Bourbaki

§ 8. Homological dimension

In this section, we adopt the conventions of § 5.

1. Projective Dimension of a Module

Definition 1

Let MM be an AA-module. The projective dimension of MM, denoted by dpA(M)\mathrm{dp}_A(M), is the greatest lower bound in Z‾\overline{\mathbf{Z}} of the lengths of projective resolutions of MM* (X, p. 48).

We thus have dpA(0)=−∞\mathrm{dp}_A(0) = -\infty, dpA(M)≥0\mathrm{dp}_A(M) \geq 0 if M≠0M \neq 0. For MM to be projective, it is necessary and sufficient that dpA(M)≤0\mathrm{dp}_A(M) \leq 0.

Lemma 1

If dpA(M)<n<+∞\mathrm{dp}_A(M) < n < +\infty, then ExtAn(M,N)=0\mathrm{Ext}_A^n(M, N) = 0 for every AA-module NN and TornA(P,M)=0\mathrm{Tor}_n^A(P, M) = 0 for every righthand AA-module PP. This follows immediately from the fact that MM has a projective resolution of length <n< n and from X, p. 100, Theorem 1.

Proposition 1

Let MM be an AA-module and nn an integer ≥0\geq 0. The following conditions are equivalent* : (i) dpA(M)≤n\mathrm{dp}_A(M) \leq n (i.e. (def. 1), MM has a projective resolution of length ≤n\leq n); (ii) ExtAr(M,N)=0\mathrm{Ext}_A^r(M, N) = 0 for every AA-module NN and every integer r>nr > n; (iii) ExtAn+1(M,N)=0\mathrm{Ext}_A^{n+1}(M, N) = 0 for every AA-module NN; (iv) for every exact sequence

0→K→Pn−1→…→P0→M→0 0 \to K \to P_{n-1} \to \ldots \to P_0 \to M \to 0
où les PiP_i sont projectifs, KK is projective.

(i) ⇒ (ii): this follows from Lemma 1. (ii) ⇒ (iii): this is trivial. (iii) ⇒ (iv): in the situation of (iv), for every AA-module NN there is an isomorphism of ExtA1(K,N)\mathrm{Ext}_A^1(K, N) onto ExtAn+1(M,N)\mathrm{Ext}_A^{n+1}(M, N) (X, p. 128, Corollary 4); if (iii) holds, then

ExtA1(K,N)=0 \mathrm{Ext}_A^1(K, N) = 0
for every NN and KK is projective (X, p. 93, Proposition 10). (iv) ⇒ (i): consider the exact sequence (X, p. 50)
0→Zn−1(M)→Ln−1(M)→…→L0(M)→M→0. 0 \to Z_{n-1}(M) \to L_{n-1}(M) \to \ldots \to L_0(M) \to M \to 0 .
If (iv) holds, then Zn−1(M)Z_{n-1}(M) is projective and MM has a projective resolution of length ≤n\leq n.

Corollary 1

Let (Mi)i∈E(M_i)_{i \in E} be a family of AA-modules. We have

dpA(⨁i∈EMi)=sup⁡i∈EdpA(Mi). \mathrm{dp}_A \left( \bigoplus_{i \in E} M_i \right) = \sup_{i \in E} \mathrm{dp}_A(M_i) .
This follows from the equivalence of conditions (i) and (iii) of Proposition 1 and Proposition 7 of X, p. 89.

In the following statement, we agree that ±∞+1=±∞−1=±∞\pm \infty + 1 = \pm \infty - 1 = \pm \infty.

Corollary 2

Let

0→M′→M→M′′→0 0 \to M' \to M \to M'' \to 0
be an exact sequence of AA-modules.

a) We have

dpA(M)≤sup⁡(dpA(M′),dpA(M′′)). \mathrm{dp}_A(M) \leq \sup (\mathrm{dp}_A(M'), \mathrm{dp}_A(M'')) .
Equality occurs as soon as dpA(M′′)≠dpA(M′)+1\mathrm{dp}_A(M'') \neq \mathrm{dp}_A(M') + 1.

b) We have

dpA(M′′)≤sup⁡(dpA(M),dpA(M′)+1). \mathrm{dp}_A(M'') \leq \sup (\mathrm{dp}_A(M), \mathrm{dp}_A(M') + 1) .
Equality occurs as soon as dpA(M)≠dpA(M′)\mathrm{dp}_A(M) \neq \mathrm{dp}_A(M').

c) We have

dpA(M′)≤sup⁡(dpA(M),dpA(M′′)−1). \mathrm{dp}_A(M') \leq \sup (\mathrm{dp}_A(M), \mathrm{dp}_A(M'') - 1) .
Equality occurs as soon as dpA(M)≠dpA(M′′)\mathrm{dp}_A(M) \neq \mathrm{dp}_A(M'').

Let us prove for example a), the proofs of b) and c) being analogous. If N is an arbitrary a-module and n an integer ≥0\geq 0, we have an exact sequence

ExtAn+1(M′′,N)→ExtAn+1(M,N)→ExtAn+1(M′,N)→ExtAn+2(M′′,N) \mathrm{Ext}_A^{n+1}(M'', N) \to \mathrm{Ext}_A^{n+1}(M, N) \to \mathrm{Ext}_A^{n+1}(M', N) \to \mathrm{Ext}_A^{n+2}(M'', N)
→ExtAn+2(M,N). \to \mathrm{Ext}_A^{n+2}(M, N) .
If dpA(M′),dpA(M′′)≤n\mathrm{dp}_A(M'), \mathrm{dp}_A(M'') \leq n, then ExtAn+1(M′,N)=0\mathrm{Ext}_A^{n+1}(M', N) = 0 and ExtAn+1(M′′,N)=0\mathrm{Ext}_A^{n+1}(M'', N) = 0 (prop. 1), hence ExtAn+1(M,N)=0\mathrm{Ext}_A^{n+1}(M, N) = 0 and dpA(M)≤n\mathrm{dp}_A(M) \leq n (prop. 1), so that
dpA(M)≤sup⁡(dpA(M′),dpA(M′′)). \mathrm{dp}_A(M) \leq \sup (\mathrm{dp}_A(M'), \mathrm{dp}_A(M'')) .

