betaFeedback

WorksMathematics

Embedding into quotient

Abstract

We study Bass's finiteness problem for embeddings into proper quotients. Given a commutative ring R, a reduced finitely generated R-algebra A, and an ideal J◃AJ\triangleleft A, we ask whether an R-algebra monomorphism A↪RA/JA\hookrightarrow_R A/J forces J=0. We give affirmative answers by presenting two arguments. Also, we discuss not only in the reduced case, but also in several non-reduced cases: when R is artinian and A is finite over R; when R is zero-dimensional and A is finite projective over R; and in a series of Cohen–Macaulay settings. In the module-theoretic direction, for a noetherian reduced ring, and a finitely generated, and torsion-free A-module M, we show that M↪RM/JMM\hookrightarrow_R M/JM implies that J is contained in a minimal prime of A; if M is faithful, then J=0. Also, dual versions investigated. Examples illustrate the necessity of hypotheses.

Introduction

This paper starts with the following implicit question of Bass [1].

Question 1.1. If AA is a finitely generated reduced commutative RR-algebra and A↪RA/JA \hookrightarrow_R A/J is an RR-algebra monomorphism, must J=0J = 0?

Bass proved:

Theorem 1.2. (Bass, see [1 Main result]). Let RR be a commutative ring, let AA be a reduced finitely presented RR-algebra, and let JJ be an ideal of AA. If there is an injective RR-algebra monomorphism A→A/JA \to A/J, then J=0J = 0.

Also, he showed that the finitely generated hypothesis is necessary. Indeed, for A=R[X1,X2,…]A = R[X_1, X_2, \ldots], there is an RR-algebra isomorphism A≅RA/(X1)A \cong_R A/(X_1), but AA is not finitely generated. He also showed the importance of the reduced assumption (see [1 Example 1.3]).

The purpose of the present paper is to study Question 1.1 and several variants, both in the algebra and module settings. We begin with the following result (later we improve it by replacing domain with reduced):

Theorem 1.3. (See Theorem 2.5). Let RR be a domain, let AA be a reduced finitely generated RR-algebra which is torsion-free as an RR-module, and let JJ be an ideal of AA. If there exists an RR-algebra monomorphism φ:A↪RA/J\varphi: A \hookrightarrow_R A/J, then J=0J = 0.

2020 Mathematics Subject Classification. 13A15, 13E15.

Key words and phrases. Cohen-Macaulay; embeddings and quotients; finiteness properties; reduced rings; nilpotent. In Section 2 we also treat non-reduced algebras and prove an affirmative answer in several cases. We then relate Question 1.1 to (co-)Hopficity. Recall that an RR-algebra AA is Hopfian if every surjective RR-endomorphism A→AA \to A is injective. If A≅RA/JA \cong_R A/J for some nonzero ideal JJ, then AA admits a surjective non-injective RR-endomorphism; moreover, A̸≅AA/JA \not\cong_A A/J. We also prove several further positive results. The zero-dimensional case is considered in Lemma 2.8 and Theorem 2.7. In the Cohen–Macaulay direction, Proposition 2.12 shows that, under the natural maximal Cohen–Macaulay hypotheses, an RR-algebra monomorphism A↪RA/JA \hookrightarrow_R A/J again forces J=0J = 0. The proof uses induction on dim⁡R\dim R, reduction modulo a maximal regular sequence, and Krull’s intersection theorem. In the opposite direction, Proposition 2.14 presents a vanishing result for coker⁡(φ:A↪RA/J)\operatorname{coker}(\varphi: A \hookrightarrow_R A/J) under a constructive hypothesis starting with a non-Cohen–Macaulay ring RR; this again forces J=0J = 0.

In Section 3 we show:

Theorem 1.4. (See Theorem 3.1). Question 1.1 is true.

Section 4 develops a module-theoretic analogue. For a noetherian reduced ring RR, an ideal J⊲RJ \lhd R, and a finitely generated torsion-free RR-module MM, Theorem 4.1 proves that

M↪RM/JM⟹J⊆p for some p∈Min⁡(R).M \hookrightarrow_R M/JM \quad\Longrightarrow\quad J \subseteq\mathfrak{p}\text{ for some }\mathfrak{p} \in\operatorname{Min}(R).

If MM is additionally faithful over RR, then J=0J = 0. We record a connection to local cohomology in Proposition 4.3. We also discuss the corresponding result for divisible modules; see Proposition 4.7. A non-noetherian example shows that the noetherian hypothesis cannot be dropped.

Theorem 1.5. (See Theorem 4.9). Let R⊆AR \subseteq A be an extension of noetherian reduced rings such that AA is torsion-free as an RR-module. Let J⊲AJ \lhd A, and let MM be a finitely generated RR-module which is torsion-free as an AA-module. If M↪RM/JMM \hookrightarrow_R M/JM, then J⊆pJ \subseteq\mathfrak{p} for some p∈Min⁡(A)\mathfrak{p} \in\operatorname{Min}(A).

We also record a rank version and give an example showing that the finiteness of MM as an RR-module is essential.

Finally, Section 5 is devoted to the dual of Question 1.1: when does the existence of an RR-linear epimorphism A↠JA \twoheadrightarrow J or J↠AJ \twoheadrightarrow A force J≅RAJ \cong_R A? The answer depends heavily on the hypotheses: over a domain AA finite over RR, any such epimorphism is automatically an isomorphism (see Theorem 5.3); in the artinian case, an epimorphism J↠AJ \twoheadrightarrow A is necessarily an isomorphism (see Proposition 5.6), while an epimorphism A↠JA \twoheadrightarrow J need not be. In the Cohen–Macaulay setting, we show that an epimorphism J↠AJ \twoheadrightarrow A is impossible under natural conditions (see Proposition 5.7). As an application, we show:

Corollary 1.6. (See Corollary 5.10). Suppose R=AR = A is a local ring of dimension one and P∈Ass⁡(A)P \in\operatorname{Ass}(A). Then there is no π:P↠RA\pi: P \twoheadrightarrow_R A.

We present several examples to show that the assumptions are needed and that the results are sharp.

All rings in the sequel are commutative. We consider both noetherian and non-noetherian rings. The term quasilocal denotes a ring with a unique maximal ideal. By local ring we mean a noetherian quasilocal ring. We follow the notation and conventions of [2, 4] for homological and commutative algebra.

Embeddings into proper quotients

Recall from Question 1.1 we are inserted to RR-algebra embedding A↪A/JA \hookrightarrow A/J. The phenomenon is closely related to the notion of Hopficity. Recall that an RR-algebra AA is called Hopfian if every surjective RR-endomorphism A→AA \to A is injective.

Proposition 2.1. Let AA be an RR-algebra and let 0≠J⊂A0 \ne J \subset A be an ideal. If A≅RA/JA \cong_R A/J, then AA admits a surjective RR-algebra endomorphism which is not injective. Moreover, A̸≅AA/JA \not\cong_A A/J as AA-modules.

Proof. Let θ:A/J→≅RA\theta: A/J \xrightarrow{\cong_R} A be an RR-algebra isomorphism, and let q:A→A/Jq : A \to A/J be the canonical quotient map. Then θ∘q:A→A\theta\circ q : A \to A is a surjective RR-algebra endomorphism with ker⁡(θ∘q)=J≠0\ker(\theta\circ q) = J \ne0. Thus it is not injective. Finally, suppose that A≅AA/JA \cong_A A/J. Since JJ annihilates A/JA/J, AA-linearity would imply that JJ annihilates AA. Hence J=J⋅1A=0J = J \cdot1_A = 0, a contradiction. □

We record the following easy observation.

