NOTRE DAME MATHEMATICAL LECTURES Number 10 TOPICS IN LOCAL ALGEBRA LECTURES DELIVERED AT THE UNIVERSITY OF NOTRE DAME by DR. JEAN DIEUDONNE Professor of Mathematics, University of Nice, France and Supple: and Supplen DR. MARIO BORELLI Assistant Professor of Mathematics, University of Notre Dame NOTRE DAME, INDIANA 1967Copyright © 1967 UNIVERSITY OF NOTRE DAME PRESS Notre Dame, Indiana Library of Congress Catalog Card Number: 67-90420 Printed In the United States of America‘PREFACE ‘These are notes of lectures which I gave at the University of Notre Dame during the fall of 1966; they have been written and prepared for publication by Dr. M. Borelli, Assistant Professor at the University of Notre Dame, whom I heartily thank for the care with which he has accomplished his task and the many hours he has devoted to it. The lectures were intended as an introduction to modern algebraic geometry, in order to familiarize with some of its most important concepts mathematicians who have had no previous contect with that theory. The scope of the book prevented me from giving anything like a complete exposition, and I have accordingly sup- pre: ed a large number of proofs, all of which can be found either in Bourbaki's "Algépre commutative" or in Grothendieck “zléments de Géométrie algébrique". On the other hand, the proofs which are given have been made as explicit as possible, and Dr. Borelli has teken great pains to spell out many details which would be taken for granted by anybody having some familiarity with the material. As the title indicates, the concepts which are studied are those which have to do with the properties of an algebraic variety (or scheme) at a point, or equivalently with the local ring of the scheme at that point, The most important of these concepts are dimension, depth, regularity, normality and completeness, and they are most of the time studied for noetherian local rings. In the study of the dimension of @ module M over a noetherian semilocal ring A (§1 and 2) we prove the Krull-Chevalley-Samuel theorem, which gives three different interpretations of dimension, namely as the Krull dimension, as the leading coefficient of theHilbert-Samuel polynomial, and es the smallest number of elements XyreeeeX, Of A such that the Module M/x, M+...4x,M has finite length. The general form of the Hauptidealsatz is proved in §2. §3 4m devoted to the notion of depth and the study of the properties of Cohen-Mecauley rings. Regular rings are defined in §4. Here, in addition to giving the usual definition and properties of reguler local rings, we characterize the regular locel rings of classical Algebraic Geometry as those rings whose corresponding points are simple, i.e, the corresponding Jacobian matrix has maximal rank. ‘The cohomological dimension of a ring is defined, and the Hilbert Serre theorem concerning it 1s stated (but not proved), In this same §4 we characterize reduced and normal noetherian rings, the latter characterization due to Serre. §5 concerns itself with the behavior of the above mentioned notions under local, flat morphisms, and in §6 we apply the results of §5 to the study of the completion and normalization of @ noetherian local ring. ‘The main results of §6 are Cohen's Structure Theorem for noetherian complete local rings, and Negate's theorem that every noetherian, complete, local integral domain ts Japanese. The notes end with the definition of Grothendieck's excellent rings and the statement of the theorem that localizations of finitely generated algebras over excellent rings are again excellent. I have tried to give to the notes a geometrical flavor, in as much as possible, by examining, with examples and figures, most of ‘the above notions in the context of classical Algebraic Geometry over the complexes.PresiCG Prerequisites... -. + Geometric Notions. . . - Introduction... ee §1. 6: $3. Se. §5. 5A. 5B. §6. 6a. 6B. Dimension Theory. . CONTENTS Hilbert-Samuel Polynomial Depth cise Regular Rings . . . Behavior Under Local Homonorphism Behavior of dimension . Behavior of depth. .... Completion and Normalization. Completion. .... Normalization... 10 ub 20 44 57 79 80 87 100 100 + 108PREREQUISITES ‘The essential prerequisites for these notes are contained in Bourbaki, "Commutative Algebra", Chapters I through IV. Results from Chapters V through VII will sometimes (but not often) be referred to. We shall denote them throughout by B.C.A., 80 that when we write, say, Proposition 4, B.C.A., III, 3, 2 we mean proposition 4 to be found in Bourbaki's “Commutative Algebra", Chapter ITI, §3, no 2. We begin by recalling some of the elementary fundamental notions of Commutative Algebra and modern Algebraic Geometry. No attempt at proofs will be made here, most proofs being available either from the above mentioned chapters of Bourbaki, or from Grothendieck's EGA, We consider only commutative rings A with unit element, and only ring homomorphisms such that 1a» 1. Unless otherwise specified, the rings considered will be noetherian. This means that the set of ideals of A satisfies the ascending chain condition, or equivalently, that every ideal of A admite a finite basis. deals. If A has © unique maximal ideal (when no danger of ambi- guity exists, ideal will always mean proper ideal), A is said to ve @ local ring. We call A a Jacobson ring if every prime ideal pC A is the intersection of the maximal ideaiscontaining tt, p= Cm. ‘The radical of A, rad(A), 1s defined as the intersection of all the maximal ideals of A, rad(A) = om. mck ‘The nilradical of A, tL(A), 18 the intersection of allprime ideals of A, #(A)- Pp . H(A) te easily seen to p prime consist precisely of the nilpotent elements of A. When 4A) = (0) 4.e. when A has no nilpotent elements, A is said to be reduced. If A 4s a Jacobson ring rad(A) = #(A), but already when A is a nontrivial local ring (1.e. not a field) rad(A) = ¥ 4(A) in general, where mm denotes the unique maximal ideal of A. One result which will be used often 1s the following Nakayama's Lemma. Let A be a ring, M, N two finitely generated Amodules, Let u:M +N be an A-morphism, and let be an ideal of A with w C rad(A). If u@ id,j iM @ (A/@) > N @ (A/a) 1s surdective. so is u. Iet A be a ring, S a multiplicatively closed subset of A. On the set-theoretical product AxS define the following equivalence relation (a, 8) ~ (a, 8!) > there exists s" ¢S with s"(as! - ats) = 0. One easily checks that the following operations (a, 8) + (at, 8!) = (ae! + ats, 28!) (a, 8) + (at, 8!) = (aa', ss’) define a ring structure on the set of equivalence classes of AxS. We denote such ring by Ag, and call it the localization of Aat S, We denote the equivalence cless of (a, s) by a/s. ‘The homomorphism t:A > Ag defined by +(2) = as/s (for any s € S) turns Ag into an A-module. We caution that t need not be injective.Let M be an A-module, We define My = Ay @ gM. It 1s easy to check that My can be obtained also by repeating verbatin the above procedure for the construction of Ag, simply substituting M for A. In the category of rings and ring homomorphisms, A, can be more simply defined es follows: the localization of A at S consists of a ring 0, and @ homomorphism p ¢ Hom(A, C) such that for all rings B the function Hom(¢, B) + Hom(A, B)! is bijective, where Hom(A, B)' consists of all those morphisms u ¢ Hom(A, B) such that ell elements of u(S) are units in B. In most applications to Algebraic Geometry the set S is of one of two types. In the first, S consists of the non negative powers of an element t ¢ A, and we write A, instead of of A. In this case we use the notation Ap instead of Ay 5 « et M be a finitely generated A-module, We define Ass(M) = {Pp a prime ideal of A | p is the annihi- lator of some x eM, x ¥ O} Supp(M) = (Pa prine idea of A| Ap © gk xO). Aes (M) 16 @ finite set when A is noetherian, and is re- lated to Supp(M) by the following property: the minimal primes of Ass(M) coincide with the minimal primes of Supp(M). We call support of MIn the case that M= A the following statements are true: 1) Ase(A) = the prime ideals (1solated and imbedded) corresponding to (0). 3) {the minimal primes of Supp(A)} = {the isolated primes of (0)} = {the minimal primes of a}. 