WorksMathematics
Vasconcelos' conjecture on the module of differentials
Abstract
Let be a field and let be a local -algebra essentially of finite type whose module of Kähler differentials has finite projective dimension. A conjecture of Vasconcelos, in the form studied by Herzog and Briggs, asserts that is then a complete intersection when . We give a complete proof, in particular one independent of the Eisenbud–Mazur conjecture, that is a reduced complete intersection when , and, over an arbitrary field, when some minimal prime of is a generic point at which is a field separably generated over ; in the latter case and is essentially smooth at all of its generic points. In positive characteristic , Vasconcelos' conjecture for reduced algebras holds for every such algebra exactly when .
Introduction
Let be a field and let be a local ring essentially of finite type over , that is, a localization of a finitely generated -algebra at a prime ideal. Vasconcelos conjectured that the projective dimension of the module of Kähler differentials can only take the values , and , for affine algebras over an arbitrary field and leaving questions of separability aside [12 §4, (C2)]. In the form studied by Herzog [7 §1, (C2)] and Briggs [4], for , the conjecture asserts that
if , then is a complete intersection. (C2)
See the introduction of [4] for the history of this conjecture and of its companion [12 §0, (C1)]: for an ideal of finite projective dimension in a local ring ,
if , then is generated by a regular sequence. (C1)
In characteristic zero, the graded cases of both conjectures were settled by Avramov and Herzog, (C1) in [AH94, Thm. 3] and (C2) in [AH94, Thm. 1]. Briggs proved (C1) in full [4 Thm. A]. For the module of differentials, Briggs proved the conjecture for algebras without non-trivial evolutions [4 Thm. 3.4], so that in characteristic zero it would follow from the Eisenbud–Mazur conjecture on evolutions and symbolic squares. Since the Eisenbud–Mazur conjecture has been shown to fail [9], that reduction does not prove Herzog’s (C2), and a different argument is needed.
In this paper we prove the conjecture for ordinary Kähler differentials in characteristic zero, and we determine exactly what survives in positive characteristic.
Throughout, denotes a local ring with maximal ideal and residue field , and is the residue field of a ring at a prime ideal . For a finite module over we write for its projective dimension, for its minimal number of generators and for its set of associated primes; is the height of an ideal , and is the common length of the maximal -regular sequences in a proper ideal [8 Thm. 16.7]. If is Cohen–Macaulay, for instance regular, then [8 Thms. 17.3 and 17.4]
for every prime ideal .
A ring is non-trivial if it is not the zero ring. A generic point of a ring is a minimal prime ideal of it.
In positive characteristic we write for the prime field and for the subfield of -th powers of a field .
A local -algebra is essentially smooth if it is a localization of a smooth -algebra, and a local -algebra essentially of finite type is called a complete intersection over if it is of the form with a localization of a polynomial ring over at a prime ideal and an -regular sequence. Whether a given presentation with regular has generated by a regular sequence does not depend on the presentation [8 Thm. 21.2(ii)].
Theorem 1.1. Let be a field of characteristic zero and let be a non-trivial local -algebra essentially of finite type. If , then
is a reduced complete intersection over , and
.
Reducedness is a conclusion here, not a hypothesis. In positive characteristic the conclusion can fail even for domains, and the correct statement involves the generic points of .
Theorem 1.2. Let be any field and let be a non-trivial local -algebra essentially of finite type with . Suppose that for some minimal prime of the local ring is a field which is separably generated over . Then
is a reduced complete intersection over ,
the ring is essentially smooth over for every minimal prime of , and
.
Corollary 1.3. Let be a perfect field and let be a non-trivial local -algebra essentially of finite type with . If the local ring is reduced for some minimal prime of , then
is a reduced complete intersection over ,
the ring is essentially smooth over for every minimal prime of , and
.
In particular every reduced such is a complete intersection over .
For reduced algebras the dividing line is the imperfection degree of the base field .
Theorem 1.4. Let be a field of positive characteristic . The following are equivalent.
(1) .
(2) Every non-trivial reduced local -algebra essentially of finite type with is a complete intersection over .
Outline of the proof
Outline of the proof. The proof descends in three steps. Differentials occur only in the first step.
Step 1: from differentials to a quotient of the conormal module. Present as with a localization of a polynomial ring in variables over . The conormal sequence
shows that the image of , the module generated by the rows of the Jacobian matrix, is a quotient of with exactly when . At a generic point of whose local ring is a separably generated field, the Jacobian criterion says that has rank at least ; hence needs at least generators. The conjecture is thereby reduced to a statement in which differentials do not appear: a quotient of of finite projective dimension cannot have more than generators, and if it has that many, then is generated by a regular sequence.