Observation 2.2. Suppose there is A′A' between R⊆AR \subseteq A so that for some J′⊆A′J' \subseteq A' we have J′A=JJ'A = J and that A′⊆AA' \subseteq A is faithfully flat. If Question 1.1 is valid for (R,J′⊆A′)(R, J' \subseteq A') then it is valid over (R,J⊆A)(R, J \subseteq A).

Corollary 2.3. Let A′A' be a finitely related reduced RR-algebra and J′J' be an ideal. Set A=A′[Z]/(Zn)A = A'[Z]/(Z^n) and J=J′AJ = J'A. If A↪RA/JA \hookrightarrow_R A/J is an RR-algebra monomorphism, then J=0J = 0.

Proof. Recall that AA is free over A′A', because whose base is given by {1,Z,…,Zn−1}\{1, Z, \ldots, Z^{n-1}\}. It is easy to see that A↪A/J=(A′→A′/J′)⊗A′AA \hookrightarrow A/J = (A' \to A'/J') \otimes_{A'} A. By faithfully flatness, A′→A′/J′A' \to A'/J' is injective. By [1 Main result], J′=0J' = 0. Then by Observation 2.2 we get the claim. □

Before presenting the next result, we present a useful example.

Example 2.4. (Warm-up) Let RR be a domain and set A=R[X,Y,Z]/(XY,Z2)A = R[X,Y,Z]/(XY,Z^2). Write xx and yy for the images of XX and YY in AA. Let J=(x−y)J = (x-y). Then there is no injective RR-algebra homomorphism φ:A↪RA/J\varphi: A \hookrightarrow_R A/J.

Proof. Without loss of generality, we may replace AA with R[X,Y]/(XY)R[X,Y]/(XY) (see Observation 2.2). We are going to prove the claim without any use of Theorem 1.2. Suppose not. So, there is such a morphism φ\varphi. First observe that, as an RR-module, A=R⊕xR[x]⊕yR[y]A = R \oplus xR[x] \oplus yR[y] with (x)∩(y)=0(x) \cap(y) = 0. Then Ann⁡A(x)=(y)\operatorname{Ann}_A(x) = (y), Ann⁡A(y)=(x)\operatorname{Ann}_A(y) = (x). Put u=φ(x)u = \varphi(x), v=φ(y)v = \varphi(y). Since xy=0xy = 0, we have uv=φ(xy)=0uv = \varphi(xy) = 0. Since φ\varphi is injective and x,y≠0x,y \ne0, we have u≠0≠vu \ne0 \ne v. Then A/(x−y)≅R[x]/(x2)A/(x-y) \cong R[x]/(x^2). Every element of R[x]/(x2)R[x]/(x^2) has a unique expression a+bxa+bx with a,b∈Ra,b \in R. Suppose (a+bx)(c+dx)=0(a+bx)(c+dx)=0. Since x2=0x^2=0, we obtain ac+(ad+bc)x=0ac+(ad+bc)x=0. Hence ac=0ac=0 and ad+bc=0ad+bc=0. Since RR is a domain, ac=0ac=0 implies a=0a=0 or c=0c=0. If a=0a=0, then bc=0bc=0. Thus either b=0b=0, in which case a+bx=0a+bx=0, or c=0c=0. Similarly, if c=0c=0, then ad=0ad=0, so either a=0a=0 or d=0d=0. Consequently, if both a+bxa+bx and c+dxc+dx are nonzero, then necessarily a=c=0a=c=0. Thus every pair of nonzero elements with zero product has the form bx,dxbx,dx, with b,d≠0b,d \ne0. In particular, (bx)2=(dx)2=0(bx)^2=(dx)^2=0.

Now let u=φ(x)u=\varphi(x), v=φ(y)v=\varphi(y). We have u,v≠0u,v \ne0 and uv=0uv=0, so the preceding observation gives u2=v2=0u^2=v^2=0. But u2=φ(x2)u^2=\varphi(x^2) and v2=φ(y2)v^2=\varphi(y^2). Since x2≠0≠y2x^2 \ne0 \ne y^2 in AA, this contradicts the injectivity of φ\varphi. Hence A̸↪RA/(x−y)A \not\hookrightarrow_R A/(x-y). □

Theorem 2.5. Let RR be a domain, let AA be a reduced finitely generated RR-algebra which is torsion-free as an RR-module, and let JJ be an ideal of AA. If there exists an RR-algebra monomorphism φ:A↪RA/J\varphi: A \hookrightarrow_R A/J, then J=0J=0.

Proof. Let K=Frac⁡(R)K = \operatorname{Frac}(R). Since AA is torsion-free over the domain RR, the canonical map

A⟶K⊗RA=:AKA \longrightarrow K \otimes_R A =: A_K

is injective. Hence, J=0J=0 if and only if JK:=K⊗RJ=0J_K := K \otimes_R J = 0. Since KK is flat over RR, tensoring φ\varphi with KK gives a KK-algebra monomorphism

φK:AK⟶AK/JK.\varphi_K : A_K \longrightarrow A_K/J_K.

Moreover, AKA_K is a finitely generated reduced KK-algebra. Indeed, AKA_K is obtained from AA by base change to the field KK, and reducedness is preserved under field extensions. Since KK is a field, every finitely generated KK-algebra is finitely presented. Thus AKA_K is a finitely presented reduced KK-algebra. Bass’ theorem (see Theorem 1.2) therefore applies to φK:AK→AK/JK\varphi_K : A_K \to A_K/J_K, and yields JK=0J_K=0. Finally, JJ is an RR-submodule of the torsion-free RR-module AA, and hence JJ is torsion-free. Therefore K⊗RJ=0K \otimes_R J=0 implies J=0J=0. □

Corollary 2.6. Let RR be a ring and let AA be a finitely generated RR-algebra which is a domain. If J⊆AJ \subseteq A is an ideal and there exists an RR-algebra monomorphism A→RA/JA \to_R A/J, then J=0J=0.

Proof. Let ρ:R→A\rho:R\to A be the structure homomorphism and put p=ker⁡(ρ)\mathfrak{p}=\ker(\rho). Since AA is a domain, p\mathfrak{p} is prime. Set R0=R/pR_0=R/\mathfrak{p}. Then R0R_0 is a domain and the structure map factors as R→R0↪AR\to R_0\hookrightarrow A. Since the given map is an RR-algebra homomorphism, it is also an R0R_0-algebra homomorphism. Moreover, AA is a finitely generated R0R_0-algebra and is a domain. Hence Theorem 2.5 applies and gives J=0J=0. □

Artinian rings are far from be reduced. This suggests the next result:

Lemma 2.7. Let RR be an artinian ring and let AA be a finite RR-algebra. If JJ is an ideal of AA and there exists an RR-algebra monomorphism φ:A↪RA/J\varphi:A\hookrightarrow_R A/J, then J=0J=0. In particular, the conclusion holds without assuming that AA is reduced.

Proof. Since RR is artinian and AA is finite as an RR-module, AA is an Artinian RR-module. In particular, AA has finite length as an RR-module. The quotient A/JA/J also has finite length. Since φ\varphi is an RR-algebra monomorphism, it is in particular an injective homomorphism of RR-modules. Hence ℓR(A)≤ℓR(A/J)\ell_R(A)\leq\ell_R(A/J), with the convenience that ℓ(−)\ell(-) means the length function. On the other hand, the canonical exact sequence

0⟶J⟶A⟶A/J⟶00\longrightarrow J\longrightarrow A\longrightarrow A/J\longrightarrow0

gives ℓR(A)=ℓR(J)+ℓR(A/J)\ell_R(A)=\ell_R(J)+\ell_R(A/J). Consequently, ℓR(A/J)≤ℓR(A)\ell_R(A/J)\leq\ell_R(A). Combining this with the previous inequality ℓR(A)≤ℓR(A/J)\ell_R(A)\leq\ell_R(A/J) yields ℓR(A)=ℓR(A/J)\ell_R(A)=\ell_R(A/J), and so ℓR(J)=0\ell_R(J)=0. Therefore J=0J=0. □

Here, we relate the problem to co-Hopfianity, indeed, this is a natural extension from artinian rings to zero-dimensional rings (not necessarily noetherian):

Lemma 2.8. Let RR be a ring such that every prime ideal of RR is maximal. Let AA be a finite RR-algebra. Suppose JJ is an ideal of AA and there exists an RR-algebra monomorphism A↪RA/JA\hookrightarrow_R A/J. If J=(0:a)J=(0:a) then J=0J=0.