4) We give some examples of the above notions, again without any attempts at proofs. The local rings most commonly met in Algebraic Geonetry are of the form Ap where p is a prime ideal of A. It is inmediate to check that the complement of pA, in Ap consists of unite, whence pAp 1s the unique maximal ideal of Ap. ‘An example of @ Jacobson ring is aiven by A/p . where A is a finitely generated algebra over an algebraically closed field k, and p is a prime ideal of A. Finally, we leave es an exercise to the reader to prove that, 1f Me a finitely generated A-module with annihilator #, then Ass(M) = {the prime ideals corresponding to an irredundant primary decomposition of a}. GEOMETRIC NOTLONS Tet A be a ring. We recall that Spec(A) is defined, as a set, to consist of ell the prime ideals p of A. Such set is made into @ topological space by defining a subbasis of open sets(which actually turns out to be a basis) as follows: we define, for t € A, D(t)= { pe Spec(a) | t¢ pt, and consider the collection (D(t)),¢q a8 the subbasis in question, (That it is a basis is easily seen from D(st) = D(s) ND(t), s, t € A.) The resulting topology on Spec(A) is usually called the Zariski topology. Equivalently, we can define the Zariski topology by determining what the closed subsets are, Here we take any ideal @ C A and define V(a) = { pe Spec(A) | @C p}. The collection of sets (V(#t)) is easily seen to satisfy the axioms of closed sets in a topology, and one then shows that Spec(A) - D(t) = V(tA) Spec(A) - v(@) = L)o(t) fe whence the two topologies are actually the same. Since A is noetherian, Spec(A) is a noethertan topological space, i.e. the open subsets of Spec(A) satisfy the maximal condition, or, equivalently, the closed subsets of Spec(A) satisfy the minimal condition. Hence Spec(A) is the finite union of its irreducible components. We caution that Spec(A) is however highly non-Hausdorf?. In fact one eastly sees that CB => view.) DV(S), hence a point p € Spec(A) has in general a closure distinct from p, in fact equal to V(p). p 1s hence a closed point if, and only if,the ideal p 1s maximal, However, given two distinct points Pp, Y of Spec(A), we can find an element t ¢ A which belongs to one but not the other of the two ideals (we can't tell which though), whence an open subset D(t) which contains one point but not the other. In other words Spec(A) is a 7, (Kolmogoroff) topological space. We also remark that the only closed, irreducible compo- nents of Spec(A) are precisely the closures of the minimal prime ideals of A, and that the only closed irreducible subsets of Spec(A) are precisely the subsets of the form V(¢L), where 0 is any ideal in A with a prime radical. In fact V(e) = V(va@), for oC A. With every ring A we have made correspond a certain topo- logical space, Spec(A). We ask the question: given the topolo- gical space Spec(A), can we recover A? Unfortunately not, since, e.g., all fields have honeomorphic Spectra. The notion which is missing. in order to obtain an adequate dictionary between the algebraic and the geometric languages is the notion, due to Serre, of the sheaf of local rings of Spec(A). This is a sheaf A which can be defined in one of two equivalent ways 1) As a presheaf A(D(t))= A, te A 2) As an espace dtale’, the stalk Ap of A over the point p € Spec(A) is given by Ap = Ap. One can easily prove that Ay = lim Ay, where the houomorphisms tep Ag 7 Ag are given by a/c" at"/(st)", Hence the two definitions are indeed equivalent. We now have associated with every ring A two objects,namely the topological space Spec(A) and the sheaf of local rings A over Spec(A). Given the pair (Spec(A), A), it is now easy to recover A, namely A = A, = A(D(1)) = A(Spec(A)), which is the totality of sections of A over Spec(A). The pairs (Spec(A), K) are the objects in the category of affine schemes, whose morphisms we now discuss. To describe the morphisms in the category of affine schenes, let (Spec(A), A), (Spec(B), B) be two objects in the category. Let 9:A +B be a ring homomorphism. Over Spec(A) we define the sheaf of rings 9,(B), given by 9,(5)(D(t)) = Bact) 7 BD(e(t))), tA, Then the function 9*:Spec(B) + Spec(A) given by 9%(p) = 9 *(p) 4s continuous, as te seen from the formula (PY *@(t)) = D(@(t))- Furthermore define 9:A + o,(B) by defining @(D(t)) say > Bo(t) 28 follows: a/t? ames 9(8)/p(t"). ‘Yo tne ring nomomorpnism @ we nave associates a pair or functions (9*, $). Such pairs are precisely the morphisms in the category of affine schenes. Our dictionary is now adequate, since in fact one can prove that the category of affine schemes is the dual (in the categorical sense) of the category of rings. APPENDLX Tet € be the field of complex numbers, R= €[X,,-...X,], @ an ideal of R such that @ = v@ . Define V(a@) as follows V(0e) = (0), ---.%,) © OP | F(xy,---5%,) = 0 for all f ewe}.This is the classical notion of an affine variety (in fact, to be strictly classical one should take @ to be a prime ideal), and it is well known that the points of V(#) are in a 1-1, onto correspondence with the maximal ideals of the ring A = R/#. So the classical notion of an affine variety corresponds simply to the set of closed points of Spec(A). In defining Spec(A) as we have, we have in fact added to the classical notion of point a lot of other "undrawable" points, namely the prime ideals of Avwhich are not maximal, We can ask: a) What are the advantages of such addition? b) If such addition is indeed advantageous, how could classical geometers get along without 1t? The answer to b) is simple: R/@ is a Jacobson ring, and the knowledge of its maximal ideals determine its prime ideals. observations: 1) We are not limited to rings of the form R/& , and, were it so, &% can be arbitrary, whence R/@ may have zero divisors (Serre's point of view) and, more strikingly, nilpotent elements. 2) Prime ideals have a "good" functorial behavior (e.g.,the inverse image of @ prime ideal under a ring homomorphism is egein prime), while maximal ideals do not. 3) The notion of a "ringed space", i.e. a topological space X and a sheaf of rings over X, is the natural tool to give an intrinsic geometric definition of projective varieties (which are definitively not affine}4) The possibility that Ré have nilpotent elements has Drought the solutions of long standing conjectures, unsolved until now, ‘There 1s one notion that seems to be lost in the transi~ tion from the classical case to Spec(A). In the classical case an elenent of A identifies a regular function over V(e), with a well defined vaiue f(x) ¢ €, at each x € V(oe). Can an element f € A be considered as @ function over Spec(A)? Most definitely, but the value field may change with the point P © Spec(A). More precisely, Ap /pA, 18 a f1eld, which we denote by k(p), and we define the value of f at p as the image of £/1 under the canonical morphist Ap >Ap/pAp. It is trivial to see that, when A= R/@ , and p is a maximal ideal of A, then k(p) = €, which throws a better light on the classical situation. We point out that, if f ¢ A is nilpotent, we have the highly non-classical situation of getting t(p) = 0 for all pc Spec(A), but f #0.INTRODUCTION The content of these lectures will be the study of some of the most significant properties (from a geometrical point of view) of local rings. We are limiting ourselves to local rings because, as it appears from the prerequisites, we shall be able to describe and discuss most of their properties without any need for the notion of abstract scheme, which is considerably more general and deeper reaching than the notion of Spec(A). First a bit of notations. When x denotes a point of Spec(A), by definition x is a prime ideal of A. However, to Gistinguish the instances when we are looking at x as a point of Spec(A) from when we are looking at x as a prime ideal of A, we write in the latter case J, for x. Thus the stalk of K over x is written Ay . We also will write, say, (x, O,) Ik instead of (Spec(A), A), and then the stalk of Oy over x € X Will be written as Oy .. Tet (X, Oy) be an affine scheme, t.e. (X, 0x) = Gpec(A), A) for some ring A. Why do we call the rings aed "local"? From classical topological knowledge one would like to say that, in Oy, there 18 information available about the nature of the neighborhoods of x. This is, in a sense, true, but must be taken with @ grain of salt. More specifically, we have OK," 7 i= ! a Ay» and A, = K(D(t)) = r(D(t), oy). Here we have written lim for "direct limit" and T(D(t), Oy) for the sections of O, over D(t). Hence 0, , gives us as much information about the neighborhoods of x (the D(t)'s), as a direct limit can give about its "preimages". For t ¢ J, we have canonical homomorphisms A, -+A; , hence canonical xun morphisms (in the category of affine schemes) (spec(As ), ay) — (Spec(A,), Ay) = (D(t), O,1D(t)). Hence, keeping in mind the duality (in the categorical sense) of the category of affine schemes and the category of rings, we have (Spee(Ay Ys Ay) = A (4), O10) and the menber on the right 1s C\ D(t). In this case, however, hy A Dd(t) ¥ x, An fact equals tH, So, while the term local is somewhat justified, it is definitely not to be understood to mean “a property holding La the local ring of a point x holds in a neighborhood of x". What is more likely to happen is the following: we have a morphism (@, ):(X. Oy) + (¥, Oy) of affine schemes. A certain property holds both for Oy , and Oy, 9/,)- Then there exists a neighborhood V of x such that the property holds doth for Oy <1 and Oy g(r)» if x! ranges over V. What is, then, the information available in the space Spec, z© Tet us look at some examples. Recall, first of all, that the prime ideals of A, are ina 1-1, onto correspondence with the x prime ideals of A contained in J,. Hence, as a set, Spec(A, ) cs is in a 1-1, onto correspondence with the irreducible closed subsets of Spec(A) containing x. 1) Spec(k), where k is a field, is quite simple. It consists of one point.a2 2) Ay 1s a diserete valuation ring. Here spec(A; ) x x consists of two points, one of which, x, (the maximal ideal) is closed, and the other (the (0) ideal) is open and generic . 