particular if is a formally smooth -algebra), then .

This follows from (20.5.7, (ii)).

Proposition (20.6.3).

Let be a field, an extension of . For to be a separable extension of , it is necessary and sufficient that (in other words, that the canonical homomorphism be injective, or equivalently (20.4.8), that every derivation of into extend to a derivation of into itself).

Indeed, it is equivalent to say that is separable over or a formally smooth -algebra (19.6.1). On the other hand, if is the prime field of , then is separable over , hence a formally smooth -algebra, so it amounts to the same to say that is a formally smooth -algebra or a -algebra formally smooth relative to . Finally, to say that is a -algebra formally smooth relative to is equivalent, by (20.5.7, (ii)), to saying that the homomorphism is left-invertible; but since is a field, this last condition is equivalent to saying that the kernel of , that is to say , is zero.

(20.6.4) Consider a commutative diagram

                  u'      v'
            A' ────▸ B' ────▸ C'
            ▴        ▴        ▴
            │f       │g       │h
  (20.6.4.1)│        │        │
            A ────▸  B ────▸  C
                u       v

of homomorphisms of commutative rings. The commutativity of the corresponding diagram (20.5.7.3) entails the existence of a unique -homomorphism

canonically deduced from (20.6.4.1) and making commutative the diagram

                                   v_{C/B/A}             u_{C/B/A}
   0 → Υ_{C/B/A}   →  Ω^1_{B/A} ⊗_B C  ──────────▸  Ω^1_{C/A}  ──────────▸  Ω^1_{C/B}   → 0
       │                │                            │                       │
       ▾                ▾                            ▾                       ▾
  (20.6.4.3)
   0 → Υ_{C'/B'/A'} →  Ω^1_{B'/A'} ⊗_{B'} C'  ──▸   Ω^1_{C'/A'}  ──▸        Ω^1_{C'/B'} → 0
                                       v_{C'/B'/A'}              u_{C'/B'/A'}

The datum of the homomorphism (20.6.4.2) is moreover equivalent to that of a -homomorphism

  (20.6.4.4)   Υ_{C/B/A} ⊗_C C' → Υ_{C'/B'/A'}

which, composed with the canonical homomorphism , recovers (20.6.4.2). It is clear that (20.6.4.2) enjoys an evident transitivity property, allowing one to say that is a functor in the triple .

(20.6.5) It will be convenient for the sequel, under the conditions of (20.6.1), to introduce a (chain) complex of -modules whose terms vanish except in degrees 0 and 1, where we take

                          K_0(C/B/A) = Ω^1_{C/A}
  (20.6.5.1)
                          K_1(C/B/A) = Ω^1_{B/A} ⊗_B C

the differential being . This permits one to write (up to canonical isomorphisms) and as the homology modules of this complex:

  (20.6.5.2)   H_0(K_•(C/B/A)) = Ω^1_{C/B},        H_1(K_•(C/B/A)) = Υ_{C/B/A}.

Likewise:

Proposition (20.6.6).

Under the hypotheses of (20.6.1), for every -module one has canonical -isomorphisms

  (20.6.6.1)   H^0(K_•(C/B/A), L) ≅ Der_B(C, L)
  (20.6.6.2)   H^1(K_•(C/B/A), L) ≅ Exalcom_{B/A}(C, L).

Indeed, the cochain complex is none other, by virtue of (20.4.8) and (20.5.6), than the complex

  … → 0 → Der_A(C, L) → Der_A(B, L) → 0 → …

where the differential is (with the notations of (20.2.1)). The proposition then follows from the exact sequence (20.2.4.2) and the definition of the cohomology modules

  (20.6.6.3)   H^i(K_•, L) = H^i(Hom_C(K_•, L)).

If one has a commutative diagram of ring homomorphisms

   A' ────▸ B' ────▸ C'
   ▴        ▴        ▴
   │        │        │
   A ────▸  B ────▸  C

the di-homomorphisms (20.5.4.3) define a di-homomorphism of complexes of modules

and the di-homomorphisms one deduces for homology or cohomology are identified, via the formulas (20.6.5.2), (20.6.6.1) and (20.6.6.2), with the di-homomorphisms already defined in (20.5.4.3), (20.6.4.2), (20.2.1) and (18.3.7.2) (taking into account, for the last, the definition of the operator in the exact sequence (20.2.4.2)).

(20.6.7) It is known that for a complex of -modules and a -module , one has canonical homomorphisms

  α_i : H^i(K_•, L) → Hom_C(H_i(K_•), L)

(M, IV, 6). Here, the canonical homomorphism

  α_1 : H^1(K_•(C/B/A), L) → Hom_C(H_1(K_•(C/B/A)), L)

is defined immediately as obtained by passage to the quotient by the image of of the restriction homomorphism

  Hom_C(K_1(C/B/A), L) → Hom_C(H_1(K_•(C/B/A)), L),

since is none other than the kernel of ; taking (20.6.6.2) and (20.6.5.2) into account, one therefore obtains a canonical -homomorphism

  (20.6.7.1)   Exalcom_{B/A}(C, L) → Hom_C(Υ_{C/B/A}, L)

which is made explicit as follows: by virtue of (20.2.4.2), every -extension of by that is -trivial comes from the datum of an -derivation of into , hence (20.4.8) from a -homomorphism of into ; one associates to the class of this extension the restriction of to , which depends only on the class of the extension and not on the choice of .