Proof. First recall that AA is finite as an RR-module, and recall that

A↪RA/J=A/(0:Aa)=aA⊆A,A\hookrightarrow_R A/J=A/(0:_A a)=aA\subseteq A,

is injective. By the main result of [5] this map should be surjective. This implies that aA=AaA=A, i.e., aa is invariable, we conclude from this that J=(0:a)=0J=(0:a)=0. □

Proposition 2.9. Let RR be a ring, let AA be a finitely generated projective RR-module, and let JJ be an RR-submodule of AA. Suppose that there exists an injective RR-linear map φ:A↪RA/J\varphi:A\hookrightarrow_R A/J. If every injective endomorphism of AA is an automorphism, then J=0J=0.

Proof. Let π:A↠A/J\pi:A\twoheadrightarrow A/J be the canonical surjection. Since AA is projective as an RR-module, the map φ:A↪RA/J\varphi:A\hookrightarrow_R A/J lifts along π\pi. Thus there exists an RR-linear map g:A→Ag:A\to A such that π∘g=φ\pi\circ g=\varphi:

0⟶J⟶A⟶A/J⟶00\longrightarrow J\longrightarrow A\longrightarrow A/J\longrightarrow0

We claim that gg is injective. Indeed, if g(a)=0g(a)=0, then φ(a)=(π∘g)(a)=0\varphi(a)=(\pi\circ g)(a)=0. Since φ\varphi is injective, it follows that a=0a=0. Hence gg is injective. By hypothesis, every injective endomorphism of AA is an automorphism. Therefore gg is an automorphism of AA. Since φ\varphi is injective, we have ker⁡(φ)=0\ker(\varphi)=0. Since

0=ker⁡(φ)=ker⁡(π∘g)=g−1(ker⁡π)=g−1(J),0=\ker(\varphi)=\ker(\pi\circ g)=g^{-1}(\ker\pi)=g^{-1}(J),

we obtain g−1(J)=0g^{-1}(J)=0. Since gg is surjective, for every j∈Jj\in J there exists a∈Aa\in A such that g(a)=jg(a)=j. But then a∈g−1(J)=0a\in g^{-1}(J)=0, and hence j=g(a)=0j=g(a)=0. Therefore J=0J=0. □

Corollary 2.10. Let RR be zero-dimensional. Let AA be a finite RR-algebra projective as module. Suppose for some ideal JJ of AA there exists an RR-algebra monomorphism A↪RA/JA\hookrightarrow_R A/J. Then J=0J=0.

Proof. Since AA is finite and projective over the zero-dimensional ring RR, every injective RR-linear endomorphism of AA is an automorphism. Hence Proposition 2.9 applies. □

It is easy to construct rings with nilpotent elements satisfying Serre’s condition (Sn)(S_n).

Proposition 2.11. Let (R,m)→(A,mA)(R,\mathbf m) \to(A,\mathbf m_A) be an extension of local rings so that AA is a finitely generated extension and (S2)(S_2) as an RR-module. Let J∈Spec⁡(A)∘:=Spec⁡(A)∖{mA}J \in\operatorname{Spec}(A)^\circ:= \operatorname{Spec}(A) \setminus\{\mathbf m_A\}. Suppose that there exists an injective RR-linear map φ:A↪RA/J\varphi: A \hookrightarrow_R A/J. Then J=0J=0.

Proof. We proceed by induction on d:=dim⁡(R)d := \dim(R) to observe J=0J=0. First assume that d=0d=0. According to Lemma 2.7, J=0J=0. We apply the inductive hypothesis, and observe that JJ is zero over punctured spectrum of RR. It turns out that C:=coker⁡(φ)C := \operatorname{coker}(\varphi) is of finite length.

Here, we separate the case d=1d=1. This implies that J∈min⁡(A)J \in\min(A). Set P=J∩RP=J\cap R. It turns out that φJ:AJ↪RPAJ/JAJ\varphi_J : A_J \hookrightarrow_{R_P} A_J/J A_J. Recall that dim⁡(RP)=0\dim(R_P)=0. Again, we apply the case d=0d=0, and deduce that JAJ=0J A_J=0. Consequently, J=0J=0.

Then we may and do assume that d>1d>1. By the independence theorem for local cohomology modules, we have Hm0(A/J)≅HmA0(A/J)H^0_{\mathbf m}(A/J) \cong H^0_{\mathbf m_A}(A/J), and this is zero because JJ is a prime ideal and J≠mAJ\ne\mathbf m_A, which imply that depth⁡(A/J)>0\operatorname{depth}(A/J)>0. Similarly, Hm1(A)=0H^1_{\mathbf m}(A)=0 by the (S2)(S_2) condition applied along with d>1d>1, and with the convenience that dim⁡(R)=dim⁡(A)\dim(R)=\dim(A), as any finite extension is integral (see [4 9.1 (a)]). Consider the exact sequence

0⟶A→φA/J⟶C⟶0.0 \longrightarrow A \xrightarrow{\varphi} A/J \longrightarrow C \longrightarrow0.

and deduce the exact sequence

0=Hm0(A/J)⟶Hm0(C)⟶Hm1(A)=0.0=H^0_{\mathbf m}(A/J)\longrightarrow H^0_{\mathbf m}(C)\longrightarrow H^1_{\mathbf m}(A)=0.

Also, Hm0(C)≅RCH^0_{\mathbf m}(C)\cong_R C, as CC is of finite length as an RR-module. Thus C=0C=0. Here, we show J=0J=0. Suppose on the way of contradiction that J≠0J\ne0. Since C=0C=0 we have A≅RA/JA\cong_R A/J. This allows us to apply Proposition 2.1 which say AA admits a surjective RR-algebra endomorphism which is not injective. Since AA is assumed to be finitely generated as an RR-module, [4 Theorem 2.4] gives us a contradiction that we searched for it. Then J=0J=0. □

Cohen-Macaulay rings are natural generalization of artinian ring. In this regard we present:

Proposition 2.12. Let (R,m)(R,\mathbf m) be a noetherian local ring Let AA be a finite RR-algebra so that there exists an RR-algebra monomorphism φ:A↪RA/J\varphi: A \hookrightarrow_R A/J with C:=coker⁡(φ)C := \operatorname{coker}(\varphi). If CC is maximal Cohen-Macaulay over RR, then J=0J=0.

Proof. We argue by induction on d:=dim⁡(R)d := \dim(R). If d=0d=0, Lemma 2.7 implies that J=0J=0. Assume now that d>0d>0. We may choose a parameter ideal x‾:=x1,…,xd⊆m\underline{x}:=x_1,\ldots,x_d\subseteq\mathbf m which is CC-regular. Consider the exact sequence

0⟶A→φA/J⟶C⟶0(∗)0\longrightarrow A\xrightarrow{\varphi} A/J\longrightarrow C\longrightarrow0 \quad(*)

We obtain by tensoring (∗)(*) with R/(x‾)R/(\underline{x}), the following exact sequence

Tor⁡1R(R/(x‾),C)⟶A/(x‾)⟶(A/J)⊗RR/(x‾)⟶C⊗RR/(x‾)⟶0.\operatorname{Tor}^R_1(R/(\underline{x}),C)\longrightarrow A/(\underline{x})\longrightarrow(A/J)\otimes_R R/(\underline{x})\longrightarrow C\otimes_R R/(\underline{x})\longrightarrow0.