3) Aw €(X, Y], d= XA + YA. Here Speo(A ) has (0) as generic point, 3,8, a6 closed potnt, and ali other points are given by prime ideals of the form f(x, ¥). Ay» where £(X, ¥) 1s an irreducible elenent of A such x that 2(0, 0) = Tet R €[X, Y] and consider the following three cases. R/(x? - x9 - x?) Rs J, = EA + FA 1 1) A (Here ¥, ¥ denote the images of X, ¥, under the canonical morphism R + A.) 2) A= R/(x? - x3)-Rs 5, = KA + YA. 3) A= R/(K- T)Rs Jy - RA + Ta. The "geometricel" picture of Spec(A) in these cases are as follows (here only one point of Spec(A) is " ndrawable", i.e. prime but not maximal: the generic point (0)): Y Y y Case 1 case 2 Case 3 In all three cases the ideal j, is maximal tn A and is represented by the origin in the figures. Now, geometric intuttion tel1s us that, with respect to Spec(A), the origin33 has dirrerent properties in each case. However Spec(A, ) 18 x the same in all three cases 1.e. consists of two points, with one open, generic point, and the other closed. To differentiate the three cases one must hence look at the inner properties of local rings, 1t is Just not sufficient to lock at the mace Spee (A, Ix In the category of rings, locel rings from a subcategory. However, were one to take this point of view, one would get a lot more morphisms between local rings than one desires. Iet us consider what happens when we have a homomorphism 9:4 9B of arbitrary (i.e. not necessarily local) rings. If 4 © Spec(B) and p= g*(q), we have canonically a morphism p 7 3y given by a/s =» 9(a)/(8)- However @ has an additional property: aha By )= Pap. ‘This is the property one wants to have for morphisms of local rings. In short: ‘The category of local rings and local morphisms is described by: 1) The objects are local rings. 41) The morphisms are locel morphiams, i.e. the inverse image under A +B of the unique maximal ideal of Bis the unique maximal of A. E.g. The injection of a local ring with no zero divisors into ite field of fractions is not e local morphism, The category of local rings 1s not a very good one. E.g. it lacks products, it is not closed under finite extensions (i.e. @ finite extension of a local ring is not a local ring in general. It te in fact @ semi-local ring), and, ifm denotes4 the unique maximal ideal of A, Spec(A) -™ is a scheme, but not affine (that it is a prescheme is seen by Spec(A) - t = U_ v(t). tem We shall hence study the inner properties of local rings A. More specifically, we shall study: 1) Dimension theory. (Dimension. Depth. Regularity) 2) Behavior under local morphisms (Flatness, Ascent, and Descent) 3) Operations on a locel ring (Completion, Normalization, Hensclization) 4) Stability under the operations in 3. (Excellent rings) Most of the topics covered will be found, under different treatments, in M. Nagata's book "Local Rings", or J.P. Serre's Algébre locale, Multiplicites, Springer-Verlag, 1965, or E.G.A. IV. We again remind the reader that we shall limit ourselves to noetherian rings. §1. DIMENSION THEORY — GENERAL NOTIONS Iet Ade a ring. The prime ideals (Pg, Py,-.-, Py) of A are said to form a chain of length n tf PyC Py C ++. C Pps tb Definition 1.1. (Krull) The dimension of A, dim(A) is equal to the 1.u.b. of the lengths of the chains of prime ideals in A, Clearly dim(A) need not be finite. For example, if15 A= ROX), Xge-+/%q++] there are clearly chains of arbitrary length. In fact, even when A is noetherian, an example of Nagata shows that dim(A) need not be finite. It is,however, if A is a local ring. (See theorem 2.3 ahead) Definition 1.2. Let p ¢ Spec(A). Then we define dim V(p ) = aim(A/p ) Coaim Vip) = dim(Ap ) Proposition 1.1. @) dim V(p) © dim(A); bd) Codim V(p) © dim(A)s c) dim V(p) + Codim V(p) © aim(a). Proof: We have two canonical morphisms A Apis Aap and we immediately get a) from the first,b) from the second. Note that a) and b) hold also when the left-hand sides are =. Hence c) nolas ir e1tner or tne summanas on tne 1ert 1s ™% Now, any chain in A/p gives rise to a chain of equal length in A, of prime ideals containing p , and any chein in Ap gives rise to a chain of equal length in A, of prime ideals contained in p- Furthermore, we may assume that the chain in A/p of length aim(A/p ) start with (0), end the ones in Ay of length dim(Ap ) ends with pA, . Hence the corresponding combined chain in A consists of (dim V(p) + Codim V(p) + 1) distinct prime ideals, which proves c). Equally simple is the proof of the following two state- ments, proof which we leave to the reader.1) If of is any ideal of A, aim(A/@t) 3 aim(A). 2) If # 4s not contained in any minimal prime ideal of A, then dim(A/o) < dim(A). let pry € Spec(A), pC y. Achain pCP,C...Cy toe F is called a saturated chain connecting p and Y if its length cannot be increased by insertion of some prime ideals. Definition 1.3. If, for all pairs p,Y © Spec(A), all saturated chains connecting P andy have the same length, A is said to be a catenary ring. ‘An example of Nagata shows that noetherian local rings need not be catenary. Proposition 1.2. Let A be an integral local ring. ‘Then 1) If A ts catenary for all p ¢ Spec(A), Gim(A) = dim(Ap ) + aim(a/p ). ii) A ts catenary if, and only if, for all p,g © Spec(A) with PC 9, dim Ag 7 dim Ap + dim(Ay /P Ay se Proof, 1) Since A is an integral local ring, the follow ing statements hold: a) A/ps Ap are integral local rings, hence a2 dimensions involved are finite. b) Any chain in A of length equal to dim(A) 1s @ saturated chain connecting (0) and tt, (‘, denotes the unique maximal ideal of A). ¢) Statement b) above holds for Ay and A/p. ote that17 ay ~ Pap /p Statement i) now follows imediately froma), b), c) and Hy, = M,(A/p). above. 11) We begin vy observing that, 1f A is an arbitrary catenary ring, and p ¢ Spec(A), then Ap and A/p are catenary. exist between the prime ideals of Ap and A/p respectively, and the appropriate prime ideals of A. Let now p,q € Spec(A), PCY and A an integral, local, catenary ring. Then Ag is a local, integral catenary ring, and we may apply 1) to the ideal pag. So dim(Ag ) = dim(Ay /pAg ) + dim((Ay ) 9 9 Phy Y) pay The morphism 9:(Ay ) Ap given by 9((a/s)/(b/t)) = phy 7 *P atins, 96 8, sh ey, dp ta well defined (he fp) and easily seen to be an isomorphism. One part of ii) is proved. Zo prove the converse, we observe first that any saturated chain, in A, connecting p andy gives rise to a saturated chain of equal length in Ay /pAg connecting (0) and qq /piy . Hence the length s of any saturated chain in A connecting p and y is at most r= dim(ag/pay ). We assert s =r, When r=, 1 the assertion is trivially true, and we proceed by induction on r. Let POPC C Peal re be a saturated chain of length s in A connecting pp and 9.18 We have dim(Ag / py, Ag) = 1. Row din(A pe Pipe = dim(A P,.1) - aim(A 5) din(Ay ) - aim(By / Py Ay) ~ atm(Ap) = din(Ay /p Ay )eler By induction s = 1-r 1 and we are done. If g:A +B is a homomorphism, B can be considered as an A-algebra by a-b = 9(a)-b. We say that B is integral over A 4f every b ¢ B satisfies an equation of integral dependence n m1 = over A, de, D™ +a DTI +...409 = 0, a eA, n>0. Theorem 1. . (Going-up theorem). Let @:A »B be a ho nomorphism, B integral over A. Then 1) dim(B) = aim(A) (ame going-up theorem) . 