Definition (20.6.8).

Let , be two ring homomorphisms. For every -extension of by a -module that is -trivial (18.3.7), we call characteristic homomorphism of , and we denote , the -homomorphism , image of the class of under the canonical homomorphism (20.6.7.1).

One can define the homomorphism in another way:

Proposition (20.6.9).

Let be a -extension of by a -module , which is -trivial (18.3.7); then the diagram

                                                              v_{C/B/A}
  0 ──▸ Υ_{C/B/A} ────▸  Ω^1_{B/A} ⊗_B C    ────────────▸    Ω^1_{C/A}
        │                  │                                  │
        │ χ_E              │ q_{E/B/A} ⊗ 1_C                  │ ≅
        ▾                  ▾                                  ▾
  (20.6.9.1)
  0 ────▸    L    ────▸  Ω^1_{E/A} ⊗_E C    ────────────▸    Ω^1_{C/A}  ──▸  0
                       δ_{C/E/A}              p_{C/E/A}

where defines the structure of -extension on and is the augmentation homomorphism, is commutative and its rows are exact.

The lower row of the diagram is the sequence (20.5.12.5) relative to the two homomorphisms and ; since is surjective and is an -trivial extension, this sequence is exact and split by virtue of (20.5.12, (iii)). The commutativity of the right-hand square of (20.6.9.1) follows from the relation (20.5.2.7); the image under of the kernel of is therefore contained in the kernel of . On the other hand, let be an -homomorphism right inverse to , and let be the canonical injection, so that one has for , where is the -derivation of into defining the -extension ; one can write , where is a -homomorphism. By virtue of (20.5.2.6), one has

  q_{E/B/A}(d_{B/A}(b)) = d_{E/A}(q(b)) = d_{E/A}(h(v(b))) + d_{E/A}(j(f(d_{B/A}(b))))   for b ∈ B.

Let then , where and , be an element of ; one has

  (q_{E/B/A} ⊗ 1_C)(z) = ∑_i d_{E/A}(h(v(b_i))) ⊗ c_i + ∑_i d_{E/A}(j(f(d_{B/A}(b_i)))) ⊗ c_i.

In the first sum, one has , hence this sum is by virtue of (20.5.2.6). If one takes , this sum is therefore zero, and there remains, by definition of ,

  (q_{E/B/A} ⊗ 1_C)(z) = ∑_i δ_{C/E/A}(c_i · f(d_{B/A}(b_i))) = δ_{C/E/A}(f(z))

which proves the commutativity of the left-hand square in (20.6.9.1).

One will note that when one supposes only that is injective (and not necessarily left-invertible), this interpretation would still permit defining as the restriction of to .

Corollary (20.6.10).

If is an ideal of , , , and if is an -trivial -extension (18.3.7) of by , then the characteristic homomorphism is bijective.

Indeed, in the diagram (20.6.9.1), the two right-hand vertical arrows are bijective homomorphisms (20.5.12, (ii)).

Theorem (20.6.11).

Let , be two ring homomorphisms, a -module. Suppose one of the following conditions holds:

(i) is an injective -module.

(ii) is a direct factor of the -module , and is right-invertible.

Then the canonical homomorphism (20.6.7.1)

  Exalcom_{B/A}(C, L) → Hom_C(Υ_{C/B/A}, L)

is bijective.

In particular, if and are projective -modules, the canonical homomorphism (20.6.7.1) is bijective.

The fact that each of conditions (i), (ii) entails that (20.6.7.1) is bijective follows in both cases from the definition of . One will note moreover that condition (ii) is necessary and sufficient for the homomorphism (20.6.7.1) to be bijective for every -module (Bourbaki, Alg., chap. II, 3rd ed., §2, n° 1, prop. 1). If one supposes that and are projective -modules, then, in the exact sequence (20.6.1.1), is a projective -module, since the exact sequence

is split, being projective; since , the exact sequence

  0 → Υ_{C/B/A} → Ω^1_{B/A} ⊗_B C → Im(v_{C/B/A}) → 0

is split.

Corollary (20.6.12).

Suppose that is a formally smooth -algebra. Then there exists a canonical homomorphism

  (20.6.12.1)   Exalcom_B(C, L) → Hom_C(Υ_{C/B/A}, L).

Moreover, this homomorphism is bijective if one of conditions (i), (ii) of (20.6.11) is satisfied.

Indeed, it follows from the hypothesis on that , and the homomorphism (20.6.12.1) is none other than (20.6.7.1).

Corollary (20.6.13).

If is a formally smooth -algebra and if is a projective -module, the homomorphism (20.6.12.1) is bijective.

Indeed, one knows then that is a projective -module (20.4.9), the topologies being discrete.

(20.6.14) The notations remaining the same, suppose now in addition that , , are -algebras and , are -homomorphisms, which amounts to giving three ring homomorphisms

  Λ ──s──▸ A ──u──▸ B ──v──▸ C.