If dpA(M)<sup⁡(dpA(M′),dpA(M′′))\mathrm{dp}_A(M)<\sup\bigl(\mathrm{dp}_A(M'),\mathrm{dp}_A(M'')\bigr), then necessarily dpA(M)<+∞\mathrm{dp}_A(M)<+\infty. For every n>dpA(M)n>\mathrm{dp}_A(M) and every a-module N, we have

Ext⁡An+1(M′,N)≠0⟺Ext⁡An+2(M′′,N)≠0, \operatorname{Ext}_A^{n+1}(M',N)\ne0\Longleftrightarrow\operatorname{Ext}_A^{n+2}(M'',N)\ne0,

by the preceding exact sequence; by prop. 1, this immediately implies dpA(M′′)=dpA(M′)+1\mathrm{dp}_A(M'')=\mathrm{dp}_A(M')+1, since one of the quantities dpA(M′)\mathrm{dp}_A(M'), dpA(M′′)\mathrm{dp}_A(M'') is >dpA(M)>\mathrm{dp}_A(M).

Example. — Let aa be an element of A which is neither invertible nor a right zero divisor. Then dpA(A/Aa)=1\mathrm{dp}_A(A/Aa)=1.

Indeed, by the exact sequence 0⟶As→φAs⟶A/Aa⟶00\longrightarrow A_s\xrightarrow{\varphi} A_s\longrightarrow A/Aa\longrightarrow0, where φ(x)=xa\varphi(x)=xa, we have dpA(A/Aa)≤1\mathrm{dp}_A(A/Aa)\leq1. If dpA(A/Aa)<1\mathrm{dp}_A(A/Aa)<1, then A/AaA/Aa is projective, and there exists an a-linear mapping ψ:As⟶As\psi:A_s\longrightarrow A_s such that ψ∘φ=Id\psi\circ\varphi=\mathrm{Id}; this implies

1=ψ(φ(1))=ψ(a)=a.ψ(1) 1=\psi(\varphi(1))=\psi(a)=a.\psi(1)

and aa is invertible.

Proposition 2

Assume A left noetherian. Let M be a finitely generated a-module and n an integer ≥0\geq0. The following conditions are equivalent:

(i) dpA(M)≤n\mathrm{dp}_A(M)\leq n.

(i bis) M possesses a projective resolution P of length ≤n\leq n such that PiP_i is a finitely generated a-module for each ii.

(ii) Ext⁡Ar(M,N)=0\operatorname{Ext}_A^r(M,N)=0 for every finitely generated a-module N and every integer r>nr>n.

(iii) Ext⁡An+1(M,N)=0\operatorname{Ext}_A^{n+1}(M,N)=0 for every finitely generated a-module N.

(iv) Tor⁡rA(P,M)=0\operatorname{Tor}^A_r(P,M)=0 for every right a-module P and every r>nr>n.

(v) Tor⁡n+1A(A/a,M)=0\operatorname{Tor}^A_{n+1}(A/a,M)=0 for every finitely generated right ideal aa of A.

(i bis) ⇒\Rightarrow (i) : this is trivial.

(i) ⇒\Rightarrow (ii) : this follows from Lemma 1.

(ii) ⇒\Rightarrow (iii) : this is trivial.

(iii) ⇒\Rightarrow (i) : by (iii) and X, p. 107, prop. 5, we have Ext⁡An+1(M,N)=0\operatorname{Ext}_A^{n+1}(M,N)=0 for every a-module N, whence (i) by prop. 1.

(i) ⇒\Rightarrow (iv) : this follows from Lemma 1.

(iv) ⇒\Rightarrow (v) : this is trivial.

(v) ⇒\Rightarrow (i bis) : let (L,d) be a free resolution of M such that LrL_r is finitely generated for all rr (X, p. 53, prop. 6). Set K=Zn−1(L)K=\mathrm{Z}_{n-1}(L); then K is finitely generated as a submodule of Ln−1L_{n-1} and we have an exact sequence

0⟶K⟶Ln−1⟶Ln−2⟶⋯⟶L1⟶L0⟶M⟶0.(1) 0\longrightarrow K\longrightarrow L_{n-1}\longrightarrow L_{n-2}\longrightarrow\cdots\longrightarrow L_1\longrightarrow L_0\longrightarrow M\longrightarrow0. \tag{1}

By (v) and X, p. 131, cor. 3, we have Tor⁡1A(A/a,K)=0\operatorname{Tor}^A_1(A/a,K)=0 for every finitely generated right ideal aa of A. By th. 2 of X, p. 74, the a-module K is flat; since it is finitely generated, hence of finite presentation (X, p. 10, prop. 5), it is projective (X, p. 13, cor.), so (1) is a projective resolution of M.

Corollary

Suppose AA left noetherian and let C0\mathscr{C}_0 (resp. C\mathscr{C}) be the set of classes of AA-modules projective of finite type (resp. of AA-modules of finite projective dimension and of finite type). Then the homomorphism of Grothendieck groups K(C0)→K(C)K(\mathscr{C}_0) \to K(\mathscr{C}) is bijective.

This follows from X, p. 58, th. 1 (note that C0\mathscr{C}_0 and C\mathscr{C} are left exact by cor. 2).