41) If @ 4s mono, dim(A) = dim(B). Proof, i) let y be @ proper prime ideal of B. We assert: a) oy) ta bv) o (a) + ker(o) ir g + (0), and B is an integral domain. a) is trivial, since g(1) = 1 and @ is proper. To prove b) assume g” (9) = ker 9. Then MANY = (0). let beg, d+0. Tet teat Oy = 0 0 ve an equation of integral dependence of minimal degree. Now Gy € In(A) and clearly (, €Y. Hence Cy = 0, and4g ne n-2 vit 4o. 1 ve +...) = 0. this 19 @ contradiction, since B is an integral domain. To prove 1) from a) and b), let p C y be prime ideals of + B. From A->B-B/p we see that B/p ts an integral domain, ge integral over A, and that ot(p) = Ker(ce 9) og) = (coey* (g-B/p) ana gB/p + (0). Henee, from b) abave wage ol (p), and 1) fortows. Not : 4) holds under the weaker assumption that B is algebraic over A. 11) Let p g f be prime ideals of A. By theoren 1 of Chapter V, 2 of B.C.A., there exists a prime ideal Pt in B sucn tnat ae = p+ tren o(p)C pi, the morpnisa g'A/p + B/pt is mono, and B/p’ 1s integral over A/p . Now g(A/p) + (0) is a prime ideal of A/p, and hence there exists a prime ideal gy" of B/p' such that ary") = 9(8/p). We have 9" = 71 -B/pts where gy! 1s a prime ideal of B, and clearly oy") = . Since 9 (A/p) + (0) and 9! is mono, we have ay" + (0), whenes 9! ? p+ Tate implies dim(A) = dim(B) whence it) follows. Definition 1.2, gives the notion of dimension for an irreducible closed subset of Spec(A). We extend this notion to20 arbitrary closed subsets by the formula aim(V(a)) = dim(a/ee) where o is an arbitrary ideal of A. If M is a finitely generated A-module we define éim(M) = dim(Supp(t)) = dim(A/ann(0t)). Here we use the fact, mentioned in the preliminaries, that Supp(M!) 4s the closure in Spec(A) of Ass(M), and Ass(M) consists of the prime ideals associated to ann(M). If NC M is another A-module we see trivially that ” aim(x) éim(M) 4im(M/y) = aim(M) In fact ann(W) D ann(M), ann(M/y) D ann(M). A non-trivial statement, proved in Bourbaki's, chapter IV, §2, Theorem 1. dim(M) = 0 if, and only if, M has finite length, in the composition series sense. §2. HILBERT-SAMUEL POLYNOMIAL Iet H be a graded ring, i.e. where H, are (additive) groups and hyhy © Eyige £07 h,¢ Hy, h,¢ H,. Clearly H, is an Hymodule. We assume: a) Hy is an artinian ring >) H 1s generated (as an Hy-algebra) by finitely many elements of Hy.a1 An Hemodule M is called graded if M= © M,, where M, are H,, n modules and FM C Maps If Mis @ finitely generated Hmodule, then M, 1s # flnitely generated H.-module and (since H, is artinian) M, has finite dength. Definition 2.1, The Hilbert-Samuel Polynomial of M, x(M, n) is given by x(n) = length y M, for large a. Of course one needs to prove that x(M, n) is indeed a polynomial. In fact Theoren 2.1. (Hilbert) Let H, M be as stated above. Then there exists a polynomial P(X) € Q{X], which achieves integer values for integer values of X and such that, for all suffictent- ly large n, x(M n) = P(n) Proof: Since H is finitely generated over H, by Hy, we have a homogeneous epimorphism (of degree 0) g By Oyr-+0X,] SH 0 and M becomes a finitely generated Hy[X,, +K,]J-module, Now length y M, 18 independent of whether we consider M as an H- ‘0 module or an Hq[%,,+..,X,]-module (since c is onto). Hence we OX, 1. We proceed by induction on r. When r= 0, H= H, and, since may assume H= H[Xy,.Mis finitely generated by, say, m, ¢ M, , we have M, = 0 if n= mex {a}. Hence x(M, n) = 0 for n sufficiently Large. 1 Tet opi + M be elven by 9,(m) = Xm. Then 9, 1s a homogeneous morphism of degree 1 and we have O3N> aS M+C+0 ee Since length y (-) is an additive function we have 0 x(M, ne1) - xQM, mn) = x(C, #1) - X(N, n) For nN, c € C we have X,-n= 0, X,+c = 0, hence N and C are HDs »%,_1] modules, and, by induction, x(C, n+l) - x(N, n) is a rational polynomial inn, for sufficiently large n. A standard argument now shows that x(M, n) is also a rational polynomial, for n sufficiently large. For the remainder of this section we assume that A is a noetherian, semi-local ring. Definition + let y be an ideal of A. We say that y ts an ideal of definition of A, if the ring A/y is artinien. We recall here that a ring A is called artinian if it satisfies the descending chain condition or, equivelently, if every prime ideal of A ts maximal. We assert: Proposition 2,1, Let y be an ideal of A, ‘The following three conditions are equivalent. a) Y 4s an ideal of definition of A bd) A/¥ has finite length (in the composition series sense)23 c) Y DW, where wdenotes the radical of A. b) =5 a) is immediate, since A/Y satisfies both chain conditions. a) =>») follows from the fact that an artinian ring 42 alse noetherten. c) ==> a) follows from the following observation: if 7 DW and a prime ideal P contains Y, then P te one of the maximal ideals of A. To see thet a) => c) we observe first, that sinee 4/y 1s artinian, red(A/) + the set of nilpotents in A/q. Wow, clearly, red(A/7 ) = (4), where o:A + A/Y is the canonical epimorphism. If is an ideal of definition of A and M is @ finitely generated A-module, M/Y M is @ finitely generated A/Y module (in fact W/yM = M@ 4 A/4/), hence M/aM has finite length. Theoren 2.2. (Htbert-Semuel) Tet A, 4, M be as above. ‘Then a) fag M has finite length b) length a lttday M) = Py (M, n) is a polynomial in n for n sufficiently large. Proof: We prove a) by induction on n, When n= 1 the assertion is precisely the observation we made previous to the KHL yg g statement of the theorem, Clearly, for all k, a/k/a finitely generated A-module (A noetherian). Hence (W/47M) ® 5 af pager is a finitely generated A-module. The epimorphism ea) ag tryst 5 ok wa x“ given by H@ G~~» @ (here H, J denote the equivalence classes2h ofmeM, qe 9") shows thet * Wy +1 Wis a finitely generated A-module. Finally the exect sequence (4) og" wg") um/g™1 Mo w/e" M0 and the Induction assumption prove a). To prove b) we define ty $4, x sity) ey (A) = 141 = er) = 8 Cg* my where y°= A, Since H, = 4/gy is artinian, H is generated over 2 H, by finitely many elements of Hy = g/g® (any A-basis of y will do) end M! is a finitely generated H-module (any A-basis of M will do); we can apply Theorem 2.1 and get n+l y) = a polynomial in n for n >> 0. dengtn(y™ Wy (we metic n o> 0 fur "s..n sufficiently imree".) From the above exact sequence (#&) we get length (M/y nel yy _ length(/g" M) = Length(g” Way M) or Py (Wn +1) - Pg (i, n) = @ polynomial in n for n >> 0. The theorem is proved. Note: The geometrical significance of the polynomial Py (M, n) was discovered by Serre, and 1t 16 the following. Iet H, M! be as in the proof of the theorem. Let X= Proj(#), Gi= tne sheat over Proj(H) essociated to the graded module M': then for every By Py Qe, n) = S> (2) rength uh(x, F(a). i25 We do not go into further details, except to point out that, for n >> 0 HI(x, feo) = 0, which throws a better light on the somewhat unsatisfactory statement of b), (for n >> 0). let now A, y , Mbe as usual, A filtration M=M, DM) D...OM, 2... 18 called a Y -~ good filtration of Mit yM,CM,;, with equality holding for n En. We assert Proposition 2.2. Under the above hypotheses, for n >> 0 length, (4/M,) = P((M,), n) = a polynomial in n of degree and coefficient of the term of highest degree equaling those of P y (M, n). Proof: As in the proof of theorem, we prove by induction onn that M/M, has finite length. In fact M/M, is an A/y - module finitely generated, and OAM AMM MM, 0 nn Me in and y (M,/Ma41) = 0» whence M,/M,,) is an A/y ~module and has finite length. Consider now the module M, . It is a finitely generated Armodule and Hoan, rength(M, Myyn ) = @ Polynomial in n, for n >> 0. "WM, . Hence, by theorem 2.2 ‘0 The exact sequence 0%, Maeng 7 Wren, > MM, > 0 shows that length(M/M,) is a polynomial inn for n>>0. TheAnelusions nn, 0 2 MC Man Cg ™ MC My Bive exact sequences o Maan ne Mau/gy™Po MM 30 05 go WM, 7 Wen aW gg" Mao OM / YP MoWay MM, +0 whence Py (4. ntn,) = PMaan,, nin.) 2 Py Qt. n) 2 P(QH,), 9). Since Py and P are polynomials, they must have the same degree and the same highest degree coefficient, Q.E.D. Ma finttely generated A-module, Then Py, Fy , are polynomials of the seme degree. Proof: Since rad(y') = raa(y) = W we have (A 18 noetherian) gy? and gy! a9" for some m, Hence Og Wy KW ey Mo Wy M 20 whence P, ,(M, n) © P. q (M, mn) and similarly 7 Py (Mn) =P, 4(M pn) 7 a and the proposition is proved. Definition Tet A, M be given as above. Then deg P, y(which, by the proposition sbove is independent of g) is denoted by a(M). Proposition 2.4, Let A be as usual, and let OM MoM 0 be an exact sequence of finitely generated A-modules. Then, for any ideal Y of definition of A: deglPy (M) - Py (MT) ~ Py (my) € aQer) - 2 $ ac) - 2 1 7 7 Proof: By the Artin-Rees lemma (B.C.A., IIT, 3, corollary 1) the sutmodules MI, =g" MOM of Mt form & 9 - G008 fittration of MI. By proposition 2.2 we have (#) (Mt) and P(jtt,) have the same degree and the same highest degree coefficient. The exact sequence Osg "Mam 2g ha gw 30 gives an exact sequence OM gP MOM 2) gy MoM gy 8 MH 90 whence Py (hn) = By (Hy A) - POMgy ) = 0 or Po (M) - Py (m") - P(M'.) = 0 96 ) 9° Qe) Hence PO) = By OH) = By OH) and, by (*),Py") =P, (M) - Py (M") + @ polyn. of degree at most d(M') ~ 1. 