One thus has, beyond the imperfection module , the imperfection modules , and , and one has already defined canonical homomorphisms of -modules (20.6.4.2)

As in the commutative diagram (20.5.7.3)

                       Ω^1_{A/Λ} ⊗_A B  →  Ω^1_{B/Λ}  →  Ω^1_{B/A}  → 0
  (20.6.14.3)         │                   │              │
                       ▾                   ▾              ▾
                       Ω^1_{A/Λ} ⊗_A C  →  Ω^1_{C/Λ}  →  Ω^1_{C/A}  → 0

the lower row is formed of -modules; one deduces from it by tensoring a commutative diagram

                       Ω^1_{A/Λ} ⊗_A C  →  Ω^1_{B/Λ} ⊗_B C  →  Ω^1_{B/A} ⊗_B C  → 0
  (20.6.14.4)         │ =                 │                    │
                       ▾                   ▾                    ▾
                       Ω^1_{A/Λ} ⊗_A C  →  Ω^1_{C/Λ}        →  Ω^1_{C/A}        → 0

where the first row is again exact and the left vertical arrow is the identity. If one sets

  (20.6.14.5)   Υ^C_{B/A/Λ} = Ker(u_{B/A/Λ} ⊗ 1_C) = Ker(Ω^1_{A/Λ} ⊗_A C → Ω^1_{B/Λ} ⊗_B C),

one sees, taking into account the definition of , that one has a unique -homomorphism

rendering commutative the diagram

   0 → Υ^C_{B/A/Λ} → Ω^1_{A/Λ} ⊗_A C →  Ω^1_{B/Λ} ⊗_B C →  Ω^1_{B/A} ⊗_B C → 0
       │ v'           │ =                │                 │
       ▾              ▾                  ▾                 ▾
  (20.6.14.7)
   0 → Υ_{C/A/Λ}   → Ω^1_{A/Λ} ⊗_A C →  Ω^1_{C/Λ}      →  Ω^1_{C/A}        → 0

whose rows are exact.

When , we shall write in place of . If is a prime field, one has .

(20.6.15) To study the relations between the preceding modules, we shall introduce on the one hand the complex of -modules , on the other hand the complexes of -modules and (20.6.5), and in addition the following complexes of -modules. We set first of all

  (20.6.15.1)   K^C_•(B/A/Λ) = K_•(B/A/Λ) ⊗_B C.

On the other hand, we shall denote by the complex of -modules whose terms vanish except in degrees 0 and 1, where

  (20.6.15.2)   T_0(C/B/A) = T_1(C/B/A) = Ω^1_{B/A} ⊗_B C,

the differential being the identity, so that this complex is homotopic to 0; we set finally

By virtue of the trivial character of , it is clear that one has

  (20.6.15.4)   H^i(K'_•, L) ≅ H^i(K_•, L)   and   H_i(K'_•, L) ≅ H_i(K_•, L)

for every -module and every .

(20.6.16) Let us now define an exact sequence of complexes, split in each degree

  (20.6.16.1)   0 → K^C_•(B/A/Λ) ──j──▸ K'_•(C/A/Λ) ──p──▸ K_•(C/B/A) → 0

as follows: let us denote for a moment by

  f : Ω^1_{A/Λ} ⊗_A C → Ω^1_{B/Λ} ⊗_B C
  g : Ω^1_{B/Λ} ⊗_B C → Ω^1_{C/Λ}

the canonical homomorphisms and respectively, whose composite is (cf. (20.6.14.4)). One takes , , , , so that is the graph of , complementary to , and is the graph of , complementary to ; the verification of the commutativity of the diagram

   0 → K^C_1(B/A/Λ) ──j_1──▸ K'_1(C/A/Λ) ──p_1──▸ K_1(C/B/A) → 0
       │                     │                   │
       ▾                     ▾                   ▾
   0 → K^C_0(B/A/Λ) ──j_0──▸ K'_0(C/A/Λ) ──p_0──▸ K_0(C/B/A) → 0

where the vertical arrows are the differentials, is immediate.

Theorem (20.6.17).

One has an exact sequence of -modules

                                     v'             u'             ∂                v_{C/B/A}        u_{C/B/A}
  (20.6.17.1)   0 → Υ^C_{B/A/Λ}  ──────▸  Υ_{C/A/Λ}  ──────▸  Υ_{C/B/A}  ────▸  Ω^1_{B/A} ⊗_B C  ──────────▸  Ω^1_{C/A}  ──────────▸  Ω^1_{C/B}  → 0

where the boundary operator is the composite

  (20.6.17.2)   Υ_{C/B/A} ──s'──▸ Υ_{C/B/Λ} ────▸ Ω^1_{B/Λ} ⊗_B C

the second arrow being the canonical injection.

By writing the exact sequence of homology for the exact sequence of complexes (20.6.16.1), one obtains (20.6.17.1), the homology being zero in degrees other than 0 and 1; the fact that the homomorphisms of this exact sequence that come by functoriality from and are indeed those of the statement is immediate. It remains to verify that is equal to (20.6.17.2); now an element is the image under of (0, z), whence one deduces at once that is the image of under the canonical homomorphism . Our assertion follows from the commutativity of the diagram

                Υ_{C/B/A} ──s'──▸ Υ_{C/B/Λ}
                  │                │
  (20.6.17.3)     ▾                ▾
                Ω^1_{B/A} ⊗_B C → Ω^1_{B/Λ} ⊗_B C

(cf. (20.6.4.3)).

