# Full text of "Trivial Extensions of Local Rings and a Conjecture of Costa"

## See other formats

Lecture Notes in Pure and Applied in Mathematics - Dekker 231 (2002) 301-311 Trivial Extensions of Local Rings and a Conjecture of Costa S. Kabbaji'* and N. Mahdou^ ^Department of Mathematics, KFUPM, P.O. Box 5046, Dhahran 31261, Saudi Arabia ^Department of Mathematics, EST Fez-Saiss, B.P. 2202, Univ. of Fez, Fez, Morocco Abstract. This paper partly settles a conjecture of Costa on (n, (i)-rings, i.e., rings in which n-presented modules have projective dimension at most d. For this purpose, a theorem studies the transfer of the (n, (i)-property to trivial extensions of local rings by their residue fields. It concludes with a brief discussion -backed by original examples- of the scopes and limits of our results. *This project has been funded by King Fahd University of Petroleum & minerals under Project # MS/TRIVIAL/224 0. Introduction All rings considered in this paper are commutative with identity elements and all modules are unital. For a nonnegative integer n, an i?-module E is n-presented if there is an exact sequence — > -Fn-i ■■■ ^ Fq ^ E ^ in which each Fi is a finitely generated free i?-module (In [1], such E is said to have an n-presentation). In particular, "0-presented" means finitely generated and "1-presented" means finitely presented. Also, pdjtE will denote the projective dimension of E as an i?-module. In 1994, Costa [2] introduced a doubly filtered set of classes of rings throw- ing a brighter light on the structures of non-Noetherian rings. Namely, for nonnegative integers n and d, a ring R is an [n, d)-ring if every n-presented i?-module has projective dimension at most d. The Noetherianness deflates the (n, d)-property to the notion of regular ring. However, outside Noethe- rian settings, the richness of this classification resides in its ability to unify classic concepts such as von Neumann regular, hereditary/Dedekind, and semi- hereditary/Priifer rings. Costa was motivated by the sake of a deeper under- standing of what makes a Priifer domain Priifer. In this context, he asked "what happens if we assume only that every finitely presented (instead of generated) sub- module of a finitely generated free module is projective?" It turned out that a non-Priifer domain having this property exists, i.e., (In the (n, d) -jargon) a (2, l)-domain which is not a (1, l)-domain. This gave rise to the theory of (n, d) -rings. Throughout, we assume familiarity with n- presentation, coherence, and basics of the (n, (i)-theory as in [1, 2, 3, 6, 7, 8, 10]. Costa's paper [2] concludes with a number of open problems and conjec- tures, including the existence of (n, (i)-rings, specifically whether: '^Thcrc are examples of (n, d)-rings which are neither (n, d — l)-rings nor (n — 1, d)-rings, for all nonnegative integers n and d". Some limitations are immediate; for instance, there are no (n, 0)-domains which are not fields. Also, for d = or n = the conjecture reduces to "(n, 0)-ring not {n— l,0)-ring" or "(0,d)-ring not {0,d— l)-ring", respectively. Let's summarize the current situation. So far, solely the cases n < 2 and d arbitrary were gradually solved in [2], [3], and [14]. These partial results were 2 obtained using various puUbacks. For obvious reasons, these were no longer useful for the specific case d = 0. Therefore, in [14], the author appealed to trivial extensions of fields k