But, Tor⁡1R(R/(x‾),C)=0\operatorname{Tor}^R_1(R/(\underline{x}),C)=0 because x‾\underline{x} is CC-regular (see [2 Ex. 1.1.12 (a)]). Thus

A/x‾A⟶(A/J)/x‾(A/J)=(A/x‾)/(J+x‾)(A/x‾)A/\underline{x}A\longrightarrow(A/J)/\underline{x}(A/J)=(A/\underline{x})/(J+\underline{x})(A/\underline{x})

is an R/(x‾)R/(\underline{x})-algebra monomorphism. By induction, J+x‾A=x‾AJ+\underline{x}A=\underline{x}A. Hence J⊆x‾AJ\subseteq\underline{x}A. Working with x‾n:=x1n,…,xdn\underline{x}^n:=x_1^n,\ldots,x_d^n gives J⊆x‾nAJ\subseteq\underline{x}^n A. Consequently,

J⊆⋂n>0x‾nA⊆⋂n>0(x‾A)n.J\subseteq\bigcap_{n>0}\underline{x}^n A\subseteq\bigcap_{n>0}(\underline{x}A)^n.

So by Krull’s intersection theorem, we deduce that J=0J=0. □

The finiteness assumption in Proposition 2.12 is needed, even if CC is big Cohen-Macaulay:

Example 2.13. Let RR be local ring, xx is a variable. Let M=R[x]u⊕R[x]vM=R[x]u\oplus R[x]v be a free R[x]R[x]-module with basis {u,v}\{u,v\}. Consider the trivial extension A:=R[x]⋉MA:=R[x]\ltimes M and let J:=AvJ:=Av. Then A/J=R[x]⊕R[x]uA/J=R[x]\oplus R[x]u with u2=0u^2=0. Define φ:A→A/J\varphi:A\to A/J to be the RR-algebra homomorphism such that φ(x)=x2\varphi(x)=x^2, φ(u)=u\varphi(u)=u, and φ(v)=xu\varphi(v)=xu. By [1 Example 1.3] we see φ(A)=R[x2]⊕R[x]u\varphi(A)=R[x^2]\oplus R[x]u and φ(A)\varphi(A) is a direct sum, hence φ\varphi is injective. Now, consider the exact sequence

0⟶A→φA/J⟶C⟶0.0\longrightarrow A\xrightarrow{\varphi}A/J\longrightarrow C\longrightarrow0.

It follows that C=R[x]/R[x2]≅RR⊕RxoddC=R[x]/R[x^2]\cong_R R\oplus Rx^{odd} is free as an RR-module, and so CC is a big Cohen-Macaulay over RR.

In opposite direction, we have a situation when C=0C=0.

Proposition 2.14. Let (R,m)(R,\mathfrak m) be local non Cohen-Macaulay and AA be an RR-algebra finite and over RR so that there is φ:A↪RA/J\varphi:A\hookrightarrow_R A/J. If C:=coker⁡(φ)C:=\operatorname{coker}(\varphi) is of finite projective dimension, then C=0C=0. In particular, J=0J=0.

Proof. We proceed by induction on dim⁡(R)\dim(R) to observe C=0C=0. First assume that dim⁡(R)=0\dim(R)=0. According to Lemma 2.7, C=J=0C=J=0. We apply the inductive hypothesis, and observe that CC is zero over punctured spectrum. It turns out that CC is of finite length. Suppose on the way of contradiction that CC is nonzero. This implies that RR is Cohen-Macaulay (see [2 Ex. 9.4.9]). But this excluded by the assumption. So, C=0C=0. Here, we show J=0J=0. Suppose on the way of contradiction that J≠0J\ne0. Since C=0C=0 we have A≅RA/JA\cong_R A/J. This allows us to apply Proposition 2.1 which say AA admits a surjective RR-algebra endomorphism which is not injective. Since AA is assumed to be finitely generated as an RR-module, [4 Theorem 2.4] gives us a contradiction that we searched for it. Then J=0J=0.

Remark 2.15. (i) The finiteness assumption in Proposition 2.14 is needed, it is enough to start Example 2.13 with any non Cohen-Macaulay ring RR. Recall that we observed in Example 2.13 that CC were be free, and so be of finite projective dimension.

(ii) Recall that the existence of a nonzero m\mathfrak m-torsion module of finite projective dimension implies the ring is Cohen-Macaulay. This turns out that the finiteness assumption in the artinan situation (see Lemma 2.7) is needed too, as otherwise Proposition 2.14 combined with part (i) leads a contradiction.

Question 1.1 is true

In this section we show:

Theorem 3.1. Let RR be a commutative ring, let AA be a reduced finitely generated RR-algebra, and let JJ be an ideal of AA. If there is an injective RR-algebra homomorphism φ:A↪RA/J\varphi:A\hookrightarrow_R A/J, then J=0J=0.

Proof. Let ι:R→A\iota:R\to A be the structure map and let π:A→A/J\pi:A\to A/J be the quotient map. Choose a1,…,an∈Aa_1,\ldots,a_n\in A such that A=ι(R)[a1,…,an]A=\iota(R)[a_1,\ldots,a_n]. For each ii, choose fi∈R[X1,…,Xn]f_i\in R[X_1,\ldots,X_n] with

φ(ai)=π(fi(a1,…,an)).\varphi(a_i)=\pi(f_i(a_1,\ldots,a_n)).

Evaluation of polynomials over RR uses the structure map ι\iota.

Fix j∈Jj\in J. Choose g∈R[X1,…,Xn]g\in R[X_1,\ldots,X_n] such that j=g(a1,…,an)j=g(a_1,\ldots,a_n). Let c1,…,cm∈Rc_1,\ldots,c_m\in R comprise all coefficients occurring in f1,…,fn,gf_1,\ldots,f_n,g. Let R0⊆RR_0\subseteq R be the subring generated by 1R,c1,…,cm1_R,c_1,\ldots,c_m. There is a surjective ring homomorphism

Z[T1,…,Tm]⟶R0,Tℓ⟼cℓ.\mathbb Z[T_1,\ldots,T_m]\longrightarrow R_0,\qquad T_\ell\longmapsto c_\ell.

Thus R0R_0 is noetherian. Set

B=ι(R0)[a1,…,an]⊆A,L=B∩J.B=\iota(R_0)[a_1,\ldots,a_n]\subseteq A,\qquad L=B\cap J.

The ring BB is reduced because it is a subring of AA. It is a finitely generated R0R_0-algebra. The kernel of the natural surjection R0[X1,…,Xn]⟶BR_0[X_1,\ldots,X_n]\longrightarrow B is finitely generated, since its source is noetherian. Hence BB is finitely presented over R0R_0. The coefficients of gg belong to R0R_0, so j∈Lj\in L. For r∈R0r \in R_0, RR-algebra compatibility gives φ(ι(r))=π(ι(r))\varphi(\iota(r))=\pi(\iota(r)). The coefficients of the fif_i also belong to R0R_0. It follows that

φ(B)⊆π(B).\varphi(B) \subseteq\pi(B).

The restriction π∣B\pi|_B has kernel LL and induces an R0R_0-algebra isomorphism

θ:B/L→∼π(B),b+L↦q(b).\theta:B/L \xrightarrow{\sim} \pi(B), \qquad b+L \mapsto q(b).