7 The first inequality is proved. The second follows immediately from observing that 0 © Py (Mt, n) = $204, a), for n >> 0, 7 whence deg Py (ut) = deg P. 7 gy Iet M be a finitely generated A-module, and let Vys++e9¥y EW be a set of generators off. Then My M+, With this in mind we give the following: «+ Yy Mis an A/f -module and hence has finite length. Definition 24, We denote by a(M) the smallest Integer k satisfying the following condition: there exist k elements x,,....%, in 1” sucn that M/x, M+...+ x, M has finite length We are now in the position of proving the main result of dimension theory, namely Theorem 2.3. (Krull-Chevalley-Samuel) Let A be a semi— local noetherian ring, Ma finitely generated A-module. Then dim(M) = 4(M) = s(M). Proof: (Serre). We shall prove 1) aim(M) = a(t) ” 2) a(M) = 5(m) 3) s(M) © aim(). We start with the followingIemma 2.1. Let x ¢€W, consider the exact sequence 03 MoM Su Mt +0 where 9(m) = xm. Then 1) 8(M) © s(W/xM) +2 41) Let (Ppy.-++s Pq) denote those points of Supp(M) such that dim(A/p,) = dim(M), 4 = 1, -m, If xt 3 Py then aim(/xt) $ aim(e) - 2 411) deel? y (,M) ~ Fy (W/xu)] 3 a(M) - 1, where y is any ideal of definition of A, Proof: 4) et W = M/at, and let yy)++40y_ eH such that W/y, N +...+ y, N has finite length and k = 5(N). Tha tRemerpht am Wy, Ne vet¥, N > M/XM + yy Meet, proves 1). 11) We start with a word about the P,'s. By definition we have dim(M) = dim(A/arn(yy)+ Tf Py» Pgre+es Pes + 2m, denote the prime ideals associated to ann(M) in A one easily sees that dim(M) = max dim(A/ P,). 1835 Hence the prime 1deals mentioned in the statement of 11) are to be found among the points of Ass(M).30 Wo have to compare dim(A/ann(M/xM)) with dim(A/ann M) Tet Y1,-.+1 Yq be those prime ideals in A associated to enn(M/2i) and such that dim(M/xM) = dim(‘/g’). ‘Then, for some 1,, 154, $+, we have D Py. let tet 15? Pr, oS ne —< Y\, ve a chain of prime ideals of maximal length in A/ann(M/xM), i.e. k = dim(M/xM). The prime ideal % corresponds to prime ideal gy of A containing ann(M/xM) and, from k = dim(M/xM) one sees that Y= Y, for some J. We proceed in steps. Gasel. 9; 2 Pay 4j>m. Then aim(M/xM) = dim A/ 9, 3S dim a/ Pay < dim(M) and 11) 1s proved in this case. gase 2. Y 5D Pay 4y Sm, Then (since x ¢ 93) 952 Ps, and the chain B,C HS s+. shows that dim(M) 2 k + 1 and 11) is proved in this case also. 414). We have two exact sequences 09 MaN st 30 0 MM > M/xK 9 0 Now deel gy (.M) ~ Py (W/stt)) = deal (Py GM) + Py) - Pay (M)) + (Py (06) ~ Py (x) ~ Po crt) and, by proposition 2,2 the right hand side is the degree of the31 sum of two polynomials, one of degree = 4(,M) - 1 § a(M) - 1, the other of degree = d(xM) - 1 5 d(M) - 1. ‘The lemma is proved. Now we return to the proof of the theorem. 1) dim(M) © d(m). We proceed by induction on d(M). 4(4) ~ 0, thon Py (it) = constant, whonce meh length (ug ® M) = Length (/g"*" M) for n >> 0. The exact sequence Og Wa? Hy omy? Ho M/y” M0 nel nel yg, shows lenatn (g® M/g™? M) = 0 whence gy” M= g Now. we take ¥ =W, and then we have © w= (0), whence no vw" M= 0 for n >> 0. Hence M is an A/yr®-module, end since A/w" is artinian, its dimension is 0, whence dim(M) = 0. Hence 1 holds when d(M) = 0. Choose a prime P, ¢ Ass(M) such that dim(M) = dim(a/»,). Since Py is the annihilator of an element m ¢ M, the submodule W= AmC M ts isomorphic to A/P,. By proposition 2.4 we have a(n) = aq) end 1 aim(N) = aim(M) Hence it suffices to prove 1) for N. Let PoC PLC Po +++ C Pp be & chain of maximal length in A, + ot corresponding to a chain of maximal length in A/, (note that n= + is a priori possible). If Py MWC po, then Py DW, whence is maximal (because A is semi-local), a contradiction.32 Choose x ¢ Py OW, xt Po We have N/xN = (A/xA) @ 4, N and, from proposition 18 of B.C.A., IT, $l we get Supp(N/xt) = Supp(N) V(x). Hence P), Pos.++s P, © Supp(N/xN), whence dim(N/xN) 2 n- 1 (in particular, 1f dim(N/sN) 1s finite, so isn). Now trivially the hmomorphism 4/ Py ~ A/ Pq given by F w+ xB is injective, hence N= 0. By lemma 2.1 we get 4(N/xN) $ a(N) - 1 $ 4(M) ~ 2, and by induction dim(N/xN) © 4(N/xN) (and we have proved that n to finite). Now ™ dim(M) = n 3 dim(N/xN) + 1 $ a(N/xN) + 1 a(m) and 1) is proved. We observe nere tnat we nave actually snown aim(m) < + ~ 2) a(M) $ w(t). Let (x4},54 5, be elements of rf such that, letting ¢ = x; A +...+ x, A, we have length (M/@M)<+ © and n= s(M). Let @ = 06+ WMann(M). We nave ann(N/a M) 26, hence the prime ideals tn Ass(M/7M) are maximal, and therefore @ D W™ for some k, 1.0. Y 18 an tdeat of definition of A, Now clearly ¢™ M= o¢™ M, whence By hl ne 1 2 7 Mg of use) Mu. Let 25. z, be a minimal cot of generators of M over A. ‘Then the elements y, Pa igign, vy) +...+v, = mare a set of ’ Z (xy"1 ox, generators of oc” u/oc™ w over Aly. Let rength (a/y) =33 a(a<+ ©since A/y is artinian). Now length (g” Wg M) = length (@™ M/ee™! uy s m= 1 ar. =a polyn. inm of degree n- 1. m1 The exact sequence Ong Wa™) Mo M/g™ M+ H/g™ M0 shows 2). 3) s(t) © aim(m). We proceed by induction on dim(M) (which is finite by 1). dim(M) = 0. Then length (M) < + © (since A/ann M is artinian) and no elements of ware needed to have length (M/x,M+...+%,M) <+. Hence s(M) = 0 and 3) holds, Let n= dim(M)= 1, Let { Pi}ysqsq_ be those elements of Ass(M) such that dim(M) = dim(A/p,). Since n# 1 the p, are not maximal. We assert: n Py w¢ 2 Py. In fect, if WC Pe then, by proposition 2 of B.C.A., IT, §1, we haveWC Py for some 1, a contradiction, since P, is not maximal. Hence we can choose xew,x€ & Py: By lemma 2.1 we have 8(M) = 5 (M/xM) + 1 and din(M/xé) © dim(M) - 1. Hence, by induction 8 (M/xM) © dim(M/xM) and finally s(M) = s(M/xM) + 1 © dim(M/xM) + 1 5 dim(M), Q.E.D.3h Appendix We give a brief description of the geometrical meaning of the three numbers dim(M), s(M), a(M). We admit right off that d(M) is a far-reaching concept leading in particular to certain results of intersection theory, and we shall limit ourselves to a geometrical interpretation of @im(M) and s(M). dim(M) is the simplest or the two. It simply gives the maximal length of irredundant descending chains of irreducible subsets of Supp(M). (Such chains must necessarily terminate with a closed point.) s(M) has a somewhat more sophisticated interpretation. Remembering that Supp(M/xM) = Supp(M) M V(x) and thet length (M) < + © ==> dim(M) = 0 €==> dim(Supp(M)) = 0 => (by above remark) <=> Supp(M) consists of a finite number of closed points. We see that s(M) is the smallest number of “hypersurface: " (the V(x)'s) such that their intersection with Supp(M) 1s zero dimensional. There is a fourth integer that one should introduce in this connection, but which is related to the previous three, in general, by an inequality rather than equality. Iet A be a local ring, m its meximal ideal. The A-module m /m? is Clearly!) annihilated by m, hence / yy? 38 an Rim module, i.e. a vector space over k = A/ML, dim, (M/y 2) is the fourth integer we wich to consider. We aesert: Proposition 2.5. 8(A) © dim(m/,, 2).35 Proof: Let x,).+.), be elements of m such that their equivalence classes (mod m.*) forn a baste of M/y 2 over A/m . We assert that x,,...,x, forma system of generators of @. Tet Mex) A@x,AG..Ox, A Nom end let N be defined by u(ay: Tet M1 =m? C rad(A) = Mm. Now Me AAs k J ¥ + < MOM YO = M2 Hse 1ay, ed ‘m and 9, ¥ are surjective. By Nakayama's lemma we have that u is surjective, which proves that x,,...,x, form a system of generators of ffl. Hence A/x,A +...+ x,A = k and length,(k)<+% Ee (A) Sn (1h /,2)5 .E.D. fence 9(A) $n = rank, (MH/) 2) a We show with an le that s(A) < ri (mM / 2) does en. ie sl an examp: (A) < rank, ( im ) happe We observe first of all that (trivially) any set of generators of M gives rise to a set of generators of m/, 2 over k. Hence din, (M/y 2) = smallest number of generators of Mt. Tet now36 Re (x, V/(y2 _ 53) = el, yl po xR+ yk A= Rp A, m= PAD. We make (without proof) the following assertions: (b), ¢) have easy proofs) a) dimR=1 b) R 4s an integral domain ¢) p ts prime Hence it follows that p is maximal and that s(A) = a dim(A) = 1. But ™ 1s not principal, in fact dim, ,, (*e/,,2) 2. To see this, consider the diegram mae R mOR = poo We see thet dims jp, ("/yq 2) = smallest no. of generators of Mt % 2 (x, y generate m). However, were mprincipal, so would p be. Wow were it so, the inverse image of p under C[X, ¥] +R would be principal mod(¥* - x3), which is easily seen to be impossible. Hence dime(/_.2) = 2. (Note that Afm = €). From dim R= 1 one obtains dim(A) = 1, whence (A) = 1 < dimg (tt /y 2). When the local ring A 1s such that (A) = dim 4. (1/2) we say that A is a reguler local ring.3 he geometrical interpretation of the number Aimy jy (He /yy,2) 16 the following: it is the number of linearly independent linear forms (modulo forms of higher degree). This corresponds to the classical concept of the dimension