To state it, let be a local ring, an ideal and . Let
be an -linear map into a finitely generated -module, and for write for the image of in , where is the class of in . The statement does not require to be regular: it holds whenever has finite projective dimension over , with the grade of in place of its height.
Theorem 1.5. Let be a Noetherian local ring and an ideal such that has finite projective dimension over . Let be a surjective -linear map onto a finitely generated -module with .
(1) .
(2) If , then equality holds and is generated by an -regular sequence whose images form a basis of .
(3) If and for every associated prime of the -module , then is generated by an -regular sequence.
If is regular, then by Serre’s theorem [10 Tag 00O7], and ; in parts (2) and (3) the ring is then a complete intersection. Step 1 applies the theorem in this case to the image , with induced by . For and the identity one has
so part (2) recovers the theorem of Briggs [4 Thm. A], that is, (C1): if and , then is generated by a regular sequence. For regular and a free module of rank , part (2) is the classical criterion of Vasconcelos on free conormal summands [Vas68, §2, Prop.]. The Jacobian image is a priori neither free nor all of , and this is why the general statement is needed.
Granting Theorem 1.5, part (2) shows that is a complete intersection in the situation of Theorem 1.2. Only afterwards do we obtain injectivity of the conormal map, the bound , and reducedness together with smoothness at the generic points, all from one computation at the associated primes of . There the module of differentials, having finite projective dimension over a ring of depth zero, is free of the same rank as at the given generic point. Hence the Jacobian matrix of the regular generators of has an invertible maximal minor, is smooth over there, and a regular local ring of depth zero is a field. In characteristic zero the hypothesis on the generic point holds automatically: the module of differentials is free at every generic point, and in characteristic zero this already makes the local ring regular. Thus Theorem 1.1 is a corollary of Theorem 1.2. Over an imperfect field with the rank of at a generic point may drop by one, and part (3) takes over. Over every field with a one-dimensional local domain with free differentials which is not a complete intersection is obtained from the monomial curve by a purely inseparable change of coordinates (Proposition 7.2).
Step 2: from a quotient of the conormal module to a change of rings. To prove Theorem 1.5 one chooses an -regular sequence in such that are linearly independent and which cannot be prolonged by an element of with these two properties, and passes to
The residual ideal has grade
If has finite projective dimension over , two elementary facts finish the proof. First, an ideal of grade zero with a cyclic quotient of finite projective dimension is zero; so contains a nonzerodivisor unless , and prime avoidance shows that span , for otherwise the sequence could be prolonged. Second, an ideal of grade one with such a quotient is principal if has depth one at every associated prime of the quotient. The second fact rests on the factorization of ideals with finite free resolutions due to Buchsbaum and Eisenbud [2 Cor. 5.2]; this is the only result beyond the textbooks which is quoted in this paper, and it is needed only for part , that is, only for Theorem 1.4. Everything therefore depends on transferring finite projective dimension from to .
Theorem 1.6 (change of rings along a regular sequence). Let be a Noetherian local ring and an ideal such that has finite projective dimension over ; this holds for every when is regular. Let be an -linear map into a finitely generated -module with . If is an -regular sequence such that are linearly independent over , where denotes the image of in , then
The regular sequence in this theorem is chosen in advance and is not extracted from homological data, and neither freeness of nor injectivity of is assumed. This is the main technical result of the paper.
Step 3: the change of rings. Theorem 1.6 is proved by removing one regular element at a time, in the residue algebra of a free DG algebra resolution of over which has a variable of degree one for each generator of . A finite free resolution of , of length , yields for every -linear functional a derivation of of degree with whose commutator with the differential takes values in the ideal generated by the variables of degree less than . This alone makes the commutator nilpotent in each degree, within a range depending only on (Proposition 3.1); the construction originates in the proofs of [4 Lem. 2.6 and Thm. 2.7]. A derivation of this kind with forces the top group to vanish, so the projective dimension drops by one (Lemma 4.1). The derivations dual to the selected elements descend to the successive quotients, and the step can be repeated times without changing . The argument is self-contained; in particular, the Eisenbud–Mazur conjecture is not used.
Organization
The sections follow the three steps in reverse order, so that every result is proved before it is used.
Section prepares Step . It fixes the DG algebra resolution of and its residue algebra , and shows that a free DG module with triangular differential over resolves its reduction (Lemma 2.1).
Section carries out the first half of Step . It constructs the derivation of from a finite free resolution of and proves that its commutator with the differential is nilpotent within a range depending only on (Proposition 3.1).
Section completes Step . One regular element is removed in Lemma 4.1, and repeating this along the prescribed sequence proves Theorem 1.6.
Section 5 carries out Step 2. It treats the residual ideal in grades zero and one (Lemmas 5.1 and 5.2) and proves Theorem 1.5 with a maximal regular sequence whose images in are linearly independent.