2. The homomorphism Tor⁡nA(P,M)→Hom⁡A(Ext⁡An(M,A),P)\operatorname{Tor}_n^A(P, M) \to \operatorname{Hom}_A(\operatorname{Ext}_A^n(M, A), P)

Let MM a left AA-module, PP a right AA-module, nn an integer ⩾0\geqslant 0. The mapping kk-bilinéaire (X, p. 129)

cP;M,As:Ext⁡An(M,A)×Tor⁡nA(P,M)→P⊗AA c_{P;M,A_s} : \operatorname{Ext}_A^n(M, A) \times \operatorname{Tor}_n^A(P, M) \to P \otimes_A A

corresponds to a mapping kk-linéaire

(2)

Tor⁡nA(P,M)→Hom⁡k(Ext⁡An(M,A),P); \operatorname{Tor}_n^A(P, M) \to \operatorname{Hom}_k(\operatorname{Ext}_A^n(M, A), P);

moreover, if one equips Ext⁡An(M,A)\operatorname{Ext}_A^n(M, A) with the structure of right AA-module coming from the bimodule structure of AA, the image of (2) consists of mappings AA-linéaires, as is immediately verified; this yields a kk-homomorphisme, called canonical

(3)

Tor⁡nA(P,M)→Hom⁡A(Ext⁡An(M,A),P). \operatorname{Tor}_n^A(P, M) \to \operatorname{Hom}_A(\operatorname{Ext}_A^n(M, A), P).

Proposition 3

a) If dpA(M)⩽n\mathrm{dp}_A(M) \leqslant n, the canonical homomorphism (3) is injective. b) If dpA(M)⩽n\mathrm{dp}_A(M) \leqslant n, if AA is left noetherian and if MM is of finite type, the canonical homomorphism (3) is bijective.

We argue by induction on nn. If n=0n = 0, MM is projective, the homomorphism (3) reduces to the canonical homomorphism P⊗AM→Hom⁡A(M∗,P)P \otimes_A M \to \operatorname{Hom}_A(M^*, P) of II, p. 77, and the proposition follows from loc. cit., corollary. If n>0n > 0, let

(4)

0→N→L→M→0 0 \to N \to L \to M \to 0

be an exact sequence of AA-modules where LL is free (and of finite type in case b); then dpA(N)⩽n−1\mathrm{dp}_A(N) \leqslant n - 1 (X, p. 135, cor. 2, c)) and NN is of finite type in case b.

Let θ∈Ext⁡A1(M,N)\theta \in \operatorname{Ext}_A^1(M, N) be the class associated with the exact sequence (4) (X, p. 117, déf. 1). Denote

un:Ext⁡An−1(N,A)→Ext⁡An(M,A)vn:Tor⁡nA(P,M)→Tor⁡n−1A(P,N) u_n : \operatorname{Ext}_A^{n-1}(N, A) \to \operatorname{Ext}_A^n(M, A) \qquad v_n : \operatorname{Tor}_n^A(P, M) \to \operatorname{Tor}_{n-1}^A(P, N)
the mappings defined by un(α)=α∘θu_n(\alpha) = \alpha \circ \theta and vn(β)=θ∘βv_n(\beta) = \theta \circ \beta. We have
(α∘θ)∘β=α∘(θ∘β) (\alpha \circ \theta) \circ \beta = \alpha \circ (\theta \circ \beta)

for all α∈Ext⁡An−1(N,A)\alpha\in\operatorname{Ext}_A^{n-1}(N,A), β∈Tor⁡nA(P,M)\beta\in\operatorname{Tor}_n^A(P,M) (X, p. 129, prop. 6), so that the diagram

Tor⁡nA(P,M)⟶Hom⁡A(Ext⁡An(M,A),P)↓vn↓Hom⁡(un,1)Tor⁡n−1A(P,N)⟶Hom⁡A(Ext⁡An−1(N,A),P),(5) \begin{array}{ccc} \operatorname{Tor}_n^A(P,M)&\longrightarrow&\operatorname{Hom}_A(\operatorname{Ext}_A^n(M,A),P)\\ \downarrow v_n&&\downarrow\operatorname{Hom}(u_n,1)\\ \operatorname{Tor}_{n-1}^A(P,N)&\longrightarrow&\operatorname{Hom}_A(\operatorname{Ext}_A^{n-1}(N,A),P), \end{array} \tag{5}

where the horizontal arrows are the canonical homomorphisms, is commutative.

If n=1n=1, we thus have a commutative diagram:

00↓↓Tor⁡1A(P,M)⟶Hom⁡A(Ext⁡A1(M,A),P)↓v1↓Hom⁡(u1,1)P⊗AN⟶Hom⁡A(N∗,P)↓↓P⊗AL⟶Hom⁡A(L∗,P) \begin{array}{ccccc} 0&&&&0\\ \downarrow&&&&\downarrow\\ \operatorname{Tor}_1^A(P,M)&\longrightarrow&\operatorname{Hom}_A(\operatorname{Ext}_A^1(M,A),P)\\ \downarrow v_1&&&&\downarrow\operatorname{Hom}(u_1,1)\\ P\otimes_A N&\longrightarrow&\operatorname{Hom}_A(N^*,P)\\ \downarrow&&&&\downarrow\\ P\otimes_A L&\longrightarrow&\operatorname{Hom}_A(L^*,P) \end{array}

where the columns are exact; this yields the result in this case. If n≥2n\geq 2, the mappings unu_n and vnv_n are bijective (X, p. 128, cor. 4 and p. 131, cor. 3). By the induction hypothesis, the canonical homomorphism

Tor⁡n−1A(P,N)⟶Hom⁡A(Ext⁡An−1(N,A),P) \operatorname{Tor}_{n-1}^A(P,N)\longrightarrow\operatorname{Hom}_A(\operatorname{Ext}_A^{n-1}(N,A),P)

is injective (resp. bijective); the diagram (5) shows that the same holds for the canonical homomorphism Tor⁡nA(P,M)⟶Hom⁡A(Ext⁡An(M,A),P)\operatorname{Tor}_n^A(P,M)\longrightarrow\operatorname{Hom}_A(\operatorname{Ext}_A^n(M,A),P), which completes the proof.