of the tangent space. If Ais not a local ring, one can still talk about dim(A), and one trivially gets the formula dim(A) = Sup (atm (Am,)) where # ranges over the maximal ideals of A. We give a brief description of the situation when aim(A) = 0, 1. dim(A) = 0. Then A is artinian, hence semi-local. Let wW = nil radical(A). We get AM ~@A/M,, 1.e. AMT isa direct sum of fields. Spec(A) consists of a finite number of eloged potnts, and the Tneal rings are primary rings (1.0. some power of the maximal ideal 1s 0), In fact, since A is artinian, 80 te Agy, + whence (my, Aa)” = (my in n >> 0, and Q (mM, Aq)" = (0). Furthermore we have A= T(Spec A, A) = © Agus which is easily seen from the fact that Spec(A) consists of a finite number of closed points. @im(A) = 1. In this case the prime ideals of A are either minimal or maximal, and there are only finitely many minimal primes, with at least one, say P, such that aim(A/p) = 1. Tf A 4s local, all minimal primes have this38 property. There are infinitely many maximal primes, if A is not semi~local. A typical example of this case are the Dedekind rings, 1.e. noetherian, integrally closed domains A such that every prime ideal pCa, p + (0) de maximal. It follows that all local rings Ap are valuation rings We note however that, while in the case A= 6[X] all local rings Ap are isomorphic, when A= Z we obtain distinct local rings, for distinct p. One can get more one-dimensional examples in the following way: Let A be a Dedekind ring, K ite field of quotients, La finite extension of K. Then any ring B, with AC BC L, is one dimensional (and need not be Dedekind). (Krull-Akizuki theorem, B.C.A., VII, §2.) Other examples are the orders of A in L, 4.e. rings contained in A, with field of quotients L (hence not integrally closed when they are different from A). If A 4s one dimensional local ring which 1s a Dedekind domain (1.e. integrally closed), then A is a valuation ring (See Lang, "Introduction to Algebraic Geonetr; B.C.A., VI). The geometrical interpretation of the notion of Dedekind theorem 1, p. 151, or rings is seen by observing that, if A is a Dedekind domain, Syec(A) consists of one minimal prime and maximal primes whose local rings are integrally closed whence regular. Classically this corresponds to the notion of an irreducible, non-singular curve. Tet A= €[X, YI, and let #(x, Y) © @[X, ¥]. Then a classical statement in Algebraic Geometry is that the irreduc-39 ible components (in the Zariski topology) of the variety of zeros of £(X, ¥) have codimension 1. We generalize the above situation with the following: Theorem 2.4, Let A be a noetherian ring, x,,....x, € Ay Ot =x, A+..44%, A. let p be a minimal prime in Ass(A/e ). ‘Then codim(V( p)) = dim(Ap ) Sn (When n= 1 this is the well-known "Heuptidealsatz"). Broof. We have the inclusions Ay D pay Deady « Since p 1s minimal in Ass(A/¢t), there are no primes of A properly included between p and a, hence Ap [Ody has a unique prime 1deal (namely p(Ay /@&Ay })» and is therefore Artinian, whence of finite length. Now Ap /aA, = Ap/t Ap test %, Ap» whence codin(V(p)) = aim Ay = 8(ap) Sn, QED. Theorem 2.4 is an example of how we can apply our local aumension theory to @ gioval situation. Some final results concerning the notion of dimension: Theorem (Artin-Tate). Let A be a noetherian integral domain, Then the following conditions are equivalent: a) A 4s semi-local of dimension $1 b) (0) 4s an isolated point in Spec(A) c) there exists an f ¢ A euch that Ap is a field. Proof: We give a cyclic proof. a) ==> bd). Since, A 4s integral, (0) € Spec(A). Since A ie semilocal, there are a finite number of closed points, {m,},...,{#,} in Spec(A). Since dim(A) precisely of {(0)}, {m,},...,{m,} and b) follows. 1, Spee(A) consists4o b) ==>) Since (0) 1s isolated in Spec(A), and the open subsets {D(f)}p_q form a basis for the Zariski topology of Spec(A), there exists f ¢ A such that D(f) = (0). But D(f) = Spec Ap, whence Ay has only one prime ideal, namely (0), and c) follows. c) ==>) Let p + (0) be any point of Spec(A). The injection A+ Ay shows, since A, is a field, that L¢ p Ay. Hence f € p. We assert: (*) every minimal prime ideal of A/fA is maximal. In fact, since A/fA ts noetherian, let Py,..., P, be the minimal prime ideals of A/fA, Assume that one of them, say Py, 42 not maximal. Let tM D P, be maxinal, Since Py 1s k wk. IfmC U py, then 1 ‘l= P, for some J, which we have Just shown not to be the minimal, we have #. + Py d= case. Som U Py i.e. there exists g' eM such that Jeu 6) t Py dm teeeeoks Lot g € A ouch that gt = @ + A. Lot Y ve a minimal idesl of Ass(A/eA). By theorem 2.4 Coaim(v(¢)) $1, and clearly Codim(v(y)) ~ 1, since y + (0) and A is an integral domain, Therefore y is a minimal prime of A, hence re and W-A/fA is @ minimal prime of A/tA, 1.2. -A/tR = ' Y-A/th= py for some J. Clearly g ey, hence g! ¢ Py, is a contradiction. Therefore assertion (*) above is proved, and every non zero prime ideal of A is hence maximal. Furthermore the only prime ideals of A are (0) and the inverse images of Piss» Pe Hence A is semilocal and dim(A) = 2.aL Proposition 2.6. Let A be a noetherian semi-local ring, Ma finitely generated A-module, x ¢wW = rad(A). Then dim(M/xM) = dim(M) - 1 and equality holds if, and only if, x belongs to none of those minimal primes p ¢ Ass(M) such that dim(M) = éim(A/p ). Proof: By theorem 2.3 and lemma 2.1 we have dim(M/xM) = 8(M/xM) 7 e(M) - 1 = dim(M) - 1. flow assume that x belongs to none of those minimal primes p © Ass(M) such that dim(M) = dim(A/p). Again by theorem 2.4 and lemma 2.1 ve have dim(M/x) = dim(t) - 1 whence equality holds. Conversely, assume that equality holds. Let Pys-++) Py € Ass(M) such that dim(M) = aim(A/ p,), Je ly...)k. Then clearly Py t Supp(M/xM) (since, for any M, @im(M) = dim(Supp(M)= Sup (dim A/p) = — Sup (dim p)). p eSupp (M) pchss (0) More quickly, since p, € Supp(M) end Supp(M/xi) = Supp (i) V(x), x ¢ Py. QED. We define a notion extensively used in Algebraic Geometry. Definition 2.5. Let A be a noetherian semi~local ring. A set of elements x,,...,%, € wW is called a system of parameters of the finitely generated A-module M if n= dim(Mt)4a and M/x, M+...+ x, Whas finite length. Note that, by the remark preceding definition 2.5 and theorem 2.4 every A-module admits a system of parameters. We prove Proposition 2.7. Let A, M be as in the above definition. Tet xy5.0.5%, EW. Then dim(M/x) M+...4 x, M) Bn- and equality holds if, and only if, the system x,,...,x, can be imbedded in a system of parameters of M. Proof: We proceed by induction on k. When k = 1 the inequality holds by Proposition 2.6. Furthermore equality holds 1f and only 1f x belongs to none of the primes p in Ass(M) with dim(M) = dim(A/p). Let Xpoeeeo%q_z Ew such that s(M/xM) =n ~ 1, (M/xM)/x, (M/xM) +. x, (MAM) haa finite length. (See defintttan 2.5) Then ine MY 2 5) Xp Xye+++9%,_1 $8 8 system of parameters of M. Conversely, if x con be imbedded in a system of parameters, cay x, %1.--->%, 2. then s(M/xM) 4 - 1 and, by Proposition 2.6, dim(M/xM) = n- 1. QED. ‘The equality My Mee ty Me Oy Mee tog MA (MA MQ 4M) shows, by the induction assumption, the desired inequality. Assume now dim(M/x, M+... %M) =n - K. ‘Then, letting N= M/x, 4 im(W/xy N +..0¢ XN) = (n= 1) - (e- 1) and43 (n-3)- (2) 2 aim(N) ~ (2) 2 dim(M) — 1 ~ (62) - nk whence dim(N) - k+1=n- k or dim(N)=n~-1. By the induction assumption, {x»,...,%,} can be imbedded in a system of parameters of N, say (Xgrerea% ps Xpuyereee%qh (here we must use dim(N) = n- 1). Then clearly {x,, X5,...,%,} 18 a system of parameters of M. Conversely, 12 (x), Xg1-++ 9% qs Xypo+soX_} is @ system of parameters of M, let N= M/x, M. Then N/x, N +...+x, N has finite length, whence s(N) #n- 1. By Proposition 2.6 we have n- 1 = dim(M) - 1 © dim(N) = s(N) $n whence dim(N) = n- 1. Hence {xp,--+s%,s+++s%,) 48 a system of parameters of N, and, by the induction assumption @im(N/xp N+...4 %_ N) = (n- 1) - (k- 1) en-k The proposition is proved. We finish this section with a few remarks about the nature of the function ¥:Spec(A) +N given by 1p) = aim(ap) where A is any noetherian ring. It is obviously not continuous, otherwise it would have to be constant when Spec(A) is connected (e.g. when A is en integral domain), and trivial examples show this is not the case (say A= XIX, YJ). We do nevertheless have some information, namely, by proposition 1.1, atm(Ap ) * aim(A)au and dim(A/p ) $ dim(a). The latter is geometrically interpreted as follows: If x ¢ J, then aim(V(J,)) $ aim(v(dy)). Dimension is a very coarse invariant, i.e. were we to consider the equivalence classes of affine varictice of a given dimension, we would obtain huge classes of highly non isomorphic varieties. $3. DEPTH ‘The next numerical invariant we shall study in the notion of depth, We assume throughout this section that A is a noetherien local ring with maximal ideal 7, and that M is a finitely generated A-module. Definition 3.1. a) an element x ¢ A is called M-reguler if the homomorphism @:M —M given by 9(m) = xm is injective. b) a sequence {x,,...,x,} of elements of A is called M regular if x, is M/x, M+...