by infinite-dimensional A;-vector spaces, and hence constructed a (2, 0)-ring (also called 2- von Neumann regular ring) which is not a (l,0)-ring (i.e., not von Neumann regular). This encouraged further work for other trivial extension contexts. Let ^ be a ring and E an ^-module. The trivial ring extension of ^ by £^ is the ring R = A ex E whose underlying group is ^ x £^ with multiplication given by (a, e)(a',e') = {aa',ae' + a'e). An ideal J of R has the form J = I (X E' , where / is an ideal of A and E' is an A-submodule of E such that IE C E' . Considerable work, part of it summarized in Glaz's book [10] and Huckaba's book [11], has been concerned with trivial ring extensions. These have proven to be useful in solving many open problems and conjectures for various contexts in (commutative and non-commutative) ring theory. See for instance [4, 5, 9, 12, 13, 15, 16, 17]. Costa's conjecture is still clusively outstanding. A complete solution (i.e., for all nonnegative integers n and d) would very likely appeal to new techniques and constructions. Our aim in this paper is much more modest. We shall re- solve the case "n = 3 and d arbitrary" . For this purpose. Section 1 investigates the transfer of the (n, (i)-property to trivial extensions of local (not necessarily Noetherian) rings by their residue fields. A surprising result establishes such a transfer and hence enables us to construct a class of (3, (i)-rings which are neither (3, d — l)-rings nor (2, ci)-rings, for d arbitrary. Section 2 is merely an attempt to show that Theorem 1.1 and hence Example 1.4 are the best results one can get out of trivial extensions of local rings by their residue fields. 1 Result and Example This section develops a result on the transfer of the (n, (i)-property for a par- ticular context of trivial ring extensions, namely, those issued from local (not necessarily Noetherian) rings by their residue fields. This will enable us to con- struct a class of (3, d)-rings which are neither (3, d — l)-rings nor (2, (i)-rings, for d arbitrary. 3 The next theorem not only serves as a prelude to the construction of ex- amples, but also contributes to the study of the homological algebra of trivial ring extensions. Theorem 1.1 Let {A, M) he a local ring and let R = Ao: AjM he the trivial ring extension of A hy A/M. Then 1) R is a (3,0) -ring provided M is not finitely generated. 2) R is not a (2, d)-ring, for each integer d >0, provided M contains a regular element. The proof of this theorem requires the next preliminary. Lemma 1.2 Let A he a ring, I a proper ideal of A, and R the trivial ring extension of A hy A/I. Then pdR{I oc A/I) and hence pdR{0 oc A/I) are infinite. Proof. Consider the exact sequence of i?-modules 0^ I <xA/I ^ R^ R/{I oc A/I) Wc claim that R/{I oc A/ 1) is not projective. Deny. Then the sequence splits. Hence, / oc A/I is generated by an idempotent element (a, e) = (a,e)(a, e) = (a^,0). So / oc A/ 1 = R{a,0) = Aa oc 0, the desired contradiction (since A/I ^ 0). It follows from the above sequence that pdR{R/{I cxA/I)) = l+pdR{I cxA/I). (1) Let (xj)igA be a set of generators of / and let R^^^ be a free i?-module. Consider the exact sequence of i?-modules ^ Ker{u) R^^^ ®R^I (xA/I ^0 where u{{ai, ej)igA, (ao, eo)) = ^{ai, ei){xi, 0) + (oq, eo)(0, 1) = (^ aiXi, ao) since Xi & I for each ieA. Hence, Ker{u) = {U(x {A/I)^^^) © (/ oc A/I) 4 where U = {(ai)ieA G ^'^'^V ^ajXi = 0}. Therefore, we have the isomor- ieA phism of i^-modules I A/I ^ {R(^^/{U cx (A/I)^^^)) © {R/{I oc A/I)). It follows that pdn{R/{I ^ A/I)) < pdnil <x A/I). (2) Clearly, (1) and (2) force pdR{I oc A/ 1) to be infinite. Now the exact sequence of i?