Section 6 carries out Step 1. It computes the rank of the Jacobian image at a generic point (Lemma 6.1) and proves Theorem 1.2, Theorem 1.1 and Corollary 1.3.
Section 7 carries out Step 1 in positive characteristic. Theorem 1.5(3) gives Corollary 7.1 and the implication of Theorem 1.4; Proposition 7.2 gives the converse.
Declaration of generative AI: As the outcome of about three weeks of intensive discussion between the second author and a ChatGPT model, a proof was proposed which rests on two deep results, Briggs’ theorem on radical elements of the homotopy Lie algebra [4 Thm. 2.7] and a theorem of Briggs, Grifo and Pollitz on cohomological support varieties [3]. A careful study of their proofs showed which parts of them the argument actually needs, and this led to the comparatively elementary proof given here, which uses neither result. The authors reviewed and edited the content as needed and take full responsibility for the content of the article.
Conventions and resolutions
This section fixes conventions and constructs the resolution of in which the computations of § 3 and § 4 take place. All rings are commutative and Noetherian. In a minimal free resolution of a finite module over a local ring all differentials vanish after tensoring with ; therefore, if for some , then .
DG algebras and modules
Let be a quotient of a local ring and let . Finite projective dimension of over the quotient will be detected on . To compare it with we resolve over by an algebra in which is the differential of a variable ; dividing out and then yields a resolution over . Resolutions of this kind were introduced by Tate [11 §2]. The multiplication is what makes derivations available, and a derivation of a free algebra is determined by its values on the variables.
We use homological grading, with differentials of degree . Every differential is denoted by , the degree of a homogeneous element by , and the component of degree of a graded object by ; is the homology of in all degrees together. A DG algebra is strictly graded-commutative and concentrated in nonnegative degrees. A DG module over a DG algebra satisfies
and we put
A derivation of degree from to is a map , linear over the ring , which lowers degrees by one and satisfies
For a ring and a set of variables of positive degrees, denotes the free strictly graded-commutative -algebra on : polynomial in the variables of even degree and exterior in those of odd degree. A monomial is a product of variables; the monomials form a -basis of . A derivation of degree from to a graded module is determined by its values on , and these may be prescribed arbitrarily in the right degrees.
A free resolution by a DG algebra
Let be a local ring, an ideal with generators , not necessarily minimal, and . Let be a free DG -algebra on variables , with in degree , such that
Such an algebra exists [11 Thm. 1]: start from the Koszul complex and, for in turn, adjoin variables of degree whose differentials are cycles representing generators of the -module of the algebra built so far. The elements of degree at most are unchanged by this step, so the homology in degrees below is unchanged and becomes zero. Each is a free -module and , so is a free resolution of over . We put
Then , and every is a cycle of because . For we write for its class in . When an -linear map into a finitely generated -module is given, denotes the image of in , as in the Introduction.
A DG module as in the next lemma is built by adjoining free generators one degree at a time, and such a module cannot distinguish quasi-isomorphic algebras: this is the DG form of the fact that tensoring with a bounded below complex of free modules preserves quasi-isomorphisms. We use the lemma twice: to lift a finite resolution over to a DG module over , and to compute over a quotient of by a regular element.
Lemma 2.1. Let be a surjective quasi-isomorphism of DG algebras. Let be a DG -module which is graded free on homogeneous basis elements of nonnegative degrees, such that the differential of every basis element is a -linear combination of basis elements of smaller degree. Then is a surjective quasi-isomorphism.
Proof. Let , an acyclic DG ideal. The kernel of is the direct sum of the over the basis elements , and it suffices to show that it is acyclic. Let be a homogeneous cycle in it, let be the largest degree of a basis element occurring in , and write the corresponding part of as with . Since involves only basis elements of degree less than , the coefficient of in is ; hence . Choose with . Then is a cycle in the kernel which involves only basis elements of degree less than . As involves finitely many basis elements, of nonnegative degrees, repeating this step exhibits as a boundary. □
Derivations from a finite free resolution
This section constructs, from a finite free resolution of , the derivations with which a regular element will be divided out in § 4.
Motivation
3.1. Motivation. Keep the notation of § 2.2, let be an -linear map into a finitely generated -module with , and suppose that is a nonzerodivisor of with . Put . We want to show that .
The prototype is the argument of Vasconcelos for a free module [Vas68, §2, Prop.]. The homomorphism , , is well defined. Since is a nonzerodivisor on and on , a finite free resolution of over reduces modulo to a free resolution of over , so
If is free, there is an -linear map taking to , and composing it with splits the map . Thus is a direct summand of over , and . When is not free there is no such splitting: the only functionals at our disposal are the maps with . The construction of this section is what remains of the splitting over the residue field.