Corollary (20.6.18).

(i) The sequence of -modules

                              v'              u'             s'
  (20.6.18.1)   0 → Υ^C_{B/A/Λ}  ──▸  Υ_{C/A/Λ}  ──▸  Υ_{C/B/Λ}  ──▸  Υ_{C/B/A} → 0

is exact.

(ii) If is a -algebra formally smooth relative to , one has .

(i) In (20.6.17.1), the image of under is the kernel of , hence by definition.

(ii) The hypothesis entails that the sequence

  0 → Ω^1_{A/Λ} ⊗_A B → Ω^1_{B/Λ} → Ω^1_{B/A} → 0

is exact and split (20.5.7); by tensoring with , the sequence

  0 → Ω^1_{A/Λ} ⊗_A C → Ω^1_{B/Λ} ⊗_B C → Ω^1_{B/A} ⊗_B C → 0

therefore remains exact, whence our assertion.

Corollary (20.6.19).

Let be a field, , two extensions of such that .

(i) If is a separable extension of , one has (in other words, the canonical homomorphism is injective).

(ii) Conversely, if is a separable extension of and one has , then is a separable extension of .

If is the prime field of , one has by (20.6.18) the exact sequence

If , then , whence (i) by virtue of (20.6.3); conversely, if and , one has , whence (ii) by virtue of (20.6.3).

Corollary (20.6.20).

(i) If is a separable algebraic extension of , one has .

(ii) Let be a field of characteristic 0, an extension of . For , it is necessary and sufficient that be an algebraic extension of . In particular, for a field of characteristic 0 to be such that , it is necessary and sufficient that be an algebraic extension of (cf. (21.4.4) and (21.7.5)).

(i) For every , let be the minimal polynomial of over ; since and , one has , and our assertion follows from (20.4.7).

(ii) There exists a pure extension of such that and such that is an algebraic extension of . Since is separable over , it follows from (20.6.19, (i)) that the sequence (20.5.7.2)

  0 → Ω^1_{L/k} ⊗_L K → Ω^1_{K/k} → Ω^1_{K/L} → 0

is exact, and from (i) that . The relation is therefore equivalent to , and since is a pure extension of , it follows from (20.5.10) that the relation is equivalent to .

Remarks (20.6.21).

(i) Since is the kernel of and is the kernel of , one has a canonical homomorphism

  (20.6.21.1)   Υ_{B/A/Λ} ⊗_B C → Υ^C_{B/A/Λ}.

This homomorphism is bijective when the sequence

  (20.6.21.2)   0 → Υ_{B/A/Λ} ⊗_B C → Ω^1_{A/Λ} ⊗_A C → Ω^1_{B/Λ} ⊗_B C → Ω^1_{B/A} ⊗_B C → 0

is exact, which occurs in the following cases:

is a flat -module.

2° The -modules and are flat; for then so is , and the sequence (20.6.21.2) is then exact by virtue of .

(ii) Consider a commutative diagram of ring homomorphisms

  Λ' ────▸ A' ────▸ B' ────▸ C'
  ▴        ▴        ▴        ▴
  │        │        │        │
  Λ  ────▸ A  ────▸ B  ────▸ C

Then the definitions of (20.6.16) show that one has a commutative diagram of complexes (where the vertical arrows come from (20.6.6.4))

   0 → K^C_•(B/A/Λ) → K'_•(C/A/Λ) → K_•(C/B/A) → 0
       │              │              │
       ▾              ▾              ▾
   0 → K^{C'}_•(B'/A'/Λ') → K'_•(C'/A'/Λ') → K_•(C'/B'/A')

whence, by passage to homology, a commutative diagram

  (20.6.21.3)
   0 → Υ^C_{B/A/Λ}      → Υ_{C/A/Λ}      → Υ_{C/B/A}      → Ω^1_{B/A} ⊗_B C    → Ω^1_{C/A}    → Ω^1_{C/B}    → 0
       │                  │                │                │                    │              │
       ▾                  ▾                ▾                ▾                    ▾              ▾
   0 → Υ^{C'}_{B'/A'/Λ'} → Υ_{C'/A'/Λ'}  → Υ_{C'/B'/A'}  → Ω^1_{B'/A'} ⊗_{B'} C' → Ω^1_{C'/A'} → Ω^1_{C'/B'} → 0

One has an analogous commutative diagram for (20.6.18.1).

Proposition (20.6.22).

Let , be two ring homomorphisms, an ideal of , the quotient ring . Suppose that is an -trivial -extension of by . Then one has the exact sequence

                                    v'           χ_E             δ_{C/B/A}                  v_{C/B/A}
  (20.6.22.1)   0 → Υ^C_{B/A/Λ}  ──▸ Υ_{C/A/Λ} ──▸ 𝔍/𝔍²  ──▸  Ω^1_{B/A} ⊗_B C  ─────────▸  Ω^1_{C/A}  → 0.