3. Homological dimension of a ring

Definition 2

We call the homological dimension of AA and denote by dh⁡(A)\operatorname{dh}(A) the supremum in Z\mathbf{Z} of the set of integers nn for which there exist two AA-modules M and N such that Ext⁡An(M,N)≠0\operatorname{Ext}_A^n(M,N)\neq 0.

We have dh⁡(0)=−∞\operatorname{dh}(0)=-\infty, dh⁡(A)≥0\operatorname{dh}(A)\geq 0 if A≠0A\neq 0. We shall see below that dh⁡(A)=1\operatorname{dh}(A)=1 if AA is principal and is not a field, and that, if K is a commutative field

dh⁡(K[X1,…,Xn])=n. \operatorname{dh}(K[X_1,\ldots,X_n])=n.

Proposition 4

Let nn be an integer ≥0\geq 0. The following conditions are equivalent:

(i) dh⁡(A)≤n\operatorname{dh}(A)\leq n,

(ii) for every AA-module M, we have dpA(M)≤n\mathrm{dp}_A(M)\leq n,

(ii′) for every finitely generated AA-module M, we have dpA(M)≤n\mathrm{dp}_A(M)\leq n,

(iii) for every exact sequence

0→K→Pn−1→Pn−2→…→P0 0 \to K \to P_{n-1} \to P_{n-2} \to \ldots \to P_0

where the PiP_i are projective, KK is projective,

(iv) for every exact sequence

I0→I1→…→In−1→N→0 I^0 \to I^1 \to \ldots \to I^{n-1} \to N \to 0

where the IiI^i are injective, NN is injective,

(v) every AA-module possesses an injective resolution of length ≤n\leq n.

The equivalence of conditions (i), (ii) and (iii) follows from prop. 1. We have obviously (ii) ⇒\Rightarrow (ii’). It is therefore enough to proving (ii’) ⇒\Rightarrow (iv) ⇒\Rightarrow (v) ⇒\Rightarrow (i).

(ii’) ⇒\Rightarrow (iv): with the notation of (iv), let KK be the kernel of I0→I1I^0 \to I^1. By X, p. 128, cor. 4, we have for every AA-module MM an isomorphism ExtA1(M,N)→ExtAn+1(M,K)\mathrm{Ext}_A^1(M, N) \to \mathrm{Ext}_A^{n+1}(M, K); it then follows from (ii’) that ExtA1(M,N)\mathrm{Ext}_A^1(M, N) is zero for every finitely generated AA-module MM. By X, p. 93, prop. 11, this implies that NN is injective, whence (iv).

(iv) ⇒\Rightarrow (v): let MM be an AA-module. Applying (iv) to the exact sequence

0→M→I0(M)→I1(M)→…→In−1(M)→Kn−1(M)→0 0 \to M \to I^0(M) \to I^1(M) \to \ldots \to I^{n-1}(M) \to K^{n-1}(M) \to 0

of X, p. 52, we conclude that Kn−1(M)K^{n-1}(M) is injective, whence (v).

(v) ⇒\Rightarrow (i): this follows from X, p. 100, th. 1.

Remarque 1

If dh(A)≤n<+∞\mathrm{dh}(A) \leq n < +\infty, we have Torn+1A(P,M)=0\mathrm{Tor}_{n+1}^A(P, M) = 0 for every AA-module MM and every right AA-module PP, since dpA(M)≤n\mathrm{dp}_A(M) \leq n (cf. lemma 1).

Remarque 2

For dh(A)\mathrm{dh}(A) to be finite, it is necessary and sufficient that dpA(M)\mathrm{dp}_A(M) be finite for every nonzero AA-module MM. This indeed follows from the preceding and from X, p. 135, cor. 1.

Corollaire

Suppose AA is left noetherian and let nn be an integer >0> 0. The following conditions are equivalent:

(i) dh(A)≤n\mathrm{dh}(A) \leq n,

(ii) for every pair of a-modules MM and NN of finite type, we have ExtAn+1(M,N)=0\mathrm{Ext}_A^{n+1}(M, N) = 0,

(iii) for every left a-module MM of finite type and every right a-module PP of finite type, we have Torn+1A(P,M)=0\mathrm{Tor}_{n+1}^A(P, M) = 0.

This follows from Props. 2 and 4.

Remark

By the equivalence of (i) and (iii), we have dh(A)=dh(A∘)\mathrm{dh}(A) = \mathrm{dh}(A^\circ) if AA is left and right noetherian. This equality is not valid in general (X, p. 204, Exercise 20).

Proposition 5

Suppose AA is left noetherian and of finite homological dimension and let C0\mathscr{C}_0 (resp. C\mathscr{C}) be the set of equivalence classes of finitely generated projective a-modules (resp. of a-modules). Then the canonical homomorphism of Grothendieck groups K(C0)→K(C)K(\mathscr{C}_0) \to K(\mathscr{C}) is bijective.

This follows from X, p. 137, Cor.

4. Rings of Homological Dimension 0

Proposition 6

The following conditions are equivalent : (i) every a-module is projective, (ii) every a-module is injective, (iii) every ideal of AA is an injective module, (iv) dh(A)≤0\mathrm{dh}(A) \leq 0, (v) AA is semisimple, (vi) AA is noetherian and every a-module is flat; (vii) every complex of a-modules is split, (viii) every exact sequence of a-modules is split.

By Prop. 4, we have (i) ⇔ (ii) ⇔ (iv); by Corollary 1 to Prop. 4, we have (vi) ⇒ (iv). By X, p. 35, Example 4, we have (i) ⇒ (vii); since (ii) ⇒ (iii) and (vii) ⇒ (viii) are trivial and since (viii) ⇒ (i) follows from II, p. 39, Prop. 4, it remains to be shown (iii) ⇒ (v) and (v) ⇒ (vi); the last assertion follows from VIII, § 5, No. 1, Props. 1 and 2; finally, if every ideal of AA is injective, it is a direct factor in AA, hence (iii) ⇒ (v).