-modules O^IoiA/I^R^O(xA/I^O, where v{a, e) = (a, e)(0, 1) = (0, a), easily yields pdR{0 oc A/I) = oo, complet- ing the proof of Lemma 1.2.2. □ Proof of Theorem 1.2.1. 1) Suppose M is not finitely generated. Let -ffo(/ 0) be a 3-presented i2-module and let (2;i)i=i,...,n be a minimal set of generators of Hq (for some positive integer n) . Consider the exact sequence of i?-modules ^ i/i := Ker{uo) ^ R^ ^ Hq ^ n where no((rj)j=i^..._n) = ""^^riZi. Throughout this proof we identify i?" with •i=i oc {A/My^. Our aim is to prove that Hi = 0. Deny. By the above exact sequence, Hi is a 2-presented i?-module. Let {xi,yi)i=i^,„^m be a minimal set of generators of Hi (for some positive integer m). The minimality of (2;j)j=i,...,n implies that C M" oc {A/M)'^, whence Xi G M" (and G (A/M)") for i = 1, . . . , m. Consider the exact sequence of i?-modules ^ i72 := ifer(«i) ^ ^ Hi ^ Q rn in where ui((ai,ei)i) = '^iai,ei){xi,yi) = '^{aiXi,aiyi), since Xj e M" for each i=l 1=1 m i. Then, H2 = U (x (A/M)"^, where f/ = {(aj)i=i,...,„i G A'^/^aiXi = and i=l ^^Oiyi = 0}. By the above exact sequence, H2 is a finitely presented (hence i=l generated) i?-module, so that (via [11, Theorem 25.1]) ?7 is a finitely generated 5 ^-module. Further, the minimahty of {xi^yi)i=i^,„^m yields U C M"*. Let {ti)i=i,...,p be a set of generators of U and let {fi)i=p+i,...,p+m be a basis of the (A/M)-vector space (A/M)"^. Consider the exact sequence of i?-modules O^Hs:= Ker{u2) ^ ^ H2 ^ where p p+m p p+m U2{{ai,ei)i) = '^{ai,ei){ti,0) + ^ {ai,ei){0, fi) = (^aiU, ^ a^/j), i=l i=p+l i=l i=p+l since ti € M™ for each i = 1, . . . ,p and is a basis of the (yl/M)-vector space (A/M)"^. It follows that H3 ^ {V oc {A/M)p) (M™ oc (^/M)'"), where V = {(aj)j=i_...^p G ^^aji^ = 0}. By the above sequence, is a 1=1 finitely generated i?-module. Hence M oc A/M is a finitely generated ideal of R, so M is a finitely generated ideal of A by [11, Theorem 25.1], the desired contradiction. Consequently, Hi = 0, forcing Hq to be a free i?-module. Therefore, every 3-presented i?-module is projective (i.e., i? is a (3,0)-ring). 2) Assume that M contains a regular element m. We must show that R is not a (2,d)-ring, for each integer d > 0. Let J = R{m,0) and consider the exact sequence of i?-modules Ker{v) R ^ J ^ where v{a,e) = (a, e)(m, 0) = (am, 0). Clearly, Ker{v) = oc (A/M) = R{0, 1), since m is a regular element. Therefore, Ker{v) is a finitely generated ideal of R and hence J is a finitely presented ideal of R. On the other hand, pdii{Ker{y)) = pdniQ a A/M) = 00 by Lemma 1.2.2, so pdji{J) = 00. Finally, the exact sequence O^J^R^R/J^O yields a 2-presented R-module, namely R/J, with infinite projective dimension (i.e., R is not a (2,d)-ring, for each d > 0), completing the proof. □ 6 We are now able to construct a class of (3, c?)