In terms of resolutions, we want to compare the homology of , which is and is bounded when , with the homology of , which will turn out to be (Lemma 4.1). As a graded algebra is the tensor product over of and the exterior algebra on , and the simplest way to pass from to is to contract against : one looks for a derivation of of degree with . If commuted with the differential in the graded sense, that is, if
vanished, then would split as a complex into and a shifted copy of , and boundedness would pass from to at once. This is the chain-level form of the splitting above, and it is what happens when is free. In general . The map which induces on lowers degrees by two; it plays the role, on Tor, of the operators of degree two attached to the hypersurface by Gulliksen [Gul74, Thm. 3.1] and Eisenbud [5 §1], for which finite projective dimension over is a nilpotence property. We use this comparison only as a guide and do not need to make it precise. The task is to choose in such a way that is visibly nilpotent in each degree. The values are forced on us only through a functional on the conormal module. Indeed, a variable of degree two has with , and for to have no component in degree zero one needs
where is the image of in ; so should come from an -linear map . Extending to the higher variables of means solving one equation per variable, and over these equations have no reason to be solvable. The idea, which goes back to the proofs of [4 Lem. 2.6 and Thm. 2.7], is to solve them one level higher. If the functional factors as through the module , with a finite free resolution of length , one first lifts to a derivation from into a DG module lifting ; there the equations are solvable because is a resolution of . One then composes with the projection of onto given by . The only step which does not commute with differentials is this last projection, and it fails to commute only through basis elements of of degree at least two, with defects of positive degree at most . Thus takes values in the ideal generated by the variables of degree less than , and on each of these variables a bounded number of applications of reaches degree zero. This is all that is needed for nilpotence.
The construction
Proposition 3.1. Let , , , and be as in § 2.2. Let be an -linear map into an -module which has a resolution by finitely generated free -modules . For every -linear map there is a derivation of degree such that
and such that the commutator
satisfies
for all integers and with
Moreover is a chain map and a derivation of of degree , its image is contained in the ideal of generated by the variables of degrees , and if .
Proof. Write for the augmentation and choose bases of the .
Step 1: lifting the resolution. We construct a DG -module , graded free on a finite basis with in degree and in bijection with the chosen basis of , such that the differential of each element of involves only and such that . Put for . Suppose that the differential has been defined on , so that the free module on these elements is a DG module with , the part of in degrees less than . By Lemma 2.1, applied to the augmentation , the projection is a surjective quasi-isomorphism. For with image , the element is a cycle. Lift it to an element of of degree ; the differential of this lift lies in the acyclic kernel of the projection and is a cycle there, so after subtracting a primitive in the kernel we obtain a cycle lifting . Put . This defines . By Lemma 2.1 again, is a surjective quasi-isomorphism, hence so is
Only the basis of is finite; itself is not bounded.
Step 2: lifting the conormal map. We construct a derivation of degree with
Choose with . The first identity holds on because and vanishes on . Let with , and suppose that has been defined on the variables of degree less than and satisfies on the subalgebra they generate, which contains . Then , so is a cycle of of degree . If it is a boundary because . If , write with ; then and
so is again a boundary, since induces . Choose with . Since is additive, -linear and satisfies the product rule , it vanishes on once it vanishes on the variables. The map does not depend on .
Step 3: projecting to . Put , a graded free -module on , and . Let be the -linear map of degree zero with
and put . Then is a derivation of degree , and (3.1) follows from (3.3). By (3.3),
The defect satisfies for . It vanishes on . On its value is
because on . On with its value is an element , of positive degree at most . Let be the ideal generated by the variables of degrees . A monomial of positive degree contains a variable of degree at most , so every homogeneous element of of positive degree at most lies in . Hence
and if , in which case . As the commutator of two derivations of odd degree, is a derivation of degree , and
Step 4: powers of . If , then and (3) holds. Let and put . For a variable of degree at most , the element lies in by (6) and has degree
As has no nonzero elements of degree zero, . We claim that
For this is (6). Let and , and write , a finite sum over variables of degrees with . Since is a derivation of even degree,
The terms with vanish. For the factor lies in , by (6) if , and
by induction. Hence , which proves (7). Finally, is generated in positive degrees, so has no nonzero elements of degree . If , then , and (7) gives
which proves (3). □
Change of rings
One regular element
The next lemma isolates what a derivation of degree with nilpotent commutator gives. Its proof is the chain-level form of the argument by the change-of-rings exact sequence below. Let be a quotient of a local ring with , and let be a nonzerodivisor of which maps to zero in . For there is an exact sequence
so is surjective onto for , and the group vanishes as soon as some power of kills . This general sequence is not used below. In the proof its role is played by the short exact sequence of complexes given by multiplication by , and the connecting map of that sequence is the map induced by for any derivation with .