5. Rings of Homological Dimension 1

Proposition 7

The following conditions are equivalent : (i) dh(A)≤1\mathrm{dh}(A) \leq 1, (ii) every submodule of a projective module is projective, (ii') every ideal of AA is projective, (iii) every quotient of an injective a-module is injective, (iv) for every projective complex CC, there exists a homomorphism φ:C→H(C)\varphi : C \to H(C) such that H(φ)=1H(C)H(\varphi) = 1_{H(C)}, (v) for every injective complex CC, there exists a homomorphism ψ:H(C)→C\psi : H(C) \to C such that H(ψ)=1H(C)H(\psi) = 1_{H(C)}.

(i) ⇔ (ii) ⇔ (iii) : this follows from Prop. 4 of X, p. 138. (ii) ⇒ (iv) : let CC be a projective complex. If (ii) is satisfied, the submodule B(C)B(C) of CC is projective, hence (iv) by X, p. 35, Remark b). (iii) ⇒ (v) : let CC be an injective complex. If (iii) is satisfied, the quotient Bn(C)B_n(C) of Cn+1C_{n+1} is injective for all nn, hence (v) by X, p. 35, Remark a). (iv) ⇒ (ii) : let PP be a projective a-module, MM a submodule of PP, and i:M→Pi : M \to P the canonical injection. Let p:L→Mp : L \to M be a surjective homomorphism from a free module LL onto MM. Consider the projective complex CC such that C1=LC_1 = L, C0=PC_0 = P, Ci=0C_i = 0 for i≠0,1i \ne 0, 1, and d1=i∘pd_1 = i \circ p. If (iv) is satisfied, let φ:C→H(C)\varphi : C \to H(C) be a homomorphism such that H(φ)=1H(C)H(\varphi) = 1_{H(C)}. Since H1(C)=Ker pH_1(C) = \mathrm{Ker}\ p, φ1\varphi_1 is a projector of LL onto Ker p\mathrm{Ker}\ p, so the exact sequence

0→Ker p→L→pM→0 0 \to \mathrm{Ker}\ p \to L \xrightarrow{p} M \to 0

is split and MM is isomorphic to a direct factor of LL, hence projective.

(v)⇒(iii)(v)\Rightarrow(iii) : let II be an injective module, MM a quotient of II, and π:I→M\pi : I \to M the canonical projection. Let i:M→Ji : M \to J be an injective homomorphism of MM into an injective module JJ. Consider the injective complex CC such that C0=IC^0 = I, C1=JC^1 = J, Ci=0C^i = 0 for i≠0,1i \ne 0, 1, and d0=i∘πd^0 = i \circ \pi. If (v) is satisfied, let ψ:H(C)→C\psi : H(C) \to C be a homomorphism such that H(ψ)=1H(C)H(\psi) = 1_{H(C)}. Since

H1(C)=Coker⁡i, H^1(C) = \operatorname{Coker}i,

ψ1\psi^1 is a section of the canonical projection J→Coker⁡iJ \to \operatorname{Coker}i, and MM is a direct factor in JJ, hence injective.

(ii)⇒(ii′)(ii)\Rightarrow(ii') : this is trivial.

(ii′)⇒(ii)(ii')\Rightarrow(ii) : this follows from VII, § 3, Corollary 1 to Theorem 1.

Example

If AA is principal, then dh⁡(A)≤1\operatorname{dh}(A) \leq 1.

Remark

If AA is (commutative) a domain, the preceding conditions are also equivalent to the following :

(iii') : every divisible a-module is injective,

(vi) : every torsion-free a-module is flat and AA is noetherian.

Domains satisfying these conditions are called Dedekind rings (cf. X, p. 204, Exercise 12 and AC, VII, § 2, No. 2, Th. 1).

Corollary

Let AA be a ring of homological dimension ≤1\leq 1, CC a complex of projective a-modules, C~\widetilde{C} a complex of projective right a-modules, C′C' a complex of injective a-modules, PP a right a-module, MM a left a-module, and nn an integer. Then there are split exact sequences

0→⨁p+q=nHp(C~)⊗AHq(C)→γHn(C~⊗AC)→α⨁p+q=n−1Tor⁡1A(Hp(C~),Hq(C))→0, 0\to\bigoplus_{p+q=n}H_p(\widetilde C)\otimes_A H_q(C)\xrightarrow{\gamma}H_n(\widetilde C\otimes_A C)\xrightarrow{\alpha}\bigoplus_{p+q=n-1}\operatorname{Tor}_1^A(H_p(\widetilde C),H_q(C))\to0,
0→∏pExt⁡A1(Hp(C),Hn−p−1(C′))→βHn(Homgr⁡A(C,C′))→λ∏pHomgr⁡A(Hp(C),Hn−p(C′))→0, 0\to\prod_p\operatorname{Ext}_A^1(H_p(C),H^{n-p-1}(C'))\xrightarrow{\beta}H^n(\operatorname{Homgr}_A(C,C'))\xrightarrow{\lambda}\prod_p\operatorname{Homgr}_A(H_p(C),H^{n-p}(C'))\to0,
0→P⊗AHn(C)→γHn(P⊗AC)→αTor⁡1A(P,Hn−1(C))→0, 0\to P\otimes_A H_n(C)\xrightarrow{\gamma}H_n(P\otimes_A C)\xrightarrow{\alpha}\operatorname{Tor}_1^A(P,H_{n-1}(C))\to0,
0→Ext⁡A1(Hn−1(C),M)→βHn(Homgr⁡A(C,M))→λHom⁡A(Hn(C),M)→0. 0\to\operatorname{Ext}_A^1(H_{n-1}(C),M)\xrightarrow{\beta}H^n(\operatorname{Homgr}_A(C,M))\xrightarrow{\lambda}\operatorname{Hom}_A(H_n(C),M)\to0.