Consequently

ψ=θ−1∘φ∣B:B⟶B/L\psi=\theta^{-1}\circ\varphi|_B:B\longrightarrow B/L

is an injective R0R_0-algebra homomorphism. Indeed, φ∣B\varphi|_B is injective and θ\theta is an isomorphism. Bass’s theorem (see Theorem 1.2), applied to BB and LL, yields L=0L=0. Therefore j=0∈B⊆Aj=0\in B\subseteq A. Since j∈Jj\in J was arbitrary, J=0J=0. □

Remark 3.2 (Characteristic). The homomorphism Z→R0\mathbb{Z}\to R_0 sends kk to k⋅1R0k\cdot1_{R_0} and need not be injective. In characteristic n>0n>0, the polynomial-ring surjection above factors through (Z/nZ)[T1,…,Tm](\mathbb{Z}/n\mathbb{Z})[T_1,\ldots,T_m]; for prime n=pn=p, this is Fp[T1,…,Tm]\mathbb{F}_p[T_1,\ldots,T_m]. Thus the proof applies in every characteristic. The structure map R0→BR_0\to B may also have a kernel.

Toward module version

The preceding results suggest that the essential obstruction to embedding a module into a proper quotient is controlled by the zero-divisors of the base ring. We begin with the following observation.

Proposition 4.1. Let RR be a noetherian reduced ring, J◃RJ\triangleleft R an ideal, and MM a finitely generated torsion-free RR-module. If M↪R(M/JM)M\hookrightarrow_R(M/JM), then J⊆pJ\subseteq\mathfrak p for some p∈Min⁡(R)\mathfrak p\in\operatorname{Min}(R). Assume in addition MM is faithful. Then J=0J=0.

Proof. Let φ:M⟶M/JM\varphi:M\longrightarrow M/JM be an RR-linear injection. For every j∈Jj\in J and m∈Mm\in M, we have jm∈JMjm\in JM, and hence jφ(m)=φ(jm)=0j\varphi(m)=\varphi(jm)=0 in M/JMM/JM. Since φ\varphi is injective, it follows that jm=0jm=0 for every m∈Mm\in M. Therefore JM=0JM=0. We claim that every element of JJ is a zero-divisor of RR. Indeed, suppose that j∈Jj\in J is regular. Since MM is torsion-free, multiplication by jj on MM is injective. But jM=0jM=0, and M≠0M\ne0, a contradiction. Hence J⊆Zd⁡(R)J\subseteq\operatorname{Zd}(R). Since RR is noetherian and reduced, Zd⁡(R)=⋃p∈Min⁡(R)p\operatorname{Zd}(R)=\bigcup_{\mathfrak p\in\operatorname{Min}(R)}\mathfrak p. Moreover, Min⁡(R)\operatorname{Min}(R) is finite. Consequently,

J⊆⋃p∈Ass⁡(R)p=⋃p∈Min⁡(R)p.J\subseteq\bigcup_{\mathfrak p\in\operatorname{Ass}(R)}\mathfrak p=\bigcup_{\mathfrak p\in\operatorname{Min}(R)}\mathfrak p.

By prime avoidance, there exists p∈Min⁡(R)\mathfrak p\in\operatorname{Min}(R) such that J⊆pJ\subseteq\mathfrak p.

Assume now MM is faithful. Then

[M↪RM/JM⟹J=0].[M\hookrightarrow_R M/JM\quad\Longrightarrow\quad J=0].

The first paragraph gives JM=0JM=0. Since MM is faithful, Ann⁡R(M)=0\operatorname{Ann}_R(M)=0, and hence J=0J=0. □

Remark 4.2. The faithfulness assumption is important since J:=ann⁡(M)J:=\operatorname{ann}(M). Then M≅M/JM=M/0=MM\cong M/JM=M/0=M.

The preceding proposition uses reducedness only through the description of the zero-divisors as the union of the minimal primes. We next give a different argument which, in dimension one, does not require AA to be reduced.

Proposition 4.3. Let R→AR\to A be local and finite morphism of noetherian local rings where dim⁡(A)=1\dim(A)=1 so that there exists an RR-algebra monomorphism φ:A↪RA/J\varphi:A\hookrightarrow_R A/J. Then J⊆pJ\subseteq\mathfrak p for some p∈Min⁡(A)\mathfrak p\in\operatorname{Min}(A).

Proof. Suppose not. By ht⁡(J)>0\operatorname{ht}(J)>0, and so JJ is mA\mathfrak m_A-primary. Then by Grothendieck’s vanishing theorem, HmA1(A/J)=0H^1_{\mathfrak m_A}(A/J)=0. It is easy to see rad⁡(mRA)=mA\operatorname{rad}(\mathfrak m_R A)=\mathfrak m_A. By the independent theorem for local cohomology modules, HmAi(−)≅RHmRi(−)H^i_{\mathfrak m_A}(-)\cong_R H^i_{\mathfrak m_R}(-). In particular, HmA1(A/J)=0H^1_{\mathfrak m_A}(A/J)=0. Now, we use the natural short exact sequence of RR-modules 0→A→A/J→C→00\to A\to A/J\to C\to0, to deduce the exact sequence

HmR0(C)⟶HmR1(A)⟶HmR1(A/J)=0.(1)H^0_{\mathfrak m_R}(C)\longrightarrow H^1_{\mathfrak m_R}(A)\longrightarrow H^1_{\mathfrak m_R}(A/J)=0. \tag*{(1)}

Now, recall from finiteness of R→AR \to A that A/JA/J and so CC is finitely generated over RR. Consequently, HmR0(C)H^0_{\mathfrak m_R}(C) is finitely generated. Then HmR1(A)H^1_{\mathfrak m_R}(A) is finitely generated as an RR-module. Therefore, HmAdim⁡(A)(A)H^{\dim(A)}_{\mathfrak m_A}(A) is finitely generated as an AA-module. This contradiction shows that J⊆pJ \subseteq\mathfrak p for some p∈Min⁡(A)\mathfrak p \in\operatorname{Min}(A). □

Question 4.4. Is in Proposition 4.1 the noetherian assumption important?

The answer is affirmative.

Example 4.5. Let kk be a field and set R=k[X1,X2,…]/(XiXj∣i≠j)R = k[X_1, X_2, \ldots]/(X_iX_j \mid i \ne j). For each n≥1n \ge1, let pn=(xi∣i≠n)\mathfrak p_n = (x_i \mid i \ne n). Then R/pn≅k[xn]R/\mathfrak p_n \cong k[x_n], so each pn\mathfrak p_n is prime. Moreover, RR is reduced and Min⁡(R)={pn∣n≥1}\operatorname{Min}(R) = \{\mathfrak p_n \mid n \ge1\}. Let

J=(x1,x2,…)andM=R/J.J = (x_1,x_2,\ldots) \qquad\text{and} \qquad M = R/J.

Since JM=0JM = 0, the canonical map M→M/JMM \to M/JM is an isomorphism. We claim that MM is torsion-free. Every element of RR can be written in the form

r=c+f1(x1)+⋯+ft(xt),r = c + f_1(x_1) + \cdots+ f_t(x_t),

where c∈kc \in k and fi(xi)∈(xi)f_i(x_i) \in(x_i). If c=0c = 0, then, for any n>tn > t, r∈pnr \in\mathfrak p_n, and hence rr is a zero-divisor. Consequently, every regular element of RR has nonzero constant term. Now M=R/J≅kM = R/J \cong k. If r∈Rr \in R is regular, its action on MM is multiplication by its nonzero constant term. Hence multiplication by rr on MM is injective. Thus MM is torsion-free. However, for every n≥1n \ge1, xn∈Jx_n \in J but xn∉pnx_n \notin\mathfrak p_n. Therefore J⊈pnJ \nsubseteq\mathfrak p_n for every $n \ge 1. Since the pn\mathfrak p_n are precisely the minimal primes of RR, we obtain J⊈pJ \nsubseteq\mathfrak p for every p∈Min⁡(R)\mathfrak p \in\operatorname{Min}(R). Thus M↪RM/JMM \hookrightarrow_R M/JM does not imply the conclusion of Proposition 4.1 when RR is not noetherian.