Lemma 4.1. Let be a local ring, let be a non-trivial quotient of with , and let be a nonzerodivisor of . Let be a free DG -algebra with and for , having a variable of degree one with , and put .
(1) The DG algebra is a free DG algebra over on the variables , and it is a free resolution of over . In particular .
(2) Suppose that is a derivation of of degree with and that satisfies for some integer . Then
Proof. (1) Let , the Koszul complex on . Since is a nonzerodivisor, is a surjective quasi-isomorphism. As a graded -module, is free on the monomials in the variables other than , and the differential of such a monomial of degree is a -linear combination of monomials of degree less than . By Lemma 2.1,
is a quasi-isomorphism. The target is as a graded algebra, so it is a complex of free -modules resolving , and tensoring it with gives .
(2) We have : the nonzerodivisor annihilates the non-trivial ring , so is not free over . By (1) and the remark at the beginning of § 2 it suffices to prove . Recall that for .
As a graded algebra is free over the exterior algebra on the cycle , the projection has kernel , and . Hence left multiplication by gives a short exact sequence of complexes
where with differential . Its long exact homology sequence [10 Tag 0111] contains the exact sequences
so is surjective for . Now
so the derivation preserves the ideal and induces a chain map on . To compute , let be a cycle, lift it to and write . Then , while gives
that is,
Thus is the map induced by . Consequently
is surjective, as a composite of surjections. It is also zero, because is the image of . Hence .
Note that is applied only to elements of ; it need not induce a map on . The lemma holds in every characteristic and involves no completeness hypothesis.
Proof of Theorem 1.6
If there is nothing to prove. Let and extend the given sequence to a finite generating list of . Form and as in § 2.2. Since are linearly independent in , there are -linear maps with
Let be the length of a finite free resolution of over . Proposition 3.1 gives derivations of with for , whose commutators satisfy whenever .
For put
We prove by induction on that is a free DG -algebra on the variables which resolves , and that
For this is the choice of . Let and assume the statement for . The image of in is a nonzerodivisor which maps to zero in , and it is the differential of the variable of . Lemma 4.1(1), applied to , and this element, gives the assertion about . The residue algebra of is . The derivation maps to for , since ; hence it induces a derivation of , still with . Its commutator with the differential is the map induced by , so its -th power vanishes in degree whenever . Take , which is finite by the induction hypothesis, and any with . Lemma 4.1(2) gives
This completes the induction, and (8) for is the theorem. □
The module and its resolution are never replaced by quotients in this induction; what changes is only the algebra on which the fixed derivations act. This is essential. An induction on which replaces , and by , and does not work, because the last module may have infinite projective dimension. For instance, let
with the coordinatewise quotient map. Then , since and are nonzerodivisors on , and for one finds
This module has infinite projective dimension over : because the annihilator of in is , the minimal free resolution of is infinite, and is a nonzerodivisor on .
Quotients of finite projective dimension
The residual ideal
Theorem 1.6 is used as follows. In the situation of Theorem 1.5, choose a regular sequence in with independent images in , as long as possible, and pass to
Then , and the grade of the residual ideal is . The question what looks like has become the question which ideals of small grade in a local ring have a cyclic quotient of finite projective dimension. In grade zero, that is, for ideals consisting of zerodivisors, the answer is that only the zero ideal does: an ideal with a finite free resolution has a well-defined rank, namely the alternating sum of the ranks of the free modules, and this rank is seen at every associated prime. In grade one the answer is that the ideal is principal, provided the ring has depth one at every associated prime of the quotient; this follows from the factorization theorem of Buchsbaum and Eisenbud [2 Cor. 5.2]. We state the grade-zero fact for modules, because the module form is used again in § 6, at the associated primes of a complete intersection.
Lemma 5.1. Let be a local ring and a finitely generated -module with .
The modules which are free over , for prime ideals of , all have the same rank, and is free for every $\mathfrak p\in\operatorname{Ass}T.
(2) If is free of rank for some prime ideal of , then every surjection is an isomorphism.
(3) If for an ideal which is contained in an associated prime of , then .
Proof. (1) Let be the alternating sum of the ranks of the modules in a finite free resolution of . For a prime such that is free, the localized resolution is a split exact sequence of free modules, so has rank . For the ring has depth zero, so is free by the Auslander–Buchsbaum formula [10 Tag 090V].
(2) The case is trivial. By (1), is free of rank for every , so is a surjection between free modules of the same finite rank over a local ring, and therefore an isomorphism. The kernel of is a submodule of which vanishes at every associated prime of . Its associated primes are among those of [10 Tag 02M3], so it has none, and it is zero [10 Tag 0587].
(3) If , then is nonzero and cyclic, and it is free by (1), hence free of rank one. Part (2), applied to , gives .