+ x; 1 M regular, 121 5 n, Remark. Clearly every x $7 being invertible is M-regular for every module M. Hence we shall confine our attention to those M-regular elements which belong to m. With regard to b) we state, without proof, the fact that the sequence (x,.....x,} is M-regular 4f, and only if all sequences {x5(1)+-++0%9(n)} @ € S, are M-reguiar, where S, denotes the group of permutations on n symbols. (Grothendieck, E.G.A., Ch. 0, §15.1, I.H.E.S. no 20) The above statement is false if A is not noetherian.4 Clearly any sequence {x,,.-..x,} with x, $m is Mregular for every M (since M/x; M= 0), hence, keeping in mind the ebove remark, we shall confine our attention to M-regular sequences Gxyppeeoxg) with gy € Me. Definition 3.2. Depth (i) = maximal number of elements in all possible Mregular sequences (of elements of m!). We investigate first some of the properties of the notion of M-regularity, Proposition 3 1) x is Mreguier if, end only if, x¢ Up. pease (M) 2) tf x is Mreguiar, dim(M/xM) - dim() - 1. 3) any Meregular sequence is contained in a system of parameters of M. 4) the sequence {x,,....x,} 1s a maximal M-regular sequence ir, and only ir, one or tne Toliowing two equivalent conditions hold 1) Hom,(k, M/x, M+...+ x,M) $0, where k = A/mm. tt) M/x, M+...+ x, M contains a submodule isomorphic to k. 5) let Gx, Hom,(K, M/x, M+...# x, M) © Ext,"(k, M) = »%,} be an M-regular sequence. Then Ext? 1(k, M/x, M). Proof: 1) Assume x 1s M-reguler, andxe Up « pcass(™) Then x ¢ p € Ass(M), for some p. Now p is the annihilator of some m #0, mM. Therefore the homomorpht.em46 o:M +M, g(m') = 2m! is not injective (p(m) = 0, m #0). Conversely, assume x $ Up , and letm# 0, meM pcsss(M) such that x= 0, Since m+ 0, 0 + AmC M, hence Ass(Am) + (in fact M = 0 <=> Ass(M) = 9). Now Ass(Am) C Ass() trivially, and x ¢ Ann(Am), whence xe Mp .a p cAss(Am) contradiction. 2) This is an imediate consequence of 1) and proposition 2.6. 3) We prove this by induction on k, where Gn,....%%,) ts an Mevegular sequence. Tf k= 1, then x, ia Mregular and, by a 2) above and Proposition 2.7, {x,} can be imbedded in a system of parameters of M. Let k > 1. By induction assumption and Proposition 2.7, dim(M/x, M+...# xy 9 M) = aim(M) - K+ 2, and from W/x, Mte.et my M = QM/xy Meee x / p(y Me M/x, Mt...+ xy) Meregular): + X_ 1 M) and 2) above we get (since x, 15 @im(M/xy Mt..ct xy M) = dim(M) - wnence, again fron Proposition 2.7, (x, +++.%4) ean be Ambedded in a system of parameters of M. 4) We observe that a sequence {x,,...,x,} is M-regular and maximal if, and only if, the sequence {xp,...,x,} is M/x, M-regular and maximal, hence we are reduced by induction to the case r= 0, We observe furthermore that conditions 1) and 41) are obviously equivalent, since a non zero A-honomorphismur of k= A/m 1s injective. Now r= 0 (and maximality), implies that there are no M- regular elements in ff, and by 1) above T= i) FP . There- peass(m) fore m € Ass(M) and, 7% being the annihilator of some non zero x eM, k= A/™m = AxC M and ii) follows. Conversely, if MON A/m= xk, let x € M be @ generator of N. Then 7m is the annibilator of x, whence 77 ¢ Ass(M) and there are no M-regular elements in-yp, i.e. 9 is @ maximal sequence of M-regular elements, Q.e.D. 5) Let N= M/x, M. We have an exact sequence osuSuSnso where 9(m) = x, m. Hence we get 1, 3 = & cee a Ext Ce, Mm) B mxt sh *(e, a) S rl 3 r 3 r Ext, *(c, w) S ext ,"(1, M) S Ext, "(1 M) > - Now, since x, €M%, 9 = 0 (multiplication by x, annihilates all elements of k). On the other hand, by induction Ext,’ 1(k, M) = Hom(k, M/xy M+...4 x4 M) = 0 since {x,,....x, )} 18 not a maximal M-regular sequence. There~ rl = ext? fore Ext,? +(x, n) = uxt,"(k, M). As was pointed out in the proof of 4), {xp,...,x,} 1s a maximal N-regular sequence, whence we can proceed by induction and obtain Ext,’ +(x, m) 2 (ky M/x, M+ xp M) =. = Hom(k, M/x; M+...+ x, M),48 and 5) is proved. Corollary 3.1, Maximal M-regular sequences have the same cardinality, Proo: Obvious from 5). Corollary 3.2. Let = & ¥ =M, Then Depth (M) 1 Depth (M). Proof: The isomorphism MP Jy. MP = M/a Mh. AMO. .8 M/K ME. . tx shows thet any maximal M" regular sequence is a maximal M regular sequence. The corollary follows from Corollary 3.1. We now come to the main theorem concerning the notion of depth, namely: Theorem 3, Let A be @ noetherian local ring, Me finitely generated A-module. Then 1) depth (M) = 0 is equivalent to me Ass(M). 41) if x eM is Mregular then depth (M/xM) = depth (M) - 1. 3 = = 441) depth (0) ant) aim(A/p ) 1p Salant?) dim(M). 1) Tats 1s e restatement of 1), Proposition 3.1. 41) Let {x, pe t+ee%y} be a maximal M/xM- regular sequence. If x is Mregular, then {x, xp, ox,} is a maximal Meregular sequence, whence depth (M) = depth (M/xM) + 1. 411) We prove this by induction on n = depth (M), If 0, then m€ Ass(M), whence, trivially49 O= inf dim(a/p) = sup dim(A/p) = dim(M). pesss(i) p chss(it) In the induction step we shall make use of the following: Iemma 3.1. Let t em be M-regular, pe Ass(M). Then any minimal prime containing p+ At belongs to Ass(M/tM). Proof: By Proposition 4 of B.c.A., IV, §1, there exists a submodule M! CM and an exact sequence 0>M'3M—>M" 30 such that Ass(M') = {PJ}; Ass(M") = Ass(M) - {p}. By 1 of Proposition 3.1, t 1s both M'-regular and M"-regular and the diagren ° ° ° 4 4 4, os M+ HM OM 50 4 4 os M5 HM > MT 30 + L 0 mye w/t weZet 0 4 0 6 6 is obviously commutative and exact, whence Asa(M'/tM') C Ass(M/tM). We have Supp(M"/tM") = Supp(M') A V(t). If gf is a minimal prime containing p + At, then from the above / 18 a minimal prime or supp(M'/tm'), whence Y © Ass(M'/tM) and we are done. We return to the proof of 111) of theorem 3.1. Assume depth (M) = 2, Tet x €m be Mreguler, N= M/xM. By 11) of theorem 3.1, depth (N)=n- 1. Let p be any point inAss(¥), and let g ve a minimal prime containing p+ Ax. Clearly y D p (since x $ p) and by the lemma y € Ass(N). + By the induction assumption we have n- 15 ain(s/y) and clearly dim(A/y ) = aim(A/go) - 1. Hence n 5 aim(a/p), for all pp ¢ Ass(M), 111) follows. Appendix Not only is the function spec (A) > N given by a(p) = depth (Ap) not continuous, but the concept of deptn is @ considerebly more sensitive invariant than dimension. In particular depth (Ap) bears no relation to depth (i), contrary to the behavior of dimension. To see this, let A be any local fing, wiich ly an luveyrel duumin, p © Spec(A), say B = (0). Then Ap is @ field, and has hence depth 0, while depth (A) te arbitrary. On the other nand let A, be any local integral domain, #v, its unique maximal ideal, k,= Aj/p, . Consider St, the A,-module A= A, @ K,, and define on A a ring structure by defining (a, x)+(a', x") = (aa', ax' + atx). One easily checks that A is @ local ring, with m, © k, as unique maximal ideal, and that every non-unit in A is @ zero divisor, whence depth (A) = 0, However, if dim(A,) ® 2, and p, is a non zero, non maximal prime ideal of A, then P = p,@k, 1s 8 prime ° Po’ ideal in A and Ay = A, Now depth (A, 21 since A, P a ae oP. 0 is an integral domain.5 ‘The following result is due to Hartshorne and gives a geometrical significance to the notion of depth. (Hartshorne) Let A be a local ring with depth (a) = 2 ‘Then Spec(A) - {m} is a connected topological space. In particular, the local ring of the unique point of intersection of two sufficiently general planes in four dimensional affine space is a 2-dimensionel ring whose depth (vy Hartshorne's result) is £1. This shows that, in the Anequalities iii) of Theorem 3.1, strict inequality is possible. This justifies the following: Definition 3 Let A be a noetherian local ring, Ma finitely generated A-module. M is said to be a Cohen Macaulay module (C-M module) if depth (M) = dim(M), If A is an arbitrary noetherian ring (not necessarily local), A is said to be @ Cohen-Macaulayring if, for every maximal ideal mof A, whe Local Ting Ag is Conen-biacauiay. We illustrate the notion of C-M modules with a few examples. 1) dim(M) = 0, M# 0. ‘hen, from iii) of theorem 3.1, M is CM. Here the notion of C-M modules is redundant. 