But, we can reduce the finitely generated of MM to only has rank (see [2 §1.4]).

Corollary 4.6. Let RR be a noetherian reduced ring, let J⊲RJ \lhd R be an ideal, and let M≠0M \ne0 be a torsion-free RR-module and has rank. If M↪RM/JMM \hookrightarrow_R M/JM, then J⊆pJ \subseteq\mathfrak p for some p∈Min⁡(R)\mathfrak p \in\operatorname{Min}(R).

Proof. The proof of Proposition 4.1 applies verbatim. Indeed, the only properties of MM used there are M≠0M \ne0 and the fact that every regular element of RR acts injectively on MM. No finite generation assumption is required. □

Observation 4.7. Let RR be a noetherian reduced ring, J⊆RJ \subseteq R an ideal, and MM a nonzero divisible RR-module such that JM=0JM = 0. Then JJ is contained in some minimal prime ideal of RR.

Proof. If by divisible we mean divisible by every regular element of RR, then the argument is very clean. Suppose 0≠M0 \ne M is divisible and JM=0JM = 0. If j∈Jj \in J were regular, divisibility gives M=jMM = jM. But JM=0JM = 0 gives jM=0jM = 0. Hence M=0M = 0, contradiction. Thus every j∈Jj \in J is a zero-divisor. Since RR is noetherian and reduced, Z ⁣d(R)=⋃p∈Min⁡(R)pZ\!d(R) = \bigcup_{\mathfrak p \in\operatorname{Min}(R)} \mathfrak p. Because there are finitely many minimal primes, prime avoidance gives J⊆pJ \subseteq\mathfrak p for some p∈Min⁡(R)\mathfrak p \in\operatorname{Min}(R). □

We continue by extending Proposition 4.1. To this end, we record a remark.

Remark 4.8. For a general module MM, the set Ass⁡R(M)\operatorname{Ass}_R(M) may be empty or infinite. If RR is Noetherian and M≠0M \ne0 is finitely generated, then Ass⁡R(M)\operatorname{Ass}_R(M) is finite and nonempty. In the non-Noetherian setting, one sometimes uses instead the weakly associated primes: p∈wAss⁡R(M)\mathfrak p \in\operatorname{wAss}_R(M) if p\mathfrak p is a prime ideal minimal over Ann⁡R(m)\operatorname{Ann}_R(m) for some m∈Mm \in M. For more details, see [3 3.3.1].

Theorem 4.9. Let R⊆AR \subseteq A be an extension of noetherian reduced rings such that AA is torsion-free as an RR-module. Let J⊲AJ \lhd A, and let MM be a finitely generated RR-module which is torsion-free as an AA-module. If M↪RM/JMM \hookrightarrow_R M/JM, then J⊆pJ \subseteq\mathfrak p for some p∈Min⁡(A)\mathfrak p \in\operatorname{Min}(A).

Proof. First, R⊆AR \subseteq A and MM is torsion-free as an AA-module imply that MM is torsion-free as an RR-module. By [4 Page 183], a ring is reduced if and only if it satisfies (R0)(R_0) and (S1)(S_1).

(R0):Rp is regular ∀p with ht⁡(p)=0⇒Rp is a field.(R_0):\quad R_{\mathfrak p}\text{ is regular }\forall\mathfrak p\text{ with }\operatorname{ht}(\mathfrak p)=0 \Rightarrow R_{\mathfrak p}\text{ is a field}.
(S1):∀p∈Ass⁡(R)⇒p∈Min⁡(R).(S_1):\quad\forall\mathfrak p \in\operatorname{Ass}(R) \Rightarrow\mathfrak p \in\operatorname{Min}(R).

Now, let p∈Ass⁡(R)\mathfrak{p} \in\operatorname{Ass}(R). Consider the exact sequence 0→JM→M→M/JM→00 \to JM \to M \to M/JM \to0. Localizing at p\mathfrak{p}, we obtain

0→(JM)p→Mp→(M/JM)p→0.0 \to(JM)_{\mathfrak{p}} \to M_{\mathfrak{p}} \to(M/JM)_{\mathfrak{p}} \to0.

Here, vdim⁡(−)\operatorname{vdim}(-) means the vector space dimension over the field RpR_{\mathfrak{p}}. On the one hand we have the additivity property of dimension on short exact sequence. Namely, we have

vdim⁡(Mp)=vdim⁡((M/JM)p)+vdim⁡((JM)p)≥vdim⁡((M/JM)p).\operatorname{vdim}(M_{\mathfrak{p}}) = \operatorname{vdim}((M/JM)_{\mathfrak{p}}) + \operatorname{vdim}((JM)_{\mathfrak{p}}) \ge\operatorname{vdim}((M/JM)_{\mathfrak{p}}).

On the other have and in view of the injection Mp→(M/JM)pM_{\mathfrak{p}} \to(M/JM)_{\mathfrak{p}}, we have

vdim⁡(Mp)≤vdim⁡((M/JM)p).\operatorname{vdim}(M_{\mathfrak{p}}) \le\operatorname{vdim}((M/JM)_{\mathfrak{p}}).

Combining these yield vdim⁡(Mp)=vdim⁡((M/JM)p)\operatorname{vdim}(M_{\mathfrak{p}}) = \operatorname{vdim}((M/JM)_{\mathfrak{p}}). Consequently,

vdim⁡(JM)p=0⇒(JM)p=0(∗)\operatorname{vdim}(JM)_{\mathfrak{p}} = 0 \Rightarrow(JM)_{\mathfrak{p}} = 0 \qquad(*)

Since wAss⁡\operatorname{wAss} is subset of support, p∉wAss⁡(JM)\mathfrak{p} \notin\operatorname{wAss}(JM). But JM⊆MJM \subseteq M, so

wAss⁡R(JM)⊆wAss⁡R(M)⊆wAss⁡R(R)=Ass⁡R(R).\operatorname{wAss}_{R}(JM) \subseteq\operatorname{wAss}_{R}(M) \subseteq\operatorname{wAss}_{R}(R) = \operatorname{Ass}_{R}(R).

It turns out from (∗)(*) that Ass⁡R(JM)=∅\operatorname{Ass}_{R}(JM) = \varnothing and so JM=0JM = 0, because it is torsion-free over RR. In particular, it is zero as an AA-module which gives

J⊆Zd⁡A(M)=⋃p∈wAss⁡(M)p⊆⋃p∈Ass⁡(A)p.J \subseteq\operatorname{Zd}_{A}(M) = \bigcup_{\mathfrak{p} \in\operatorname{wAss}(M)} \mathfrak{p} \subseteq\bigcup_{\mathfrak{p} \in\operatorname{Ass}(A)} \mathfrak{p}.

Therefore, by prime avoidance J⊆pJ \subseteq\mathfrak{p} for some p∈Ass⁡(A)\mathfrak{p} \in\operatorname{Ass}(A).

Corollary 4.10. Adopt the above notation. If MM is in addition faithful over AA, then J=0J = 0.

But, we can reduce the finitely generated of MM to only has rank (see [2 §1.4]).

Remark 4.11. Let R⊆AR \subseteq A be a noetherian reduced ring, J⊲AJ \lhd A an ideal, and MM has rank over RR and torsion-free as an AA-module. If M↪R(M/JM)M \hookrightarrow_{R} (M/JM), then J⊆pJ \subseteq\mathfrak{p} for some p∈Min⁡(A)\mathfrak{p} \in\operatorname{Min}(A).