-rings which are neither (3, d—1)- rings nor (2, (i)-rings, for d arbitrary. In order to do this, we first recall from [14] an interesting result establishing the transfer of the (n, d) -property to finite direct sums. Theorem 1.3 ([14, Theorem 2.4]) A finite direct sum 0i<j<„ Ai is an (n, d)- ring if and only if so is each Ai . □ Example 1.4 Let d be a nonnegative integer and B a Noetherian ring of global dimension d. Let {Ao,M) be a nondiscrete valuation domain and A = Ao oc (Aq/M) the trivial ring extension of Aq by Aq/M. Let R = A x B be the direct product of A and B. Then i? is a (3,d)-ring which is neither a (3, d — l)-ring nor a (2, (i)-ring, for d arbitrary (The case d = reduces to "(3,0)-ring not (2, 0)-ring"). Proof. By Theorem 1.2.1, A is a (3, 0)-ring (also called 3- Von Neumann regular ring) which is not a (2, d')-ring for each nonnegative integer d'. Moreover, i? is a (3, d)-ring by [14, Theorem 2.4] since both A and B are (3, (i)-rings (by gnomonic theorems of Costa [2]). Further, R is not a (2,(i)-ring by [14, Theorem 2.4] (since A is not a (2, d)-ring). Finally, we claim that R is not a (3,d — l)-ring. Deny. Then 5 is a (3, ci — l)-ring by [14, Theorem 2.4]. Hence, by [2, Theorem 2.4] 5 is a (0,d — l)-ring since B is Noetherian (i.e., 0-coherent). So that gldim{B) < d — 1, the desired contradiction. □ 2 Discussion This section consists of a brief discussion of the scopes and limits of our find- ings. This merely is an attempt to show that Theorem 1.2.1 and hence Exam- ple 1.4 are the best results one can get out of trivial extensions of local rings by their residue fields. Remark 2.1 In Theorem 1.2.1, the (n, d) -property holds for a trivial ring extension of a local ring {A, M) by its residue field sans any (n, d)-hypothesis on the basic ring A. This is the first surprise. The second one resides in the narrow scope revealed by this (strong) result, namely n = 3 and d = 0. 7 Thus, the two assertions of Theorem 1.2.1, put together with Costa's gnomonic theorems, restrict the scope of a possible example to n = 3 and d arbitrary. Furthermore, since in Theorem 1.2.1 the upshot is controlled solely by re- strictions on M, the next two examples clearly illustrate its failure in case one denies these restrictions, namely, "M is not finitely generated^ and "M contains a regular eiemenfc" , respectively. Example 2.2 Let Khea field and let A = K[[X]] = K+M, where M = XA. We claim that the trivial ring extension i? of A by A/M{= K) is not an (n, d)- ring, for any integers n, d > 0. Proof. Let's first show that R is Noetherian. Let J = / oc £^ be a proper ideal of R, where / is a proper ideal of A and -E is a submodule of the simple A-module A/M (i.e., E = ot E = A/M). Since ^ is a Noetherian valuation ring, / = Aa for some a G M. Let f & A such that (a, /) G J. Without loss of generality, suppose J ^ R{a,f). Let {c,g) G J \ R{a,f), where c,geA, and let c = Aa, for some \ ^ A. Then (0,^ — A/) = {c,g) — (a,/)(A,0) G J \ R{a,f), so that we may assume c = and g ^ 0, i.e., g is invertible in A. It follows that (0,1) = (0,g){g-^,0) G J (hence E = A/M) and (a,0) = (a,/) - (0,5)(5("V)0) G J. Consequently, J = (a, 0)^2+ (0, whence J is finitely generated, as desired. Now, by Lemma 1.2.2, pdR{0 oc A/M) = pdRR{0, 1) = oo, hence gldim{R) = oo. Then an application of [2, Theorem 1.3(ix)] completes the proof. □ Example 2.3 Let X be a field and £^ be a i^-vector space with infinite rank. Let A = K (X E he the trivial ring extension of K by E. The ring A is a local (2, 0)-ring by [14, Theorem 3.4]. Clearly, its maximal ideal M = oc £^ is not finitely generated and consists entirely of zero-divisors since (0, e)M = 0, for each e G E. Let R = A oc (A/M) be the trivial ring extension of A by A/M{= K). Then i? is a (2, 0)-ring (and hence Theorem 1.2.1(2) fails because of the gnomonic property). Proof. Let H he a 2-presented i?