Since Z(C)Z(C), B(C)B(C), Z(C~)Z(\widetilde C), and B(C~)B(\widetilde C) are projective and B(C′)B(C') is injective, this follows from X, p. 78, Cor. 2, p. 96, Cor. 1, and p. 98, Cor. 2.

6. Homological Dimension of Polynomial Rings

Lemma 2

Let ρ:A→A′\rho : A\to A' be a ring homomorphism, M an a-module, M′ an a′-module. If the a-module Ad′A'_d is flat, one has dpA(A′⊗AM)≤dpA(M)\mathrm{dp}_A(A'\otimes_A M)\leq\mathrm{dp}_A(M). If the a-module As′A'_s is projective, one has dpA(M′)≤dpA′(M′)\mathrm{dp}_A(M')\leq\mathrm{dp}_{A'}(M').

The first assertion is clear if dpA(M)=±∞\mathrm{dp}_A(M) = \pm \infty; if dpA(M)=n∈N\mathrm{dp}_A(M) = n \in \mathbf{N}, there exists an exact sequence of AA-modules

0→Pn→Pn−1→…→P0→M→0 0 \to P_n \to P_{n-1} \to \ldots \to P_0 \to M \to 0

where the PiP_i are projective ; the sequence of A′A'-modules

0→A′⊗APn→A′⊗APn−1→…→A′⊗AP0→A′⊗AM→0 0 \to A' \otimes_A P_n \to A' \otimes_A P_{n-1} \to \ldots \to A' \otimes_A P_0 \to A' \otimes_A M \to 0

is exact, since Ad′A'_d is flat, and the A′A'-modules A′⊗APiA' \otimes_A P_i are projective (II, p. 89, Corollary) ; therefore dpA′(A′⊗AM)≤n=dpA(M)\mathrm{dp}_{A'}(A' \otimes_A M) \leq n = \mathrm{dp}_A(M). The second assertion is clear if dpA′(M′)=±∞\mathrm{dp}_{A'}(M') = \pm \infty; if dpA′(M′)=m∈N\mathrm{dp}_{A'}(M') = m \in \mathbf{N}, there exists an exact sequence of A′A'-modules

0→Pm′→Pm−1′→…→P0′→M′→0, 0 \to P'_m \to P'_{m-1} \to \ldots \to P'_0 \to M' \to 0 ,

where the Pi′P'_i are projective ; the underlying sequence of AA-modules is exact. Moreover each Pi′P'_i is a sub-A′A'-module direct factor of a module As′(1){A'_s}^{(1)}, hence is a projective AA-module ; therefore one has dpA(M′)≤m=dpA′(M′)\mathrm{dp}_A(M') \leq m = \mathrm{dp}_{A'}(M').

Lemma 3

Assume A commutative and let M be an a[X]-module.* a) One has dpA(M)≤dpA[X](M)≤dpA(M)+1\mathrm{dp}_A(M) \leq \mathrm{dp}_{A[X]}(M) \leq \mathrm{dp}_A(M) + 1. b) If the homothety XMX_M is injective, one has dpA(M/XM)≤dpA[X](M)\mathrm{dp}_A(M/XM) \leq \mathrm{dp}_{A[X]}(M). c) If XM=0XM = 0, one has dpA(M)+1=dpA[X](M)\mathrm{dp}_A(M) + 1 = \mathrm{dp}_{A[X]}(M). a) One has an exact sequence of A[X]A[X]-modules (III, p. 106 and VII, § 5, No. 1)

0→A[X]⊗AM→A[X]⊗AM→M→0; 0 \to A[X] \otimes_A M \to A[X] \otimes_A M \to M \to 0 ;

The assertion a) follows from X, p. 135, Corollary 2 and Lemma 2.

b) If dpA[X](M)=±∞\mathrm{dp}_{A[X]}(M) = \pm \infty, the assertion is trivial. If MM is projective and nonzero, then the AA-module M/XMM/XM is identified with A⊗A[X]MA \otimes_{A[X]} M, hence is projective, and one has dpA(M/XM)≤0=dpA[X](M)\mathrm{dp}_A(M/XM) \leq 0 = \mathrm{dp}_{A[X]}(M). Let us argue by induction on dpA[X](M)=n\mathrm{dp}_{A[X]}(M) = n, assumed >0> 0. Consider an exact sequence of A[X]A[X]-modules 0→N→L→M→00 \to N \to L \to M \to 0, where LL is a free A[X]A[X]-module ; applying to the diagram

0⟶N⟶L⟶M⟶0↓↓↓0⟶N⟶L⟶M⟶0 \begin{array}{ccc} 0 & \longrightarrow & N \longrightarrow L \longrightarrow M \longrightarrow 0 \\ & & \downarrow \quad \downarrow \quad \downarrow \\ 0 & \longrightarrow & N \longrightarrow L \longrightarrow M \longrightarrow 0 \end{array}

Proposition 2 of X, p. 4, one sees that XNX_N is injective and that one has the exact sequence

0→N/XN→L/XL→M/XM→0. 0 \to N/XN \to L/XL \to M/XM \to 0 .