The finiteness of MM as RR-module is important:

Example 4.12. Let R:=kR := k be a field, set M=A=k[x,y]M = A = k[x,y] and J=(x)J = (x). Then AA is a Noetherian reduced ring and MM is a finitely generated torsion-free AA-module. We have M/JM=A/xA≅k[y]M/JM = A/xA \cong k[y]. As kk-vector spaces, A=R[x,y]A = R[x,y] and R[y]R[y] both have countably infinite dimension over kk. Hence there exists a kk-linear isomorphism φ:A→A/xA\varphi: A \to A/xA. For example, choose a bijection π:N2→N\pi: \mathbb{N}^{2} \to\mathbb{N} and define φ(xayb)=yπ(a,b)\varphi(x^{a}y^{b}) = y^{\pi(a,b)} with (a,b≥0)(a,b \ge0), extending kk-linearly. Thus M↪RM/JMM \hookrightarrow_{R} M/JM. However, Min⁡(R)={(0)}\operatorname{Min}(R) = \{(0)\}, whereas J=(x)≠(0)J = (x) \ne(0). Consequently, J⊈pJ \not\subseteq\mathfrak{p} for every p∈Min⁡(R)\mathfrak{p} \in\operatorname{Min}(R).

The dual surjection problem

We now consider a dual version of Question 1.1. Instead of asking whether AA can be embedded into a proper quotient A/JA/J, we ask whether a nonzero ideal JJ can be mapped onto AA, or whether AA can be mapped onto JJ.

Question 5.1. Let AA be a finitely generated RR-algebra and let 0≠J⊆A0 \ne J \subseteq A be an ideal. (i): Suppose that there exists an RR-linear epimorphism A↠RJA \twoheadrightarrow_{R} J, or (ii): suppose there is an RR-linear epimorphism J↠RAJ \twoheadrightarrow_{R} A. Under what hypotheses must J≅RAJ \cong_{R} A?

Question 5.1(i) is not true even π\pi is splitting, for example let A=⨁n≥0AnA = \bigoplus_{n \ge0} A_n be graded ring and let J=⨁n>0AnJ = \bigoplus_{n > 0} A_n be the irrelevant ideal and R:=A0R := A_0. The explicit finite extension is given by Remark 5.4.

Remark 5.2. Suppose R=AR = A. If there exists an AA-linear epimorphism A↠JA \twoheadrightarrow J, then JJ is a cyclic AA-module, and hence JJ is a principal ideal. On the other hand, if there exists an AA-linear epimorphism J↠AJ \twoheadrightarrow A the epimorphism splits. Thus J≅A⊕ker⁡(π)J \cong A \oplus\operatorname{ker}(\pi). In particular, if AA is an integral domain, then

rank⁡A(J)=1=rank⁡A(A),\operatorname{rank}_{A}(J) = 1 = \operatorname{rank}_{A}(A),

and consequently ker⁡(π)=0\operatorname{ker}(\pi) = 0. Hence J≅A.J \cong A.*

Theorem 5.3. Let R⊆AR \subseteq A be integral domains such that AA is finite as an RR-module. Let 0≠J⊆A0 \ne J \subseteq A be an ideal. Then rank⁡R(J)=rank⁡R(A)\operatorname{rank}_{R}(J) = \operatorname{rank}_{R}(A). Moreover, if there exists either an RR-linear epimorphism π:A↠RJ\pi: A \twoheadrightarrow_{R} J or an RR-linear epimorphism π:J↠RA\pi: J \twoheadrightarrow_{R} A, then π\pi is an isomorphism. In particular, A≅RJA \cong_{R} J.

Proof. Let Q:=Frac⁡(R)Q := \operatorname{Frac}(R). Since AA is a domain and is finite over RR, the QQ-algebra AQ:=A⊗RQA_{Q} := A \otimes_{R} Q is a finite-dimensional domain over the field QQ. Hence AQA_{Q} is a field. Since 0≠J⊆A0 \ne J \subseteq A, we have

0≠J⊗RQ⊆A⊗RQ=AQ.0 \ne J \otimes_{R} Q \subseteq A \otimes_{R} Q = A_{Q}.

As AQA_{Q} is a field, it follows that J⊗RQ=AQJ \otimes_{R} Q = A_{Q}. Consequently,

rank⁡R(J)=vdim⁡Q(J⊗RQ)=vdim⁡Q(A⊗RQ)=rank⁡R(A).\operatorname{rank}_{R}(J) = \operatorname{vdim}_{Q}(J \otimes_{R} Q) = \operatorname{vdim}_{Q}(A \otimes_{R} Q) = \operatorname{rank}_{R}(A).

Suppose first that there exists an RR-linear epimorphism π:A↠RJ\pi: A \twoheadrightarrow_{R} J. Let K:=ker⁡(π)K := \ker(\pi). Since QQ is flat over RR, tensoring the exact sequence $0 \to K \to A \xrightarrow{\pi} J \to 0$ with QQ gives an exact sequence

0⟶K⊗RQ⟶A⊗RQ⟶J⊗RQ⟶0.0 \longrightarrow K \otimes_{R} Q \longrightarrow A \otimes_{R} Q \longrightarrow J \otimes_{R} Q \longrightarrow0.

Since A⊗RQ=J⊗RQA \otimes_{R} Q = J \otimes_{R} Q, we obtain K⊗RQ=0K \otimes_{R} Q = 0. But K⊆AK \subseteq A and AA is torsion-free as an RR-module. Hence KK is torsion-free as an RR-module, and therefore K=0K = 0. Thus π\pi is an isomorphism.

Now suppose that there exists an RR-linear epimorphism π:J↠RA\pi: J \twoheadrightarrow_{R} A. Again let K:=ker⁡(π)K := \ker(\pi). Tensoring $0 \to K \to J \xrightarrow{\pi} A \to 0$ with QQ gives

0⟶K⊗RQ⟶J⊗RQ⟶A⊗RQ⟶0.0 \longrightarrow K \otimes_{R} Q \longrightarrow J \otimes_{R} Q \longrightarrow A \otimes_{R} Q \longrightarrow0.

Since J⊗RQ=A⊗RQJ \otimes_{R} Q = A \otimes_{R} Q, we obtain K⊗RQ=0K \otimes_{R} Q = 0. As K⊆J⊆AK \subseteq J \subseteq A and AA is torsion-free over RR, we again obtain K=0K = 0. Hence π\pi is an isomorphism. Therefore, in either case, A≅RJA \cong_{R} J. □

Remark 5.4. The domain hypothesis in Theorem 5.3 cannot be omitted in general. Let A=k[x,y]/(x,y)2A = k[x,y]/(x,y)^2 and J=(x,y)J = (x,y). Then A=k⊕kx⊕kyA = k \oplus kx \oplus ky, and J=kx⊕kyJ = kx \oplus ky. Define a kk-linear epimorphism A↠kJA \twoheadrightarrow_{k} J, by assignments 1↦x1 \mapsto x, x↦yx \mapsto y and y↦0y \mapsto0. This map even admits a kk-linear section. Nevertheless, J≠AJ \ne A. Thus an epimorphism A↠RJA \twoheadrightarrow_{R} J need not imply J≅RAJ \cong_{R} A over an artinian base.

Example 5.5. Let A=R=k[x,y]/(x,y)2A = R = k[x,y]/(x,y)^2 and let J=(x)J = (x). Define an RR-linear map π:A→J\pi: A \to J by π(1)=x\pi(1) = x, π(x)=0\pi(x) = 0 and π(y)=0\pi(y) = 0. Then π\pi is surjective. However, J≠AJ \ne A. Thus Question 5.1(i) has a negative answer even when R=AR = A is artinian.