Indeed, since is surjective, one has (20.4.12). Furthermore, it follows from (20.6.10) that is canonically identified with . It then suffices to apply the exact sequences (20.6.17.1) and (20.6.18.1), noting that one has a commutative diagram

                Υ_{C/A/Λ}                ──▸ Ω^1_{A/Λ} ⊗_A C
                  │                            │
                  ▾ χ_E                        ▾
   Υ_{C/B/Λ} = Υ_{C/B/A} = 𝔍/𝔍²  ──δ_{C/B/A}──▸  Ω^1_{B/A} ⊗_B C = Ω^1_{E/A} ⊗_E C
                  │                            │
                  ▾                            ▾
                𝔍/𝔍²              ──δ_{C/B/A}──▸  Ω^1_{B/A} ⊗_B C

and using the commutativity of the diagram (20.6.17.3).

One has thus specified the kernel of in the case where there exists a ring such that is a -algebra and is -trivial; one observes that this is the case in particular when is a formally smooth -algebra (for the discrete topology).

Suppose in addition that one has a commutative diagram of ring homomorphisms

  Λ' ────▸ A' ────▸ B'
  ▴        ▴        ▴
  │        │        │ f
  Λ  ────▸ A  ────▸ B

such that is an ideal of with , and is an -trivial -extension of by . One then has a commutative diagram

  (20.6.22.2)
   0 → Υ^{C'}_{B'/A'/Λ'} → Υ_{C'/A'/Λ'} → 𝔍'/𝔍'² → Ω^1_{B'/A'} ⊗_{B'} C' → Ω^1_{C'/A'} → 0
       ▴                   ▴               ▴       ▴                       ▴
       │                   │               │       │                       │
   0 → Υ^C_{B/A/Λ}        → Υ_{C/A/Λ}    → 𝔍/𝔍²   → Ω^1_{B/A} ⊗_B C      → Ω^1_{C/A}    → 0

as follows from (20.6.21.3) and (20.5.11.3).

Corollary (20.6.23).

Under the hypotheses of (20.6.22), suppose in addition that is a formally smooth -algebra. Then one has an exact sequence

                              χ_E           δ_{C/B/A}                  v_{C/B/A}
  (20.6.23.1)   0 → Υ_{C/A/Λ}  ──▸  𝔍/𝔍²  ──▸  Ω^1_{B/A} ⊗_B C  ──▸  Ω^1_{C/A} → 0.

This indeed follows from (20.6.18, (ii)).

(20.6.24) When the hypotheses of (20.6.22) are satisfied, one says again that the characteristic homomorphism is the characteristic homomorphism of the -algebra , relative to and to the ideal (suppressing these last specifications when there is no risk of confusion); it will sometimes be denoted or .

Proposition (20.6.25).

Let , , be three ring homomorphisms, a -module. One then has an exact sequence

                                  u⁰              v⁰              ∂
  (20.6.25.1)   0 → Der_B(C, L)  ──▸  Der_A(C, L)  ──▸  Der_Λ(B, L)  ──▸
                          ──▸  Exalcom_{B/A}(C, L)  ──u¹──▸  Exalcom_{A/Λ}(C, L)  ──v¹──▸  Exalcom_{A/Λ}(B, L)  → 0

where , are the homomorphisms defined in (18.3.6.4) and (18.3.6.2), and is defined as in (20.2.2).

Indeed, since the exact sequence (20.6.16.1) is split, one deduces an exact sequence

  0 → Hom_C(K_•(C/B/A), L) → Hom_C(K'_•(C/A/Λ), L) → Hom_C(K^C_•(B/A/Λ), L) → 0.

If one applies to this complex the exact sequence of cohomology, taking into account (20.6.15.4) and (20.6.6), one obtains (20.6.25.1), since one has

  Hom_C(K^C_•(B/A/Λ), L) = Hom_B(K_•(B/A/Λ), L)

by definition; the identification of and with the homomorphisms of (18.3.4.2) follows from (20.6.6.4).

Remark (20.6.26).

*In this number, the complexes have appeared as a technical artifice destined to simplify the exposition of certain functorial behaviours. In reality, these complexes, considered as objects of the category of complexes of

-modules "up to homotopy" (that is, where the morphisms are the classes of homotopic homomorphisms), are remarkable invariants, finer than the pair formed of and . When is a prime field, and is a formally smooth -algebra (for example a regular ring of finite type over an extension of (cf. (IV, 6.8.6))), one can show that the complex can be described uniquely in terms of and (to the exclusion of ): one expresses as the quotient of a polynomial algebra over by an ideal , and one considers the complex with two non-zero terms*

  … → 0 → 𝔏/𝔏² → Ω^1_{B/A} ⊗_B C → 0 → …

(whose homology coincides indeed with that of by virtue of (20.6.23.1)). These complexes , which from the point of view of homological algebra play the role of a conormal bundle for over , will occupy an important place in the chapters of this work devoted to the duality of coherent sheaves and to the Riemann-Roch theorem.