2) dim(A) = 1, Aa noethertan local ring. Then, if A is CM, depth (A) = 1, which is equivalent to saying, since dim(A) = 1, thet #1 Ass(A). Hence a non C-M ring of dimension 1 is a local ring in which all non- units are zero divisors. For example if A = k[x, yl], where k is any field and x° y = xy® = 0, and “ty = xA + yA, one easily checks that Aj, is a non ©M ring of dimension 1.3) dim(A) = 2. Here we limit ourselves to showing that every 2-dimensional, integrally closed local integral domain is C-M. To see this, let x em, x40. Since A is an integral domain, x is A-reguler and, since A is integrally closed, none of the prime ideals associated to xA is imbedded (see B.C.A., VII, §1). Then, if p € Ass(A/xA), it follows by the Hauptidealsatz that p +m Therefore m= UP is impossible, and a fp edss (A/xA) yemiyt VU cun be found. Therefore depth (A) = z ee i and hence depth (A) = 2 = dim(A) which proves our assertion. We now investigate some of the consequences of knowing that a ring A, or a module M, are C-M. Proposition 3.2. Let M be a C-M A-module. Then 1) For every pe Ass(M), 4im(A/p) = dim(M) = depth (i) 2) The following three conditions are equivalent: (4) x de Mereguler (44) dim(M/m) = aim(M) - 2 (444) x belongs to no prime of Ass(M) 3) If x is Mregular, M/xM is a CM module Proof: 1) ie a trivial consequence cf the definition of CM modules and of (iii) of Theorem 3.1, 2) (4) implies (41) by (11) of Theorem 3.2, and (1) is equivalent to (111) by 1) of Proposition 3.1. It remains to prove that (ii) implies (1). ‘his follows immediately rrom 1)53 above (all primes in Ass(M) are equidimensional) and proposition 2.6, 3) By (141) of Theorem 3.1 we have depth (M/xM) = depth (M) - 1 = aim(M) - 1 = dim(M/xM) hence M/xM is a C-M module. We state without proof (an easy application of proposition 2,7 and 3.1) the generalization of 2) and 3) above to M-regular sequences. Proposition 3.3. Tat M be a G-M module, Then the follow ing three conditions are equivalent: (4) Gay,++..,} to an M-reguler sequence (14) @im(M/xy M+. t x, M) = aim(M) ~ r (141) (xy,++.5x,) is embeddable in « system of pareneters. surtnermore, ir {x,,.+.,%,} is an M-regular sequence, then M/x, Mt..+ x, Mis a CM module. Proposition 3.4, A module M, for which conditions (1), (11), (441) of the previous proposition are equivalent, and such that M/x, M+...+ x, M is C-M whenever {x,,...,x,} is an M regular sequence, is a -M module, Proof: Let n= dim(M). If n= 0 there is nothing to prove, Assume n2 1, let {x,,...,x,} be a system of parameters of M, Since (111) ==> (1), x, is Mregular and M/x, Mis @ GM module. Now since x, is Mregular x, # Up , whence dim(M/x, M) = dim(M)- 1. Therefore pcdss(M)54 aim(M) = dim(M/xyM) + 1 = depth(M/eM) + 1 = deptn(M) and M is C-M, Q.E.D. Corollery + If Mis a C-M module, every maximal M-regular sequence is e syst of parameters and conversely. Proof; Obvious, Remark. If A is a (not necessarily local) 0-M integral domain, and x € A, x +0, clearly x is A-reguler, whence A/A 1s again CM. Since k[X,,...,%,] 18 a C-M ring (we shall prove this later), it follows from the above remark that, if £(Xys.++0K,)s E(Xys++erX,) are relatively prime irreducible X,], then K[Xy,.+.5%,]/(f) 6) 4s again CM. This throws a better light on example 2) given after elements of ¥1%)5. definition 3.3. We now examine the behavior of the notion of C-M under localization, We have Fropoattion 3.5. Let M be a GM module, pp © Supp(M). Then Y) Mp "Me sp is 8 CW module 2) aim(H) = aim(my,) + aim(M/ px) Proof: We shall obtain proposition 3.5 as @ consequence of the following: Proposition 3.6. Iet M be a C-M module, p € Supp(M), r= dim() - dim(M/p™). Then 1) ‘There exists an M-regular sequence (x,,....x,} with 4 ep and35 2) any such sequence gives dim(M/x, M+...+ x, M) = dim(M/pM) = dim(A/p ). Proof: To prove 1) we proceed by induction on r, When r= 0 the etatement 1s trivial. Tet r 21. Then dim(M/pM) < @im(M), hence p § Ass(M) (since the primes in ‘Acs (M) are equidimenstonal), and therefore Ue. (™) a ) pe gy osu Tet x, © Pp, x, & U + Then x, is M-reguler and the 1 Pe 9 ckekn) 1 module N= M/x, Mis C-M. Furthermore dim(N) = dim(M) - 1 and N/pN = M/pM. We can hence apply the induction assumption to N and find an N-regular sequence {x,,...,x,} with x, «fp. Now trivially (x,,....x,} 1s an M-regular sequence with x, «p, and 1)1s proved. 2) Let {x,,...,%,} be an M-regular sequence with xy ¢ Pp. Let P= M/x, M+..+ x, M (P= Mit r= 0). Now P/pP = M/ pu and from proposition 3.3 we get that ain(r/ pe) = aim(uj - 1 = aim(y) and that P is @ C-M module. Now clearly p € Supp(P), hence p> Po for some Po € Ass(P) dim(P/pP) = dim(P) = dim(A/p') for ali p' ¢€ Ass(P) (since Pis CM). Since clearly p C Ann(P/pP) Furthermore we have dim(P) = dim(P/p P) © aim(a/p ) and dim(&/p) # aim(a/ py) = dim(P). Hence p= Py tc. p< 4s5(P) and 2) follows. We now prove Proposition 3.5. Let x,,...,x, be an M-36 regular sequence inp, where r ~ dim(M) - dim(M/pé). Since localization is a flat operation we have that the images of Ayrissty in pilyp are still an Np -regular sequence. Hence by proposition 1.1 aan(tty,) S aim(m) - aim(Wip Maré deptn(Me,) = aam(tly ) whence 1) and 2) of proposition 3.5 follow. Corollary 3.4, If A is a local C-M ring, A is catenery, and for every local epimorphism A» B, B is catenary. Proof: The quotient of a catenary local ring by a prime ideal being catenary, it is enough to prove A is catenary. Let W be a minimal prime ideal of A, p 1 Y two prime ideals of A such that CPC. Then Ay ata(ay de ama )+ aims, ” 8g) by proposition 3.5. If A'= A/W", and pt, of! are the images of }: 7 iAl, this relation is equivalent to is a CM ring end dim(A',,) = dim(Atg ,) + dim(A', \/pt At.) 1 , 77 Remark. The notion of C-M rings still is insufficient to distinguish the three local rings considered in the introduction, dee. OX, ¥I/(x? - x3 - x*)3 efx, vI/(¥? - x3)s efx, YI(K- ¥), localized at the origin. One easily checks thet all three are CM rings, following the procedure used in the remark after Corollary 3.3. We shall obtain one notion which distinguishes the three local rings in the next section.51 $4, REGULAR RINGS We let A be @ noetherian local ring, tL its maximel ideal, k= Ate, We denote by 3,(/y2) the symetric algebra of the K-vector space Mt/m_2. If rank,(#/,2) = r one trivially has 8, (/py 2) = KL over ky ", are indeterminates reese] © where ty We proceed to define a homomorphism : er, = 8m} 8284 (Mb?) > Crop (A) = Bom 1 as follows Tet %s..05K, be @ K-basis of trt/yq 2, and let x1++.5%, € ML de their representatives. By Nekayana's Lenma (see the renark on page 35) x,)+..4x, forms a set of generators of #t. Hence m* te generated by elenente of the tom 3%... 2€with vy 08... 27; @, +...+4, = 1.0 is defined by 0(%, ~ ... K, 7) = the class of @. @ x, 1}... x, 7 mod mitt, mrivietay © ie a homogeneous homomorphism of degree 0, and an epimorphism. Theorem 4,1. Let A be a noetherian local ring of dimension n, ML its maximal ideal k = A/mm. The following four conditions are equivalent. 8) 0:5, (Ml/y 2) > eryy (A) 18 biJective D) rank, (Mty_ 2) = c) tL ts generated by n elements 4) There existe an A-regular system which generates #t.58 Proof: b) ==> c) follows from the remark above that every K-basis of Athy? lifts back ( in m) to a set of generators of ‘m (by Nakayama's Lemma). Conversely, any set of generators of 4W gives rise (mod 1°) to a set of generators of t/q 2 over Ky whence rank, (ttt/yy2) 5 rank, (fit/,2) 2 n, whence c) ==> b). We have proved >) <=> ce). a) ==> 4). Let Z. But, by proposition 2.5, ip © Mya? de | Dasis Of M/ yy over k, We use the symbol % for Z 1... Z, 7, 3, and al Sap ti. ay. Tet z © be representatives of a ir r eee «0%. We already know thet 2),...,2, generate % (Nokayona's Lemma), and shall show that they form an A-regular sequence. We begin by asserting that, obviously, P= 5 lea a 2 (mod mJ*t) lea where &, € k = A/ttt, cy € A, their representatives. Hence, since 9 1s injective, the relation = c, 2 emi, clea lol=3 implies 6( = T, 2") = 0, whence c, ¢ fh. jal=5 Assume now that z,,...,2, do not form an A-regular sequence. Then, for some j, 1 J £4, there exists anx¢ A, xb Az) +..4A 2, and x2, € Az) tt A 2s .1- That, is we ‘e1 have an equation of the form wey Vy By teeth Ty Bye Since 6 1s surjective, we have, for some t,59 EB cate! (mod. 7**?) lal=t where et least one c, for ana witha, =a, =... a, =0 is He such that cy #7. However, in the expression of Vy teeth Vp ey a8 Edge” (mod. mmi*®), all the Jal St+2. coefficients 4, such that d, + 77 correspond to multiindices a for which Gy, dps-+.,05 1 are not all 0, We thus reach a contradiction, 4) 9c). let 24,....2, be an A-reguler sequence which forms a set of generators of mm. Then, by proposition 2.5, + & rank (7m /m 2) ® 0 and by the definition of depth (A) and theorem 3.1 n® depth (A) # r. Hence r= rank,(M/p,2) = n, and c) follows: c) da). We proceed by contradition, 1.e, we assume ker © +0, For brevity's sake we write S = 8 (yy 2) G= er, (A). We have the exact sequence oJ 284040