Equivalently, a nonzero proper ideal with a cyclic quotient of finite projective dimension contains a nonzerodivisor; this ideal form is classical.
Lemma 5.2. Let be a local ring and an ideal with . If for every associated prime of , then for a nonzerodivisor of .
Proof. The ideal is nonzero, since otherwise the associated primes of , at which has depth zero, would be associated to . It has a finite free resolution, so by [2 Cor. 5.2] we have for a nonzerodivisor and an ideal which is either or contains a -regular sequence of length two. Multiplication by is an isomorphism of onto , hence
If , choose . Then contains a -regular sequence of length two, which is also -regular, so , while because . This contradiction shows and .
If is Cohen–Macaulay, the hypothesis of Lemma 5.2 says that every associated prime of has height one.
Proof of Theorem 1.5
Choose an -regular sequence in such that are linearly independent in and which is maximal with these two properties. It exists, since the empty sequence has them and linear independence bounds the length by . Let
Then is non-trivial, and Theorem Theorem 1.6 gives . If , then is contained in no associated prime of by Lemma Lemma 5.1(3); by prime avoidance [10 Tag 00DS] it contains a nonzerodivisor of , whose lift to prolongs to an -regular sequence in . Hence
The images span. We show that span . If , then generate , and their images span because is surjective. Let , and suppose that the span of is a proper subspace. The set
is an ideal of , and because is surjective. Let be the inverse images in of the associated primes of ; none of them contains , since is contained in no associated prime of . By prime avoidance [10 Tag 00DS] there is outside . Then is -regular and are linearly independent, contradicting maximality. Thus
Parts (1) and (2). A regular sequence in has length at most [8 Thm. 16.7], so , which is (1). If , then , and (5.1) gives . Hence , and the independent elements form a basis of . This proves (2).
Part (3). Suppose and for every . Then . The associated primes of the -module are the ideals with . For such the sequence is -regular and , so [10 Tag 090R]
Lemma 5.2 gives with a nonzerodivisor of . For a lift of , the sequence is -regular and generates . □
Only one regular sequence is chosen in this proof. The three parts correspond to three possibilities for the residual ideal: its grade cannot be negative, it vanishes in grade zero, and it is principal in grade one under the depth condition of part (3). When has grade two or more there is no such conclusion: in Proposition 7.2 the ring is regular, the residual ideal has height two and .
Differentials
The Jacobian image
Let be a field and a non-trivial local -algebra essentially of finite type. Fix a polynomial ring , an ideal of it contained in a prime ideal , and put , such that . Let
Then is a regular local ring, so that Theorem 1.5 applies with grade , and is free on . The conormal sequence of reads
We call the image of the Jacobian image, and we write for regarded as a map onto it; it plays the role of in Theorem 1.5. Thus
Since is free, if and only if . In that case is a quotient of the conormal module with finite projective dimension, and Theorem 1.5 applies. The number is the minimal number of generators of the module generated by the rows of the Jacobian matrix of a generating set of ; it is at least the rank of that matrix at the closed point. Theorem 1.5(1) says that
and what has to be shown is that is not smaller. Since the number of generators of a module can only drop under localization, it suffices to bound the rank of from below at a generic point, where it is computed by the classical Jacobian criterion.
For a homomorphism of rings we write for the first homology of the naive cotangent complex of over [10 Tag 00S0]. It vanishes when is a localization of a polynomial ring over [10 Tag 08Q1, Tag 00S7], it is when is surjective with kernel [10 Tag 07BP], and for a field extension it vanishes if and only if is separable over [10 Tag 0322]. For homomorphisms such that is a free -module there is the exact Jacobi–Zariski sequence [10 Tag 00S2]
Lemma 6.1. Let be a minimal prime of such that is a field, let be its inverse image. Then
and if is separably generated over . Consequently
Proof. Since is a field, , and the localization of at is the map
with image . In the Jacobi–Zariski sequence (11) of the first term vanishes, because is a localization of a polynomial ring 、】【 over , and maps to by (12). Hence is the kernel of (12). The ring is a regular local ring [10 Tag 0AFS] of dimension , so the source of (12) has dimension , and . If is separably generated over , then it is separable over [10 Tag 030X], so . Finally , and because . □
The vector space depends only on the field extension , and it vanishes exactly when is separable over [10 Tag 0322]. Its dimension is what the Jacobian matrix fails to see in positive characteristic. Localizing (10) at gives the first of the equalities
and the second is the dimension formula for , which is an equality because a field is universally catenary [10 Tag 02IJ]. Comparing with Lemma 6.1,
which is Cartier’s equality [8 Thm. 26.10]; there it is stated over a perfect base field, here the prime field of , with the kernel of in place of ; the two agree by the Jacobi–Zariski sequence of , since . It is not used in this paper.