Since LL is free over A[X]A[X], L/XLL/XL is free over AA and one has

dpA[X](N)=n−1,dpA(M/XM)≤1+dpA(N/XN); \mathrm{dp}_{A[X]}(N) = n - 1 , \quad \mathrm{dp}_A(M/XM) \leq 1 + \mathrm{dp}_A(N/XN) ;

since dpA(N/XN)≤n−1\mathrm{dp}_A(N/XN) \leq n - 1 by the induction hypothesis, one deduces dpA(M/XM)≤n\mathrm{dp}_A(M/XM) \leq n, which was to be proved.

c) The assertion is trivial if dpA[X](M)=±∞\mathrm{dp}_{A[X]}(M) = \pm \infty, and also if dpA[X](M)=0\mathrm{dp}_{A[X]}(M) = 0 (which is impossible since XM=0XM = 0). We may therefore suppose dpA[X](M)=n>0\mathrm{dp}_{A[X]}(M) = n > 0. Considering as above an exact sequence 0→N→L→M→00 \to N \to L \to M \to 0, where LL is a free A[X]A[X]-module, we obtain an exact sequence of AA-modules

0→M→N/XN→L/XL→M→0. 0 \to M \to N/XN \to L/XL \to M \to 0 .

By b), we have dpA(N/XN)≤dpA[X](N)=dpA[X](M)−1=n−1\mathrm{dp}_A(N/XN) \leq \mathrm{dp}_{A[X]}(N) = \mathrm{dp}_{A[X]}(M) - 1 = n - 1; since dpA(L/XL)=0\mathrm{dp}_A(L/XL) = 0, we deduce from the preceding exact sequence, applying XX twice, p. 135, cor. 2, that dpA(M)≤n−1\mathrm{dp}_A(M) \leq n - 1. But, by a), we have

dpA(M)≥dpA[X](M)−1=n−1, \mathrm{dp}_A(M) \geq \mathrm{dp}_{A[X]}(M) - 1 = n - 1 ,

whence c).

Theorem 1

Suppose AA commutative. Then

dh(A[X])=dh(A)+1. dh(A[X]) = dh(A) + 1 .

For every A[X]A[X]-module MM, we have (lemma 3)

dpA[X](M)≤dpA(M)+1≤dh(A)+1 \mathrm{dp}_{A[X]}(M) \leq \mathrm{dp}_A(M) + 1 \leq dh(A) + 1

therefore dh(A[X])≤dh(A)+1dh(A[X]) \leq dh(A) + 1; conversely, if MM is an AA-module, let M‾\overline{M} be the A[X]A[X]-module obtained by equipping MM with the structure for which XM=0XM = 0, then (lemma 3)

dpA(M)=dpA[X](M‾)−1≤dh(A[X])−1, \mathrm{dp}_A(M) = \mathrm{dp}_{A[X]}(\overline{M}) - 1 \leq dh(A[X]) - 1 ,

therefore dh(A)≤dh(A[X])−1dh(A) \leq dh(A[X]) - 1.

Corollary 1

Suppose AA commutative. We have

dh(A[X1,...,Xn])=dh(A)+n. dh(A[X_1, ..., X_n]) = dh(A) + n .

This follows from the theorem by induction on nn.

Corollary 2

Let KK be a commutative field* (resp. a principal ideal domain or a Dedekind ring which is not a field). Then dh(K[X1,...,Xn])dh(K[X_1, ..., X_n]) is equal to nn (resp. n+1n + 1). This follows from the fact that dh(K)=0dh(K) = 0 (resp. dh(K)=1dh(K) = 1).

7. Homological dimension of graded modules

In this section, we suppose that AA is a graded ring with degrees ≥0\geq 0. We denote by (An)n∈Z(A_n)_{n \in \mathbf{Z}} its grading; we have therefore An=0A_n = 0 for n<0n < 0, A0A_0 is a subring of AA, J0=⨁n>0AnJ_0 = \bigoplus_{n > 0} A_n is a two-sided ideal of AA and the graded quotient ring A/J0A/J_0 is identified with A0A_0.

Lemma 4

Let MM be a graded AA-module bounded below (X, p. 56). If A0⊗AM=0A_0 \otimes_A M = 0, then M=0M = 0.

Since A0⊗AMA_0 \otimes_A M is isomorphic to M/J0MM/J_0 M, this is none other than ll, p. 171, prop. 6.

Lemma 5

Let MM be a graded AA-module bounded below, and let

s:A⊗A0M/J0M→M s : A \otimes_{A_0} M/J_0 M \to M
be a graded AA-homomorphism such that 1⊗As:A0⊗A(A⊗A0M/J0M)→A0⊗AM1 \otimes_A s : A_0 \otimes_A (A \otimes_{A_0} M/J_0 M) \to A_0 \otimes_A M is the canonical isomorphism. Then ss is surjective. If Tor1A(A0,M)=0\mathrm{Tor}_1^A(A_0, M) = 0, ss is bijective.

We have an exact sequence

0→Ker s→A⊗A0MJ0M→M→Coker s→0 0 \to \mathrm{Ker}\ s \to A \otimes_{A_0} M J_0 M \to M \to \mathrm{Coker}\ s \to 0
and the graded AA-modules Ker s\mathrm{Ker}\ s and Coker s\mathrm{Coker}\ s are bounded below. From the exact sequence A0⊗A(A⊗A0M/J0M)→1⊗sA0⊗AM→A0⊗ACoker s→0A_0 \otimes_A (A \otimes_{A_0} M/J_0 M) \xrightarrow{1 \otimes s} A_0 \otimes_A M \to A_0 \otimes_A \mathrm{Coker}\ s \to 0, we deduce that A0⊗ACoker s=0A_0 \otimes_A \mathrm{Coker}\ s = 0, hence ss is surjective (Lemma 4). We then have an exact sequence
Tor1A(A0,M)→A0⊗AKer s→A0⊗A(A⊗A0M/J0M)→1⊗sA0⊗AM. \mathrm{Tor}_1^A(A_0, M) \to A_0 \otimes_A \mathrm{Ker}\ s \to A_0 \otimes_A (A \otimes_{A_0} M/J_0 M) \xrightarrow{1 \otimes s} A_0 \otimes_A M .
If Tor1A(A0,M)=0\mathrm{Tor}_1^A(A_0, M) = 0, then A0⊗AKer s=0A_0 \otimes_A \mathrm{Ker}\ s = 0 and ss is injective (Lemma 4).