20.7. Generalizations to topological rings

(20.7.1) It follows at once from the definitions that if, in (20.5.2) and (20.5.3), the rings , , are supposed to be topological rings and the ring homomorphisms , continuous, then the homomorphisms and are continuous (one must naturally take on the tensor product topology).

Moreover, is a strict surjective morphism of topological -modules; for it is thus for the canonical homomorphism , taking into account the definition of the tensor product topology; by passage to the quotients one deduces that the corresponding homomorphism is a strict surjective morphism, and is the restriction of this last to .

In (20.5.5), if and are topological -algebras, and if and are equipped with the tensor product topologies, the canonical homomorphism (20.5.5.1) is a topological isomorphism, taking into account the fact that is the topological direct sum of and (20.4.8).

Proposition (20.7.2).

Let , be two continuous homomorphisms of topological rings. For the continuous homomorphism to be formally left-invertible (cf. (19.1.5)), it is necessary and sufficient that be a -algebra formally smooth relative to (19.9.1) (and a fortiori it suffices that be a formally smooth -algebra).

To say that is formally left-invertible signifies indeed, since the topologies of and are coarser than those deduced from the topology of (20.4.5), that for every discrete -module , annihilated by an open ideal of , the canonical homomorphism

  Hom.cont_C(Ω^1_{C/A}, L) → Hom.cont_C(Ω^1_{B/A} ⊗_B C, L)

is surjective (19.1.5); since Hom.cont_C(Ω^1_{B/A} ⊗_B C, L) = Hom.cont_C(Ω^1_{B/A}, L) by definition of the tensor product topology, it amounts to the same, by virtue of (20.4.8), to say that the canonical homomorphism

  Der.cont_A(C, L) → Der.cont_A(B, L)

is surjective. But the exact sequence (20.3.8.2) shows that this condition is equivalent to , that is to say precisely to the fact that is formally smooth relative to (19.9.8).

Corollary (20.7.3).

Suppose that in and in , the square of an open ideal is open. For to be a -algebra formally smooth relative to , it is necessary and sufficient that, if one denotes by a fundamental system of neighbourhoods of 0 formed of ideals of , then, for every , the homomorphism

  (20.7.3.1)   v_{C/B/A} ⊗ 1_{C/𝔎_λ} : Ω^1_{B/A} ⊗_B (C/𝔎_λ) → Ω^1_{C/A} ⊗_C (C/𝔎_λ)

be left-invertible.

One knows indeed in that case that the topology of (resp. ) is deduced from that of (resp. of ) (20.4.5); one concludes at once that the topology of is also deduced from that of ; the corollary then follows from (20.7.2) and (19.1.7).

Proposition (20.7.4).

Let be a topological ring, a topological -algebra. For to be formally unramified (19.10.2), it is necessary and sufficient that the separated completion be zero.

Indeed, it follows at once from (19.10.4) and (20.1.1) that, for to be formally unramified, it is necessary and sufficient that for every open ideal of , every open ideal of such that , and every -module , one have , that is to say for every discrete -module annihilated by an open ideal of ; by virtue of (20.4.8), this is equivalent to for such a -module, whence at once the proposition.

When is discrete, the condition in the statement of (20.7.4) is therefore equivalent to .

Corollary (20.7.5).

Let be a ring, an ideal of , an -algebra; equip with the -preadic topology, with the topology deduced from that of , and set , . Then, for to be a formally unramified -algebra, it is necessary and sufficient that (or, equivalently, that B_0 be a formally unramified A_0-algebra).

Indeed, one has by virtue of (20.5.5); to write that every open submodule of is equal to is equivalent, by (20.4.5), to writing , whence the conclusion.

Proposition (20.7.6).

Let , be two continuous homomorphisms of topological rings, and suppose that makes into a formally étale -algebra; then is a formal bimorphism (19.1.2).

Indeed, for every discrete -module annihilated by an open ideal of , one then has (20.7.4) and consequently (20.3.6.1) the canonical homomorphism

Der.cont_A(C, L) → Der.cont_A(B, L) is injective; it is moreover surjective by virtue of (20.7.2), hence bijective. It follows that the image of is necessarily dense in , otherwise the quotient of by the closure of would be separated and and would therefore have a discrete quotient , contrary to what we have just seen (taking (20.4.8) into account). Consequently , which is a formal monomorphism by virtue of (20.7.2), is also a formal epimorphism (19.1.2), hence a formal bimorphism.

Corollary (20.7.7).

Suppose that in and in the square of every open ideal is open. If is a formally étale -algebra, then, for every open ideal of , the homomorphism (20.7.3.1) is bijective.

Indeed, by virtue of (19.1.1), this homomorphism is surjective, and it is injective by (20.7.3).

Proposition (20.7.8).

Let be a continuous homomorphism of topological rings, an ideal of , the quotient topological ring , the canonical homomorphism. Then:

(i) In the exact sequence (20.5.12.1)

                    δ_{C/B/A}                v_{C/B/A}
  𝔍/𝔍²  ────────────▸  Ω^1_{B/A} ⊗_B C  ────────────▸  Ω^1_{C/A} → 0

the homomorphism is continuous and the homomorphism is a strict morphism of topological -modules.

(ii) For to be formally left-invertible (19.1.5), it is necessary and sufficient that for every discrete -module annihilated by an open ideal of , the canonical homomorphism

  (20.7.8.1)   Exalcotop_A(C, L) → Exalcotop_A(B, L)

be injective.

(i) The first assertion is evident. On the other hand, the canonical homomorphism is a strict morphism by definition of the tensor product topology, and one deduces at once (cf. (20.5.2)) that the same holds of .

(ii) To say that is formally left-invertible signifies that for every discrete -module annihilated by an open ideal of , the canonical homomorphism

  Hom.cont_C(Ω^1_{B/A} ⊗_B C, L) → Hom.cont_C(𝔍/𝔍², L)

is surjective. Now, taking into account (18.4.3) and (20.4.8), this amounts to saying that the canonical homomorphism

  Der.cont_A(B, L) → Exalcotop_B(C, L)

is surjective, and the conclusion therefore follows from the exact sequence (20.3.6.1).

The fact that (20.7.8.1) be injective is also expressed in the following manner, taking into account the definition of the two members (18.4.1): for every open ideal of , every open ideal of such that , and every -extension of by a -module , such that the inverse image of under the canonical homomorphism be -trivial, there exists an open ideal of ,

an open ideal of such that and such that the inverse image of under the canonical homomorphism be -trivial. In particular:

Corollary (20.7.9).

If the topological -algebra is formally smooth, the canonical homomorphism is formally left-invertible.

(20.7.10) In (20.6.1), when , , are topological rings and , continuous homomorphisms, one equips with the topology induced by that of ; the homomorphisms (20.6.4.2) and (20.6.4.4) are then continuous, provided the same holds for those of the diagram (20.6.4.1). Moreover, if, in (20.6.7), one assumes that is a discrete -module annihilated by an open ideal of , one deduces, by passage to the inductive limit, a canonical -homomorphism

  (20.7.10.1)   Exalcotop_{B/A}(C, L) → Hom.cont_C(Υ_{C/B/A}, L)

taking into account (18.5.3.1): for every open ideal of , every open ideal of such that , every open ideal of such that and , every -extension of by that is -trivial comes from the datum of a continuous -homomorphism of into , and the homomorphism (20.7.10.1) associates to the image in of the class of , the restriction of to , the characteristic homomorphism of , again denoted .

Proposition (20.7.11).

Suppose that the topology of is such that the square of an open ideal is open. If is a topological -algebra that is formally smooth and if is a formally projective -module, one has a canonical isomorphism

  (20.7.11.1)   Exalcotop_B(C, L) ≅ Hom.cont_C(Υ_{C/B/A}, L)

for every discrete -module annihilated by an open ideal of .

Indeed, in the exact sequence (20.3.7.1), one has (19.4.4), hence and the homomorphism (20.7.8.1) is none other than (20.7.10.1); the fact that it is bijective is deduced from (20.6.13) by passage to the inductive limit, taking into account that the topology of is then deduced from that of (20.4.5) and (19.2.4).

(20.7.12) In (20.6.14) one again equips with the topology induced by that of , and then the homomorphism (20.6.14.6) is continuous, when the rings considered are topological and the ring homomorphisms continuous.

(20.7.13) If, in (20.6.23), one supposes that the rings , , , are topological, the homomorphisms , , continuous and a discrete -module annihilated by an open ideal of , then one may pass to the inductive limit as in (20.3.6), and one has an exact sequence

  (20.7.13.1)   0 → Der.cont_B(C, L) → Der.cont_A(C, L) → Der.cont_Λ(B, L) →
                  → Exalcotop_{B/A}(C, L) → Exalcotop_{A/Λ}(C, L) → Exalcotop_{A/Λ}(B, L) → 0.

(20.7.14) Let be a topological ring, a topological -algebra, an open ideal of , an open ideal of such that ; set ,

; the kernel of the homomorphism is , from which it follows at once that the kernel of the homomorphism

  (20.7.14.1)   φ_{(𝔐', 𝔑')} : Ω^1_{B/A} → Ω^1_{B'/A'}

is ; on the other hand, the homomorphism (20.7.14.1) is surjective, as follows from (20.4.7). If (resp. ) is a second open ideal of (resp. ) such that , and , and if one sets , , one has likewise a surjective homomorphism

  φ_{(𝔐', 𝔑'), (𝔐'', 𝔑'')} : Ω^1_{B''/A''} → Ω^1_{B'/A'}

and these homomorphisms obviously form a projective system. If one remarks that the form a fundamental system of neighbourhoods of 0 in , one sees therefore that the separated completion of the topological -module is given, up to a canonical isomorphism, by

Moreover, the canonical homomorphism is the projective limit of the projective system of the , hence factors as , and since it is surjective one concludes that the canonical homomorphism

is surjective for every pair .

One may moreover in the foregoing replace everywhere by .

Finally, if is a separated and complete topological -module, every continuous -homomorphism of into extends in a unique way to a continuous -homomorphism of into , and conversely such a homomorphism gives back by composition with a continuous -homomorphism, so that one has a canonical isomorphism

  Hom.cont_{B̂}(Ω̂^1_{B/A}, L) ≅ Hom.cont_B(Ω^1_{B/A}, L).

More particularly, if is a discrete -module annihilated by an open ideal of , one sees that the canonical isomorphism (20.4.8.2) may also be written

  (20.7.14.4)   Hom.cont_B(Ω̂^1_{B/A}, L) ≅ Der.cont_A(B, L).

Proposition (20.7.15).

Let be a discrete ring, a topological adic -algebra , an ideal of definition of , . Suppose that and are B_0-modules of finite type. Then is a -module of finite type.

Since the square of every open ideal of is open, the topology of is the -preadic topology (20.4.5). Taking into account the hypothesis that is an adic ring, it therefore suffices, by virtue of , to prove

that is a B_0-module of finite type. But this follows from the hypothesis and from the exact sequence (20.5.12.1)

  𝔫/𝔫² → Ω^1_{B/A} ⊗_B B_0 → Ω^1_{B_0/A} → 0.

(20.7.16) The proposition (20.7.15) applies for example when is a field , a formal power series algebra equipped with its usual topology, a finite extension of (cf. (21.9.2)). One will note that the field of fractions of has infinite transcendence degree over ; when is of characteristic 0, one deduces at once (with the help of (20.4.13, (i)) and of (20.5.9) in particular, also using the fact that every derivation of a field of characteristic 0 extends to every extension) that is not a -module of finite type.

(20.7.17) Let , , be three topological rings, , two continuous homomorphisms; replacing , , by quotients by open ideals , , with , , so that one has homomorphisms , , one deduces canonical homomorphisms , which, by virtue of (20.5.4), form projective systems, and give in consequence, by passage to the limit, canonical homomorphisms, extensions to the separated completions of the homomorphisms of the exact sequence (20.5.7.1)

                              v_{C/B/A}             u_{C/B/A}
                Ω^1_{B/A} ⊗_B C  ────────▸  Ω^1_{C/A}  ────────▸  Ω^1_{C/B}  → 0

  (20.7.17.1)   v̂_{C/B/A} : Ω̂^1_{B/A} ⊗̂_{B̂} Ĉ → Ω̂^1_{C/A}

  (20.7.17.2)   û_{C/B/A} : Ω̂^1_{C/A} → Ω̂^1_{C/B}

and in the sequence

                              v̂_{C/B/A}             û_{C/B/A}
  (20.7.17.3)   Ω̂^1_{B/A} ⊗̂_{B̂} Ĉ  ────────▸  Ω̂^1_{C/A}  ────────▸  Ω̂^1_{C/B}  → 0

the composite of two consecutive homomorphisms is 0, but the sequence is not necessarily exact. However, if and are metrizable, the homomorphism is surjective, and is dense in : this follows at once (cf. ) from the fact that, if (resp. ) is a decreasing sequence of ideals of (resp. ) forming a fundamental system of neighbourhoods of 0, and if one sets , , the transition homomorphisms , and are surjective (20.7.14).

Proposition (20.7.18).

Let , , be three topological rings, , two continuous homomorphisms. Suppose and admissible and metrizable. For the canonical homomorphism to admit a left inverse that is a continuous -homomorphism, it is necessary and sufficient that be a -algebra formally smooth relative to .

The condition is necessary by virtue of (20.7.2) and (19.1.6). To see that it is sufficient, note that the topological -module is metrizable and complete by virtue of the hypothesis, hence , equipped with the product topology, is

metrizable and complete; it is moreover admissible, for if is an ideal of definition in , the sequence of tends to 0. Since the composite application

  D : B  ──d_{B/A}──▸  Ω^1_{B/A}  ──▸  Ω̂^1_{B/A} ⊗̂_{B̂} Ĉ = L

is a continuous -derivation of into , the continuous -homomorphism

  f : x ↦ (v(x), D(x))

of into defines on a structure of -extension. Since is a closed ideal in , it follows from (19.9.5) and from the hypothesis that the identity application (which is a -homomorphism) factors as , where is a continuous homomorphism such that ; consequently (20.1.3), is of the form , where is a continuous -derivation of into , in other words . Taking into account (20.7.14.4), one has , where is a continuous -homomorphism; but one has by definition, and since the image of under generates (topologically) the -module (20.4.7), the relation gives indeed . Q.E.D.

Corollary (20.7.19).

Under the hypotheses of (20.7.18), if one supposes in addition that in and the square of an open ideal is open, then, for to be a -algebra formally smooth relative to , it is necessary and sufficient that be left-invertible.

Indeed, the topologies of and of are then deduced from that of (20.4.5), and every -homomorphism of one into the other is necessarily continuous.

(20.7.20) Let be a topological ring, a topological -algebra, metrizable and complete, a closed ideal of , the quotient topological ring, which is metrizable and complete. Let be a decreasing fundamental system of neighbourhoods of 0 in formed of ideals, and set , , . One has a projective system of homomorphisms (20.5.11.3), from which one deduces by passing to the limit a canonical homomorphism

  δ̂_{C/B/A} : 𝔍/𝔍² → Ω̂^1_{B/A} ⊗̂_{B̂} Ĉ

and reasoning as in (20.7.17), one sees that the canonical homomorphism is surjective and that is dense in .