Proof of Theorem 1.2
Theorem 1.5 needs no injectivity of the conormal map, so it is applied first. Everything else then follows from one computation at the associated primes of . Let be the minimal prime of the theorem, let and be as in Lemma 6.1.
Complete intersection. By Lemma 6.1, where because is separably generated,
Theorem 1.5(2) gives and shows that is generated by an -regular sequence of length ; then . Thus is a complete intersection over . Clearing denominators, which are units of , we may take the generators in . Let
be the -linear map sending the -th basis vector to . It is the composite of the surjection given by the with , so its image is and its cokernel is by (10).
At the associated primes. By Lemma 6.1, , so is free of rank over . Let . By Lemma 5.1(1), is free of rank . Hence the projection splits, and its kernel is free of rank . The map maps onto this kernel, so it is an isomorphism onto it, and is a split injection. Its reduction modulo is therefore injective, so some maximal minor of the Jacobian matrix is a unit in (take if ). Let be the inverse image of in . After renumbering the variables, the -algebra is standard smooth [10 Tag 00T8], hence smooth [10 Tag 00T7], and its local ring at is regular [10 Tag 00TT]. This local ring is , because generate after localization at , hence also after localization at . A regular local ring is a domain [10 Tag 00NP], and has depth zero, so is a field which is essentially smooth over .
Conclusion. The kernel of is a submodule of , and the nilradical of is an ideal of . Both vanish at every associated prime of , the first because is injective and the second because is a field. As in the proof of Lemma 5.1(2), both are zero [10 Tag 02M3, Tag 0587]. Thus is reduced, every associated prime of is minimal, which gives the assertion on the minimal primes, and is injective. Since factors through the surjection , both maps are isomorphisms, and (6.1) becomes the exact sequence
Hence . No separability is used after the first step. □
Characteristic zero and perfect fields
In characteristic zero there is no hypothesis on generic points in Theorem 1.1, so one must show that the local rings of at its generic points are fields. They have depth zero, so their modules of differentials are free, and in characteristic zero freeness of the module of differentials characterizes regularity.
Proof of Theorem 1.1. Let be a minimal prime of and its inverse image in , so that . The ring has depth zero, so is free by the Auslander–Buchsbaum formula. Since , the local ring is therefore regular [10 Tag 00TX], hence a domain [10 Tag 00NP]. Being of dimension zero, it is a field. This field is finitely generated over , hence separably generated since [10 Tag 0322, Tag 030O]. Apply Theorem 1.2. □
Proof of Corollary 1.3. Let be a minimal prime with reduced. A reduced local ring of dimension zero is a field, and is finitely generated over as a field. Every extension of a perfect field is separable [10 Tag 030Y, Tag 030Z], and a finitely generated separable extension is separably generated [10 Tag 030O]. Apply Theorem 1.2. □
Remark 6.2. The hypotheses of Theorem 1.2 and Corollary 1.3 cannot be dropped. For every field of positive characteristic the ring
has , because the differentials of the three equations vanish, and it is not a complete intersection, because its defining ideal has height two and three minimal generators. For the residue field of its unique minimal prime is separable over ; so the hypothesis of Theorem 1.2 concerns the local ring and not only its residue field, and reducedness cannot be dropped in Corollary 1.3.
The imperfection-degree threshold
Theorems 1.1 and 1.2 settle the conjecture when some generic point is separably generated. It remains to determine what happens for reduced algebras over an imperfect field, where the Jacobian image can be too small.
One element short
When no generic point is separably generated, Lemma 6.1 still bounds from below by . If this dimension is at most one, the regular sequence of Theorem 1.5 is at most one element short, and Theorem 1.5(3) applies.
Corollary 7.1. Let be a field and let be a non-trivial reduced local -algebra essentially of finite type with . If for every minimal prime of , then is a complete intersection over .
Proof. Use the notation of §6.1. Since is reduced, every is a field and consists of the inverse images of the minimal primes of . Then . Lemma 6.1 gives
for every . If , Theorem 1.5(2) shows that is a complete intersection. Otherwise and
so for every . As is regular, this says that for every associated prime of the -module , and Theorem 1.5(3) applies. □
In positive characteristic the dimension of is bounded by the size of a -basis of : the Jacobi–Zariski sequence of embeds into , and has a basis indexed by a -basis of . This explains the form of Theorem 1.4: has dimension at most one for all extensions exactly when .