-module and let ( minimal set of generators of H. Our aim is to show that H is a projective i?-module. 8 Consider the exact sequence of i?-modules ^ Ker{u) ^ ^ H ^ Q n where ti((rj)j=i_...^„) = '^^ViXi. So, Ker{u) is a finitely presented i?-module i=l with Ker{u) = U (x E', where [/ is a submodule of A'^ and E' is a i^- vector subspace of K^. We claim that Ker{u) = 0. Deny. The minimality of (xi, . . . ,Xn) yields Ker(u) = ?7 oc ^' C (M oc A/M)i?" = (M oc yl/M)" since i? is local with maximal ideal M oc AjM. Let (j/j, /i)i=i,...,p be a minimal set of generators of Ker{u), where yi G M" and /j G K". Consider the exact sequence of i?-modules ^ Ker{v) ^ BP ^ Ker{u){= U (xE')^O p p p where f ((a,, ei)i=i,...,p) = ^(cj, ei)(yi, /j) = (^0,2/^,^0,^/,). Here too the 1=1 i=l i=l minimality of {ui, fi)i=i,...,p yields Ker{v) C (M a ^/M)^; whence, Ker{v) = p V oc (A/M)f, where V = {(ai)i=i,...,p e ^f/^a^yi = 0}(C M^). By the i=l above exact sequence, Ker{v) is a finitely generated i?-module, so that F is a finitely generated 74-module [11, Theorem 25.1]. Now, by the exact sequence p where ■i«((oi)i=i,...,p) = ^^ajj/j, ?7 is a finitely presented ^-module (since U i=l is generated by (yi)i=i,...,p). Further, U is an A-submodule of A'^ and A is a (2, 0)-ring, then U is projective. In addition, A is local, it follows that U is a finitely generated free ^-module. On the other hand, U C M" = (0 oc , so (0, e)U = for each e e E, the desired contradiction (since U has a basis). □ 9 References [1] N. Bourbaki, Algebre commutative, chapitres 1-4, Masson, Paris, 1985. [2] D.L. Costa, Parameterizing families of non-Noetherian rings. Comm. Al- gebra 22 (1994) 3997-4011. [3] D.L. Costa, S. Kabbaj, Classes of D + M rings defined by homological conditions. Comm. Algebra 24 (1996) 891-906. [4] R. Damiano, J. Shapiro, Commutative torsion stable rings, J. Pure Appl. Algebra 32 (1984) 21-32. [5] F. Dischinger, W. Miiller, Left PF is not right PF, Comm. Algebra 14 (1986) 1223-1227. [6] D.E. Dobbs, On the global dimensions oi D + M, Canad. Math. Bull. 18 (1975) 657-660. [7] D.E. Dobbs, S. Kabbaj, N. Mahdou, M. Sobrani, When is D + M n- coherent and an (n, d)-domain?. Lecture Notes in Pure and Appl. Math., Dekker, 205 (1999), 257-270. [8] D.E. Dobbs, I.J. Papick, When is D + M coherent?, Proc. Amer. Math. Soc. 56 (1976) 51-54. [9] R. Fossum, Commutative extensions by canonical modules are Gorenstein rings, Proc. Amer. Math. Soc. 40 (1973) 395-400. [10] S. Glaz, Commutative coherent rings. Lecture Notes in Mathematics, Vol. 1371, Springer- Verlag, Berlin, 1989. [11] J. A. Huckaba, Commutative rings with zero divisors. Marcel Dekker, New York, 1988. [12] G. Levin, Modules and Golod homomorphisms, J. Pure Appl. Algebra 38 (1985) 299-304. [13] C. Lofwall, On the homotopy Lie algebra of a local ring, J. Pure Appl. Algebra 38 (1985) 305-312. 10 [14] N. Mahdou, On Costa's conjecture, Comm. Algebra 29 (7) (2001) 2775- 2785. [15] I. Palmer, J.-E. Roos, Explicit formulae for the global homological di- mensions of trivial extensions of rings, J. Algebra 27 (1973) 380-413. [16] D. Popescu, General Neron desingularization, Nagoya Math. J. 100 (1985) 97-126. [17] J.-E. Roos, Finiteness conditions in commutative algebra and solution to a problem of Vasconcelos, Commutative algebra: Durham 1981 (Durham, 1981), pp. 179-203, London Math. Soc. Lecture Note Ser., 72, Cambridge Univ. Press, Cambridge-New York, 1982. 11