Question 5.1(ii) is true, if RR is artinian:

Proposition 5.6. Let RR be an artinian ring, let AA be a finite RR-algebra, and let 0≠J⊆A0 \ne J \subseteq A be an ideal. If there exists an RR-linear epimorphism J↠RAJ \twoheadrightarrow_{R} A, then this epimorphism is an isomorphism. In particular, J≅RAJ \cong_{R} A.

Proof. Since J⊆AJ \subseteq A, we have ℓR(J)≤ℓR(A)\ell_{R}(J) \le\ell_{R}(A). On the other hand, an RR-linear epimorphism J↠AJ \twoheadrightarrow A gives ℓR(J)≥ℓR(A)\ell_{R}(J) \ge\ell_{R}(A). Hence ℓR(J)=ℓR(A)\ell_{R}(J) = \ell_{R}(A). If KK denotes the kernel of the epimorphism, then $0 \to K \to J \to A \to 0$ gives ℓR(J)=ℓR(K)+ℓR(A)\ell_{R}(J) = \ell_{R}(K) + \ell_{R}(A). Thus ℓR(K)=0\ell_{R}(K) = 0, and consequently K=0K = 0. □

Proposition 5.7. Let (R,m)(R,\mathfrak m) be a noetherian local ring of positive dimension, let R→AR \to A be a finite local homomorphism, and let JJ be an ideal of AA. Assume that A/JA/J is a maximal Cohen–Macaulay RR-module. Then there is no RR-linear epimorphism J↠RAJ \twoheadrightarrow_{R} A.

Proof. Suppose, to the contrary, that there exists an RR-linear epimorphism π:J↠A\pi: J \twoheadrightarrow A. Let d=dim⁡Rd = \dim R, and choose an RR-regular system of parameters x=x1,…,xd\mathbf{x} = x_{1},\ldots,x_{d} which is also A/JA/J-regular. Tensoring the exact sequence 0→J→A→A/J→00 \to J \to A \to A/J \to0 with R/(x)R/(\mathbf{x}), and using the A/JA/J-regularity of x\mathbf{x}, gives an injection J/xJ→A/xAJ/\mathbf{x}J \to A/\mathbf{x}A (see the argument presented in Proposition 2.12). On the other hand, tensoring the epimorphism J↠RAJ \twoheadrightarrow_{R} A with R‾:=R/(x)\overline{R} := R/(\mathbf{x}) gives an epimorphism J/xJ↠R‾A/xAJ/\mathbf{x}J \twoheadrightarrow_{\overline{R}} A/\mathbf{x}A. Hence

ℓ(J/xJ)=ℓ(A/xA)(∗)\ell(J/\mathbf{x}J) = \ell(A/\mathbf{x}A) \qquad(*)

Let us apply (∗)(\ast) along with the following induced short exact sequence

0⟶J/xJ⟶A/xA⟶(A/J)/x(A/J)⟶0,0 \longrightarrow J/xJ \longrightarrow A/xA \longrightarrow(A/J)/x(A/J) \longrightarrow0,

to deduce that (A/J)/x(A/J)=0(A/J)/x(A/J)=0. By Nakayama’s lemma, A/J=0A/J=0, contrary to the assumption that A/JA/J is a nonzero maximal Cohen–Macaulay RR-module. □

Remark 5.8. Let us start Example 2.13 with a Cohen-Macaulay ring RR. Recall from Example 2.13 that A/JA/J is RR-free and so big Cohen–Macaulay. In particular, the finite assumption of Proposition 5.7 is needed, even if AA is finitely generated as an RR-algebra.

Corollary 5.9. Suppose R→AR \to A is an extension of local rings of dimension one, and P∈Ass⁡(A)∖{mA}P \in\operatorname{Ass}(A) \setminus\{\mathfrak{m}_A\}. Then there is no π:P↠RA\pi: P \twoheadrightarrow_R A.

Proof. Since P∈Ass⁡(A)∖{mA}P \in\operatorname{Ass}(A) \setminus\{\mathfrak{m}_A\} we see 0<depth⁡(A/P)0 < \operatorname{depth}(A/P). We know 1≤depth⁡(A/P)≤dim⁡(A/P)≤dim⁡(A)=11 \le\operatorname{depth}(A/P) \le\operatorname{dim}(A/P) \le\operatorname{dim}(A)=1, i.e., A/PA/P is maximal Cohen–Macaulay. It remains to apply Proposition 5.7. □

Corollary 5.10. Suppose R=AR=A is a local ring of dimension one, and P∈Ass⁡(A)P \in\operatorname{Ass}(A). Then there is no π:P↠RA\pi: P \twoheadrightarrow_R A.

Proof. By Corollary 5.9, it is enough to deal with P=m∈Ass⁡(A)P=\mathfrak{m} \in\operatorname{Ass}(A). Suppose on the way of contradiction that there is π:P↠RA\pi: P \twoheadrightarrow_R A. Let xx be so that P=(0:Rx)P=(0 :_R x), such a thing exists by definition of associated primes. Then 0=π(0)=π(xP)=xπ(P)=xR0=\pi(0)=\pi(xP)=x\pi(P)=xR. Thus x=0x=0, and so P=(0:Rx)=(0:R0)=RP=(0 :_R x)=(0 :_R 0)=R, a contradiction, as primes are proper ideals. □

The assumption P∈Ass⁡(A)P \in\operatorname{Ass}(A) in Corollary 5.10 is needed:

Example 5.11. Let A:=R:=k[[X]]A:=R:=k[[X]], and set P:=(X)P:=(X). There is a surjective RR-linear map π:P↠RA\pi:P\twoheadrightarrow_R A defined by Xf(x)↦f(x)Xf(x)\mapsto f(x).

Most of the results extend to the module setting. Here, we present a sample parallel to Proposition 2.9.

Observation 5.12. Let MM be an injective module over AA. Let J⊆AJ \subseteq A be an ideal. If there exists an epimorphism π:JM↠AM\pi:JM\twoheadrightarrow_A M then JM=MJM=M. In particular, if JJ is nilpotent, then M=0M=0.

Proof. Since MM is injective, the map π:JM↠M\pi:JM\twoheadrightarrow M extends to an AA-homomorphism π~:M↠M\widetilde{\pi}:M\twoheadrightarrow M which is epic as well:

0→JM→⊆M π↓↓π~ M=M\begin{CD} 0 @>>> JM @>{\subseteq}>> M \\ @. @V{\pi}VV @VV{\widetilde{\pi}}V \\ @. M @= M \end{CD}

Since π~\widetilde{\pi} is AA-linear, it follows that M=π(JM)=π~(JM)⊆Jπ~(M)=JMM=\pi(JM)=\widetilde{\pi}(JM)\subseteq J\widetilde{\pi}(M)=JM. By reaping this process, M=JM=J(JM)=⋯=JnMM=JM=J(JM)=\cdots=J^nM for all nn, which imply the last claim. □

References

M. Asgharzadeh

Email address: [email protected]

E. Mahdavi

Email address: [email protected]

References

  1. [1]H. Bass, A finiteness property of affine algebras, Proceedings of the American Mathematical Society, 110 No. 2 (1990), 315–318.
  2. [2]W. Bruns and J. Herzog, Cohen–Macaulay rings, Cambridge University Press, 39, Cambridge, (1998).
  3. [3]S. Glaz, Commutative coherent rings, Springer-Verlag, LNM, 1371 (1989).DOI
  4. [4]H. Matsumura, Commutative Ring Theory, Cambridge Studies in Advanced Mathematics 8, Cambridge University Press, 1986.DOI
  5. [5]Wolmer V. Vasconcelos, Injective endomorphisms of finitely generated modules, Proc. Amer. Math. Soc. 25 (1970), 900–901.DOI

Paper details

vlnc:W_rshb46s8
Version
v1 · latest
Published
Added
License
CC BY 4.0
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents