§ 6. Extension of scalars in regular algebras
1. Algebras essentially of finite type
Let be a ring. Let be a -algebra and a family of elements of ; let denote the subalgebra of generated by the . We shall say that is an essentially generating family of the -algebra if, for every element of , there exists an element of , invertible in , such that belongs to . It amounts to the same to say that, for every , there exist polynomials and in such that is invertible in and that one has .
We shall say that a -algebra is essentially of finite type if it admits a finite essentially generating family. It amounts to the same to say that there exist a -algebra of finite type and a multiplicative subset of such that the -algebra is isomorphic to .
Example 1
To say that an extension of a field is a -algebra essentially of finite type means that it is an extension of finite type in the sense of A, V, p. 11, Def. 2. The -algebra is of finite type only if it is an extension of finite degree of (V, § 3, No. 4, Cor. 3 of Th. 3).
Example 2
In order that a local -algebra be essentially of finite type, it is necessary and sufficient that it be isomorphic to a -algebra of the form , where is a -algebra of finite type and a prime ideal of (cf. II, § 2, No. 5, Prop. 11 (iii)).
Proposition 1
If the ring is noetherian, every -algebra essentially of finite type is a noetherian ring.
This follows from III, § 2, No. 10, Cor. 3 of Th. 2 and II, § 2, No. 4, Cor. 2 of Prop. 10.
The following properties are deduced from the definition:
Proposition 2
a) Every quotient algebra of a -algebra essentially of finite type is essentially of finite type.
b) Every ring of fractions of a -algebra essentially of finite type is a -algebra essentially of finite type.
c) The product -algebra of a finite family of -algebras essentially of finite type is essentially of finite type.
d) Let be a homomorphism of rings; for every -algebra essentially of finite type, the -algebra is essentially of finite type.
Corollary
Let be a -algebra essentially of finite type and a noetherian -algebra. Then the ring is noetherian.
In fact, it is a -algebra essentially of finite type (prop. 2, d)), and one applies Proposition 1.
Proposition 3
Let be a ring, a -algebra and an -algebra. If is essentially of finite type over and essentially of finite type over , then is essentially of finite type over .
Let us denote by the canonical mapping. Let be a finite essentially generating family of the -algebra , and let be the subalgebra generated by it; let be a finite essentially generating family of the -algebra . Let us denote by the sub--algebra of generated by the and the . Let ; by hypothesis there exist polynomials and in such that is invertible in and such that one has . The non-zero coefficients of and are finite in number; there exists a polynomial such that is invertible in and that and belong to . Then is invertible in , and one has
Thus the elements for and for form a finite essentially generating family of the -algebra .
Corollaire
The tensor product of two -algebras essentially of finite type is a -algebra essentially of finite type.
For let and be two -algebras essentially of finite type. Then is essentially of finite type over (Prop. 2, d)), hence over (Prop. 3).
2. Tensor products of Macaulay or Gorenstein algebras
Proposition 4
Let be a field, a -algebra essentially of finite type and a -algebra.
a) If and are Macaulay rings, the same is true of .
b) If and are Gorenstein rings, the same is true of .
Suppose that and are Macaulay rings (resp. Gorenstein rings), and let us prove that the same is true of . The ring is noetherian (No. 1, cor. of Prop. 2). The -module is free, hence flat. By Prop. 10 of § 2, No. 7 (resp. Cor. 1 of Prop. 12 of § 3, No. 8), it is enough for us to prove that is a Macaulay ring (resp. a Gorenstein ring) for every prime ideal of . The extension of is of finite type (No. 1, Prop. 2 and Example 1); we are therefore reduced to proving the statement in the case where the -algebra is a finite-type extension of .
The ring is then a free -module of finite rank, hence a Macaulay -module. By § 2, No. 6, Cor. 1 of Prop. 8, it is a Macaulay ring, which proves a). Suppose now that is a Gorenstein ring. There exist intermediate extensions , of such that
and such that is a monogenic -algebra for ; this enables us to reduce to the case where the extension of is monogenic (of finite degree). The canonical homomorphism makes into a flat -algebra and, for every , the ring , isomorphic to , is a monogenic -algebra, hence a Gorenstein ring (, n° 9, example 5). Thus is a Gorenstein ring (, n° 8, cor. 1 of Prop. 12), which completes the proof of b).
Corollary 1
Let be a field, an extension of and a -algebra essentially of finite type. For to be a Macaulay ring (resp. a Gorenstein ring), it is necessary and sufficient that the same be true of .
If is a Macaulay ring (resp. a Gorenstein ring), the same is true of by Prop. 4. Since is a faithfully flat -module, the converse follows from Prop. 10 of § 2, No. 7 (resp. Cor. 1 of Prop. 12 of § 3, No. 8).
Corollary 2
Let be a noetherian ring, and -algebras. Suppose that is flat and essentially of finite type over . If and are Macaulay rings (resp. Gorenstein rings), then is a Macaulay ring (resp. a Gorenstein ring).
Suppose first that is a field; let denote the canonical homomorphism from into , and its kernel. The homomorphism induces a homomorphism from the field of fractions of into . Then is identified with ; since , the ring is a Macaulay ring (resp. a Gorenstein ring) by Prop. 10 of § 2, No. 7 (resp. Cor. 1 of Prop. 12 of § 3, No. 8). The assertion then follows in this case from Cor. 1.
Let us pass to the general case. The -algebra is flat, and noetherian by the corollary to Prop. 2 of No. 1. For each prime ideal of , the ring , which is identified with , is a Macaulay ring (resp. a Gorenstein ring) by what precedes. We conclude by applying Prop. 10 of § 2, No. 7 (resp. Cor. 1 of Prop. 12 of § 3, No. 8).
3. Separable extension of the base field in regular¹ or normal algebras
Lemma 1
Let be a noetherian ring, the union of an increasing filtered family of noetherian subrings.
a) If the rings are regular and if is a flat -module for every , the ring is regular.
b) If the rings are normal (, No. 8), is normal.
a) Let be a maximal ideal of ; for each , let us denote by the ideal of . Since is noetherian, is of finite type and there exists an element of such that , so that the -module is isomorphic to . Since is flat over , we have (A, X, p. 141, Lemma 2). Since the rings are regular, it follows that is regular (, No. 2, Prop. 4).
b) Since the are reduced, is reduced. Let and be elements of such that is not a zero divisor and that the element of the total ring of fractions of is integral over . There exists a monic polynomial such that . Let be an element of such that the ring contains , and the coefficients of . Since is normal, there exists such that . Thus belongs to and is normal.
Lemma 2
Let be a field, and let and be extensions of . Suppose that is of finite type and that one of the extensions or is separable. Then the ring is regular.
Let denote the ring ; it is noetherian by cor. of prop. 2 (No. 2). Suppose first that the extension is separable. By A, V, p. 130, corollary, there exists a transcendence basis of such that is a finite separable extension of the pure extension . The ring , isomorphic to a ring of fractions of a polynomial ring over , is regular (, No. 2, cor. 5 of prop. 4); for every prime ideal of , the ring is isomorphic to , hence to a finite product of fields (A, V, p. 35, def. 1, p. 34, th. 4, and p. 33, prop. 5). Since is a free -module, it is a regular ring by cor. of prop. 9 of , No. 5.
Suppose now that is separable over . By the first part of the proof, the ring is regular for every finite-type subextension of . One concludes by applying lemma 1.
Proposition 5
Let be a field, a -algebra and an extension of . Suppose that is essentially of finite type or that the extension is of finite type.
¹ In this number, an algebra over a field is said to be regular if it is a regular ring. In particular, every extension of is a regular algebra. Care will be taken not to confuse this notion with that of regular extension, introduced in A, V, p. 135, def. 2, which will not occur here.
a) If the ring is regular (resp. normal), the ring is regular (resp. normal).
b) If the ring is regular (resp. normal) and if the extension of is separable, the ring is regular (resp. normal).
The -module is free, hence faithfully flat; assertion a) therefore follows from cor. of prop. 8 of I, § 3, No. 5 and from prop. 8 of § 4, No. 5 (resp. cor. 2 of th. 4 of § 1, No. 10). Under the hypotheses of b), for every prime ideal of , the ring is regular (lemma 2), and normal (§ 4, No. 2, cor. 2 of prop. 4). Assertion b) therefore follows from cor. of prop. 9 of § 4, No. 5 (resp. cor. 3 of th. 4 of § 1, No. 10).
4. Absolutely regular or absolutely normal algebras
Definition 1
Let be a field and a -algebra. One says that is absolutely regular (resp. absolutely normal) if the ring is regular (resp. normal) for every radicial extension of finite degree of .
Every absolutely regular (resp. absolutely normal) -algebra is regular (resp. normal), as is seen by taking in definition 1. Recall that the -algebras such that the ring is reduced for every radicial extension of finite degree of are the separable -algebras (A, V, p. 117, th. 2).
Example 1
If is perfect, every regular (resp. normal) -algebra is absolutely regular (resp. absolutely normal).
Example 2
Let be an artinian -algebra. If is normal, it is reduced, hence isomorphic to a finite product of extensions of (A, VIII, § 8, No. 1, prop. 2). The following conditions are equivalent:
A is separable; — A is absolutely regular; — A is absolutely normal.
In fact, if is absolutely normal, it is separable; if is separable, for every finite extension of , the ring is reduced and artinian, hence isomorphic to a product of fields, and consequently regular, so that is absolutely regular.
If moreover the -algebra is of finite degree, the preceding conditions are also equivalent to saying that is étale (A, V, p. 34, th. 4).
Example 3
Let be a regular local -algebra. If the extension of is separable, the algebra is absolutely regular. Let indeed be a finite-degree extension of . The -module is free. Since it is of finite type, every maximal ideal of lies over (V, § 2, No. 1, prop. 1). The ring , isomorphic to , is regular (No. 3, lemma 2). By the corollary to prop. 9 of § 4, No. 5, the ring is regular, which proves that the -algebra is absolutely regular.
Some authors use the terminology “geometrically regular” (resp. “geometrically normal”.
Proposition 6
Let k be a field and A a noetherian k-algebra.
a) If A is absolutely regular (resp. absolutely normal, resp. separable), the same is true of SA for every multiplicative subset S of A.
b) If is absolutely regular (resp. absolutely normal, resp. separable), for every maximal ideal m of A, then A is absolutely regular (resp. absolutely normal, resp. separable).
a) This follows from the fact that the ring is isomorphic to a ring of fractions of the ring for every extension of k.
b) Suppose that is absolutely regular (resp. absolutely normal, resp. separable) for every maximal ideal m of A. Let be a radicial extension of k of finite degree and let be a maximal ideal of . It remains to verify that the local ring is regular (resp. normal, resp. reduced).
The canonical homomorphism makes into a finite A-algebra. The ideal lies over a maximal ideal m of A (V, § 2, No. 1, prop. 1), and is isomorphic to a ring of fractions of the regular (resp. normal, resp. reduced) ring , and is therefore regular (resp. normal, resp. reduced).
Lemma 3
Let k be a field and K a finite-type extension of k. There exist an extension L of K, of finite degree, and a subextension of L over k, radicial and of finite degree over k, such that the extension L of is separable.
Let us choose a composite extension E of K and of a perfect closure of k, and elements of E such that E is a separable algebraic extension of (A, V, p. 130, cor.). Let I be a finite generating subset of K over k. The field (resp. ) is the union of the subfields (resp. ), where runs through the set of subextensions of of finite degree over k; hence one can find such a subextension such that the elements of E = belong to and that each element of I is algebraic and separable over . Then L = is a separable extension of , of finite degree over K, and is a finite radicial extension of k.
Proposition 7
Let k be a field, A and B two k-algebras one of which is essentially of finite type. Suppose that A is absolutely regular (resp. absolutely normal). If the ring B is regular (resp. normal), the same is true of the ring .
Let K be an extension of finite type of k; let us prove that the ring is regular (resp. normal). Consider indeed extensions L and having the properties of lemma 3. Then the ring is regular (resp. normal) by definition, and the ring , which identifies with , is regular (resp. normal) by Prop. 5, b) of No. 3. Consequently, is regular (resp. normal) by Prop. 5, a).
Suppose the ring B regular (resp. normal), and prove the proposition. The canonical homomorphism makes into a free B-module. For every prime ideal of B, the ring identifies with ; by the corollary of Prop. 9 of § 4, No. 5 (resp. Cor. 3 of Th. 4 of § 1, No. 10), it is enough to prove that is regular (resp. normal) for every prime ideal of B.
If the -algebra B is essentially of finite type, the extension of is of finite type and the ring is regular (resp. normal) by what we have seen above. Suppose now the -algebra A essentially of finite type; the ring is noetherian and is the union of the increasing filtered family of noetherian subrings , where runs through the subextensions of finite type of . The latter are regular (resp. normal), and one applies lemma 1 of No. 3.
Corollary 1
Let be a field. The tensor product of two absolutely regular (resp. absolutely normal) -algebras, one of which is essentially of finite type, is an absolutely regular (resp. absolutely normal) -algebra.
Let and be two -algebras satisfying the hypotheses of the corollary. Let be a finite radicial extension of . The ring is regular (resp. normal); the ring is regular (resp. normal) by Prop. 7, as is the ring which is isomorphic to it.
Corollary 2
Let be a field, an absolutely regular (resp. absolutely normal) -algebra, and an extension of . Suppose that is essentially of finite type or that the extension of is of finite type.
a) The ring is regular (resp. normal).
b) If the extension of is separable, the -algebra is absolutely regular (resp. absolutely normal).
Assertion a) results from proposition 7; assertion b) results from cor. 1 and example 2.
Corollary 3
Let be a field, a -algebra, and an extension of . Suppose that the -algebra is essentially of finite type or that the extension of is of finite type. In order that the -algebra be absolutely regular (resp. absolutely normal), it is necessary and sufficient that the -algebra be so.
Suppose absolutely regular (resp. absolutely normal) and let be a purely inseparable extension of of finite degree. The ring , isomorphic to , is regular (resp. normal) by cor. 2.
Suppose conversely that the -algebra is absolutely regular (resp. absolutely normal), and let be a purely inseparable extension of of finite degree. Let be a composite extension of and ; then the ring may be identified with , hence is regular (resp. normal); consequently, the ring is regular (resp. normal) by prop. 5, a) of No. 3.
Corollary 4
Let be a field, a -algebra essentially of finite type, and an extension of which is a perfect field. In order that be absolutely regular (resp. absolutely normal), it is necessary and sufficient that the ring be regular (resp. normal).
This results from cor. 3 and example 1.
5. Characterizations of absolutely regular algebras
Proposition 8
Let k be a field and A a k-algebra essentially of finite type. Let I denote the kernel of the homomorphism of k-algebras such that for x and y in A. The following conditions are equivalent:
(i) the k-algebra A is absolutely regular; (ii) for every regular k-algebra C, the ring is regular; (iii) the ring is regular; (iv) the ideal I of is completely secant (, No. 1, Definition 1).
Let B denote the ring and provide it with the structure of an A-algebra deduced from the homomorphism such that ; then is a homomorphism of A-algebras, and induces by passing to the quotient an isomorphism of onto A.
(i) (ii): this results from Prop. 7. (ii) (iii): it is enough to apply (ii) with , then with . (iii) (iv): the A-module B is free, hence faithfully flat. If the ring B is regular, A is regular (, No. 5, Prop. 8); the ideal I is then completely secant (, No. 3, Prop. 2). (iv) (i): let us suppose the ideal I completely secant and first prove that A is regular. Let m be a maximal ideal of A and let be the homomorphism deduced from . The maximal ideal is equal to ; applying Prop. 6 of , No. 6 to the A-algebra , one sees that the ideal n is completely secant in . Consequently (, No. 3, Prop. 3) the local ring is regular. Let us denote by the homomorphism ; since is the canonical homomorphism of A into , one has . Thus j extends to a local homomorphism of local rings from into , which makes the latter into a faithfully flat -module. By Prop. 8 of , No. 5, the ring is therefore regular. We have thus proved that A is regular.
Now let be an extension of k. The kernel of the mapping deduced from the multiplication of is none other than ; it is therefore completely secant in (, No. 6, Prop. 6). The -algebra therefore satisfies condition (iv); by what has just been seen, it is regular, and this proves (i).
Recall (A, III, p. 133-134) that the quotient endowed with the A-module structure deduced from is denoted by and called the module of k-differentials of A. When the k-algebra A is essentially of finite type, the ring is noetherian, so that the A-module is finitely generated.
One denotes by , or simply by d, the k-linear mapping from A into which associates with an element x of A the class of in . The mapping d is a k-derivation; for every A-module M and every k-derivation D : A M, there exists a unique A-linear mapping such that (loc. cit., Prop. 18).
If S is a multiplicative subset of A, the canonical -linear mapping (loc. cit., p. 136)
Let k be a field and A a local k-algebra essentially of finite type. Put n = \dim(A) + \deg.\mathrm{tr}_k(\kappa_A). Let B be a k-algebra of finite type and q a prime ideal of B such that the k-algebra A is isomorphic to (cf. No. 1, Example 2); one has n = \dim_q(B) (VIII, § 2, No. 4, Cor. 5 of Th. 3). By loc. cit., Th. 3, c) and Cor. 5, one also has
Theorem 1
Let k be a field and A a local k-algebra essentially of finite type. Put n = \dim(A) + \deg.\mathrm{tr}_k(\kappa_A).
a) One has [\kappaA \otimes_A \Omega_k(A) : \kappa_A] \geq n.*
b) The following conditions are equivalent: (i) the k-algebra A is absolutely regular; (ii) the A-module \Omegak(A) is free of rank n;* (iii) one has [\kappaA \otimes_A \Omega_k(A) : \kappa_A] = n.*
Consider the assertion (iii') one has [\kappaA \otimes_A \Omega_k(A) : \kappa_A] \leq n;* it is clear that (ii) implies (iii) and that (iii) implies (iii'). To prove the theorem it is enough to prove the implications (i) \Rightarrow (ii) and (iii') \Rightarrow (i).
Choose a k-algebra of finite type B and a prime ideal q of B such that A is isomorphic to (No. 1, Example 2). One has \dim_q(B) = n; replacing B by a ring , with , one may impose that the spectrum of B is connected and of dimension n, which we shall henceforth suppose.
(i) \Rightarrow (ii): Suppose the k-algebra A absolutely regular. As in the proof of Prop. 8, let us denote by \mu : A \otimes_k A \to A and \nu : \kappa_A \otimes_k A \to \kappa_A the homomorphisms deduced from the multiplication of A; put I = \mathrm{Ker}(\mu) and n = \mathrm{Ker}(\nu). The ideal I is completely secant; the ideal n is maximal and the local ring (\kappa_A \otimes_k A)_n is regular (loc. cit.). By Theorem 1 of § 5, No. 2, the finitely generated A-module I/I^2, which is none other than \Omega_k(A), is projective, hence free. But by the remark of § 5, No. 6, the ideal n identifies with \kappa_A \otimes_A 1, so that n/n^2 is isomorphic to \kappa_A \otimes_A \Omega_k(A). One therefore has
Let us denote by B' the ring \kappa_A \otimes_k B, and by S the image in B' of B - q. The ring \kappa_A \otimes_k A identifies with , that is to say with . There therefore exists a maximal ideal q' of B' not meeting S such that one has n = ; the ring (\kappa_A \otimes_k A)_n is then isomorphic to , and one has
Since n contains \kappa_A \otimes_k qB_q , q' contains the ideal \kappa_A \otimes q of B' , hence its inverse image in B contains q ; it is equal to q since q' does not meet S . By VIII, § 2, No. 4, cor. to prop. 5, one has
which proves (ii).
(iii') \Rightarrow (i) : Suppose that one has [\kappa_A \otimes_A \Omega_k(A) : \kappa_A] \leq n , that is to say [\kappa(q) \otimes_B \Omega_k(B) : \kappa(q)] \leq n , and let us prove that the k-algebra B is absolutely regular. Let (x_1, \ldots , x_n) be a sequence of elements of B such that 1 \otimes dx_1, \ldots , 1 \otimes dx_n generate the \kappa(q)-vector space \kappa(q) \otimes_B \Omega_k(B) . Replacing B by B_f , for a suitable element f of B - q , one may suppose that dx_1, \ldots , dx_n generate the B-module \Omega_k(B) (Nakayama lemma and II, § 5, No. 1, prop. 2). Let \bar{k} be an algebraically closed extension of k . It is enough to prove that the \bar{k}-algebra B_{(\bar{k})} is regular, since this will imply that B is absolutely regular (cor. 4 to prop. 7 of No. 4). For every canonical factor C of B_{(k)} , the differentials d(1 \otimes x_i) generate the C-module \Omega_{\bar{k}}(C) (A, III, § 10, No. 12, prop. 20). The theorem therefore follows from the two following lemmas:
Lemma 4
Let k be a field, K an algebraic extension of k , B a k-algebra whose spectrum is connected. For every canonical factor C of B_{(K)} , one has \dim(C) = \dim(B) .
Let us first treat the case where the extension K is of finite degree over k . The B-module C is a direct factor of a free module, hence finitely generated projective. Since the function p \mapsto \mathrm{rg}_p C is constant on Spec(B) (II, § 5, No. 3), the support of the B-module C is equal to Spec(B) , so that the B-module C is faithfully flat. This implies \dim(C) \geq \dim(B) (VIII, § 2, No. 1, prop. 2), hence the required equality since one has \dim(C) \leq \dim(B_{(K)}) = \dim(B) (VIII, § 2, No. 4, cor. to prop. 5).
Let us pass to the general case. Let e be the idempotent element of B_{(K)} such that C = B_{(K)}/B_{(K)}e . There exists a subextension K' of K of finite degree such that e is the image of an idempotent element e' of B_{(K')} . Put C' = B_{(K')}/B_{(K')}e' . The ring C' \otimes_{K'} K is identified with C , and C' is a canonical factor of B_{(K')} . One has \dim(C') = \dim(B) by the case already treated, and \dim(C') = \dim(C) by loc. cit. , whence the lemma.
Lemma 5
Let k be an algebraically closed field, A a finitely generated k-algebra whose spectrum is connected, n an integer and (x_1, \ldots , x_n) a finite sequence of elements of A . Suppose that A is of dimension n and that the differentials dx_1, \ldots , dx_n generate the A-module \Omega_k(A) . Then the ring A is an integral domain and regular and the A-module \Omega_k(A) is free with basis (dx_1, \ldots , dx_n) .
Let m be a maximal ideal of A such that \dim(A_m) = n . One has [A/m : k] = 1 (V, § 3, No. 3, prop. 1 (iii)), hence A = m \oplus k1_A . Let p and q denote the corresponding projections. For and in , one has
Let us now prove that the ring is an integral domain and regular. There exists a minimal prime ideal of such that . For every maximal ideal of containing , one has (VIII, § 2, No. 4, cor. 2 to th. 3), so that is regular by what we have just seen. In particular is an integral domain, which implies . Since this holds for all the maximal ideals of , one deduces that one has . But is connected, is nonempty and one has (II, § 4, No. 4, prop. 16). One deduces that , whence , which means that is an integral domain. One then has for every maximal ideal of ; applying the first part of the proof, one deduces that is regular.
Finally, let be a linear relation between the with coefficients in . If the are not all zero, there exists an index and a maximal ideal of such that does not belong to (V, § 3, No. 3, Prop. 1, (iii) and (iv)); but this contradicts the fact proved above that the classes of in are linearly independent.
Example
When is an extension of finite type of , theorem 1 yields again Cor. 1 of A, V, p. 128, taking account of Example 2 of No. 4.
Corollary 1
Let be a field and a -algebra essentially of finite type. The set of elements of such that the -algebra is absolutely regular is open in .
We may suppose that the -algebra is of finite type. The set considered is then formed of the prime ideals such that . Now the mapping is lower semi-continuous by definition, and the mapping is upper semi-continuous (Nakayama's lemma and II, § 5, No. 1, Prop. 2).
We shall see later (§ 7, No. 9, Cor. 4 of Th. 3) that under the hypotheses of Cor. 1, the set of prime ideals of such that the ring is regular is open in .
Corollary 2
Let be a field and a -algebra essentially of finite type. In order that should be absolutely regular, it is necessary and sufficient that the -module be projective and that for every minimal prime ideal of , the -algebra be separable.
Conversely, suppose that the -module is projective and that the -algebra is separable for every minimal prime ideal of . Let be a prime ideal of , and a minimal prime ideal of contained in . Since the -module is free (II, § 3, No. 2, Cor. 2 of Prop. 5), we have
The -algebra is Artinian and separable, hence absolutely regular (No. 4, Example 2). Th. 1 implies
It then follows from Th. 1 that the -algebra is absolutely regular, whence the corollary.
Remark
Suppose the -algebra absolutely regular. Then the total ring of fractions of is identified with the product of the , where runs through the set of minimal prime ideals of (IV, § 2, No. 5, Prop. 10); it is therefore a separable -algebra. Conversely, suppose that is a separable -algebra; for every minimal prime ideal of , the -algebra is a ring of fractions of (IV, § 1, No. 1, Cor. 3 of Prop. 2 and No. 3, Cor. 1 of Prop. 7), hence a separable -algebra (No. 4, Prop. 6). If moreover the -module is projective, it follows from Cor. 2 that the -algebra is absolutely regular.
Corollary 3
Let be a field of characteristic 0 and a -algebra essentially of finite type. For to be regular, it is necessary and sufficient that the -module be projective.
By Corollary 2, it is enough to prove that an artinian local -algebra such that is a free -module is a field. Now, by IX, § 3, No. 3, Theorem 1, there exists a subfield of such that . Let be the composite mapping of the projection of onto and the canonical mapping . One verifies as in the proof of Lemma 4 that is a -derivation. But every -derivation of into an -module annihilates : let in fact , and an integer such that one has . This implies in the -module the relation , hence since . Thus , hence , and finally .
Exercises
See the exercises for § 6.