Proof of Theorem 1.4, (1)⇒(2). Write with , so that by the description of differentials through -bases [10 Tag 07P2]. Let be a field extension. Every extension of the perfect field is separable [10 Tag 030Y, Tag 030Z], so , and the Jacobi–Zariski sequence (11) of gives an injection
Hence
Thus every reduced as in (2) satisfies the hypothesis of Corollary 7.1, which proves (2). □
A domain with free differentials
For the converse we need, over every field with , a reduced algebra with differentials of finite projective dimension which is not a complete intersection. The construction rests on one observation: in positive characteristic a polynomial in -th powers has zero differential, so equations in -th powers do not affect the module of differentials. Over a perfect field such equations produce nilpotents, as in Remark 6.2. Over an imperfect field one can shift -th powers by constants which are not -th powers and obtain domains. Two -independent constants suffice to transport the monomial curve , which is not a complete intersection; with only one, has dimension at most one and Corollary 7.1 applies.
Proposition 7.2. Let be a field of positive characteristic containing two -independent elements , for instance . In the polynomial ring put
let and , . Then is a one-dimensional local domain over , it is not a complete intersection, and
Proof. The monomial curve. For independent variables the homomorphism
is an isomorphism onto : the three relations reduce every monomial to a -linear combination of , and the images of these normal forms have exponents in distinct residue classes modulo three, so they are linearly independent. This holds for every , including .
Primality and dimension. Since , the ring is free of rank over , and by (16)
which is free of rank over because the three equations are monic in separate variables. In particular . Let . The generic fibre is a field if and only if , , and are -independent over , that is, if and only if their differentials are linearly independent in [10 Tag 07P2]. Since is a purely transcendental, hence separable, extension, the map is injective and , so , , are linearly independent. With respect to these three elements the differentials of , , have coefficient matrix
Its determinant is a nonzero polynomial in the transcendental element in every characteristic ( for , for ). Hence the generic fibre is a field of degree over , and the free -algebra embeds into it, so is a prime ideal of height two.
The maximal ideal. The residue ring is the field , so is maximal with , and is a regular system of parameters of the regular local ring . One has , so and is a one-dimensional local domain with .
Not a complete intersection. Let and write for the initial form of in the associated graded ring . The initial forms of the three generators are , and (note since ), which are linearly independent in . Since , the three generators are linearly independent modulo , so . An ideal of generated by a regular sequence of length has height and is minimally generated by elements, so is not generated by an -regular sequence, and is not a complete intersection [8 Thm. 21.2(ii)].
Free differentials. Each generator of is a polynomial in , and in . Hence the conormal map is zero and , proving (15).
Proof of Theorem 1.4, (2)(1). If , then contains two -independent elements , and the ring of Proposition 7.2 is a reduced local -algebra essentially of finite type with free module of differentials which is not a complete intersection. Hence (2) fails. This includes fields of infinite imperfection degree.
Shanghai Institute for Mathematics and Interdisciplinary Sciences (SIMIS), Shanghai 200433, China
Email address: [email protected]
Fakultät 8 – Mathematik und Physik, Universität Stuttgart, Stuttgart, Germany
Email address: [email protected]
References
- [2]D. A. Buchsbaum and D. Eisenbud, Some structure theorems for finite free resolutions, Adv. Math. 12 (1974), 84–139. doi:10.1016/S0001-8708(74)80019-8.DOI
- [3]B. Briggs, E. Grifo and J. Pollitz, Bounds on cohomological support varieties, Trans. Amer. Math. Soc. Ser. B 11 (2024), 703–726. doi:10.1090/btran/182.DOI
- [4]B. Briggs, Vasconcelos’ conjecture on the conormal module, Invent. Math. 227 (2022), no. 1, 415–428. doi:10.1007/s00222-021-01070-0.DOI
- [5]D. Eisenbud, Homological algebra on a complete intersection, with an application to group representations, Trans. Amer. Math. Soc. 260 (1980), no. 1, 35–64. doi:10.1090/S0002-9947-1980-0570778-7.DOI
- [7]J. Herzog, Homological properties of the module of differentials, in: Atas da 6ª Escola de Álgebra (Recife, 1980), Coleção Atas, Sociedade Brasileira de Matemática, 1981. Reprinted with an appendix by B. Briggs and S. B. Iyengar as arXiv:2502.14159.
- [8]H. Matsumura, Commutative ring theory, Cambridge Studies in Advanced Mathematics 8, Cambridge University Press, Cambridge, 1986.DOI
- [9]C. P. Mok and Y. Xu, A counterexample to the Eisenbud–Mazur conjecture, preprint, Hexagon 2610.00384v1, 2026.hub.valency.io/works/W_yyc6af9c
- [11]J. Tate, Homology of Noetherian rings and local rings, Illinois J. Math. 1 (1957), no. 1, 14–27. doi:10.1215/ijm/1255378502.DOI
- [12]W. V. Vasconcelos, On the homology of I/I², Comm. Algebra 6 (1978), no. 17, 1801–1809. doi:10.1080/00927877808822322.DOI