Proposition 8

Let MM be a graded AA-module bounded below.

a) The following conditions are equivalent : (i) MM is isomorphic to a graded AA-module of the form A⊗A0NA \otimes_{A_0} N, where NN is a graded projective A0A_0-module (resp. graded free) ; (ii) MM is a projective AA-module (resp. graded free) ; (iii) M/J0MM/J_0 M is a projective A0A_0-module (resp. graded free) and Tor1A(A0,M)=0\mathrm{Tor}_1^A(A_0, M) = 0.

b) Assume furthermore that MM has a generator system consisting of homogeneous elements of bounded degrees. Then the following conditions are equivalent : (i) The graded AA-module MM has a finite composition series whose quotients are isomorphic to graded AA-modules of the form A⊗A0NA \otimes_{A_0} N, where NN is a graded A0A_0-module that is flat ; (ii) MM is a flat AA-module ; (iii) M/J0MM/J_0 M is a flat A0A_0-module and Tor1A(A0,M)=0\mathrm{Tor}_1^A(A_0, M) = 0.

In each case, there is obviously (i) ⇒ (ii) ⇒ (iii). Therefore, it remains to prove (iii) ⇒ (i).

a) The graded AA-module A⊗A0M/J0MA \otimes_{A_0} M/J_0 M is a projective AA-module since M/J0MM/J_0 M is a projective A0A_0-module. The canonical homomorphism of AA-modules

p:M→M/J0M p : M \to M/J_0 M
is surjective ; hence there exists a graded AA-homomorphism of degree zero
s:A⊗A0M/J0M→M s : A \otimes_{A_0} M/J_0 M \to M
such that p∘s(a⊗x)=axp \circ s(a \otimes x) = ax for a∈Aa \in A and x∈M/J0Mx \in M/J_0 M.

By Lemma 5, ss is an isomorphism of AA-modules from A⊗A0M/J0MA\otimes_{A_0}M/J_0M onto MM, whence (i).

b) By the hypothesis on MM, there exist integers a,ba,b with a≤ba\leq b such that MM is generated by ⨁a≤i≤bMi\displaystyle\bigoplus_{a\leq i\leq b}M_i. We argue by induction on the positive integer b−ab-a.

If b−a=0b-a=0, then MM is generated by MaM_a and the canonical A0A_0-homomorphism

Ma⟶M/J0M M_a\longrightarrow M/J_0M
is bijective; we then deduce from the AA-homomorphism A⊗A0Ma⟶MA\otimes_{A_0}M_a\longrightarrow M defined by the AA-module structure of MM a graded AA-homomorphism
s:A⊗A0M/J0M⟶M s:A\otimes_{A_0}M/J_0M\longrightarrow M
satisfying the condition of Lemma 5. Then, by Lemma 5, ss is bijective, whence (i).

In the general case, let M(a)M^{(a)} be the (graded) sub-AA-module of MM generated by MaM_a and M′M' the quotient M/M(a)M/M^{(a)}. We have an exact sequence

0⟶M(a)→fM→gM′⟶0, 0\longrightarrow M^{(a)}\xrightarrow{f}M\xrightarrow{g}M'\longrightarrow0,
whence, since Tor⁡1A(A0,M)=0\operatorname{Tor}_1^A(A_0,M)=0 by hypothesis, exact sequences
Tor⁡2A(A0,M′)⟶Tor⁡1A(A0,M(a))⟶0(6) \tag{6} \operatorname{Tor}_2^A(A_0,M')\longrightarrow\operatorname{Tor}_1^A(A_0,M^{(a)})\longrightarrow0
0⟶Tor⁡1A(A0,M′)⟶M(a)/J0M(a)→1⊗fM/J0M→1⊗gM′/J0M′⟶0.(7) \tag{7} 0\longrightarrow\operatorname{Tor}_1^A(A_0,M')\longrightarrow M^{(a)}/J_0M^{(a)} \xrightarrow{1\otimes f}M/J_0M\xrightarrow{1\otimes g}M'/J_0M'\longrightarrow0.

But the canonical homomorphism Ma⟶M(a)/J0M(a)M_a\longrightarrow M^{(a)}/J_0M^{(a)} is bijective. It follows that the homomorphism 1⊗f:M(a)/J0M(a)⟶M/J0M1\otimes f:M^{(a)}/J_0M^{(a)}\longrightarrow M/J_0M is injective and that its image is a direct factor sub-A0A_0-module of M/J0MM/J_0M. It then follows from the exact sequence (7) that Tor⁡1A(A0,M′)=0\operatorname{Tor}_1^A(A_0,M')=0 and that the A0A_0-module M′/J0M′M'/J_0M' is flat since isomorphic to a direct factor of M/J0MM/J_0M. By the induction hypothesis (which applies to M′M' since it is generated by the Mi′M'_i for a<i≤ba<i\leq b), M′M' satisfies condition (i), hence is flat. We then deduce from the exact sequence (6) that Tor⁡1A(A0,M(a))=0\operatorname{Tor}_1^A(A_0,M^{(a)})=0, but M(a)/J0M(a)M^{(a)}/J_0M^{(a)} is identified with MaM_a which is a flat A0A_0-module (as a direct factor submodule of M/J0MM/J_0M) ; by what has already been proved, the graded AA-module M(a)M^{(a)} is isomorphic to A⊗A0MaA\otimes_{A_0}M_a, hence also satisfies (i), which completes the proof.

Corollary 1

Let MM be a graded AA-module of finite type. If the A0A_0-module M/J0MM/J_0M is projective (resp. graded free, resp. flat) and if Tor⁡1A(A0,M)=0\operatorname{Tor}_1^A(A_0,M)=0, then the AA-module MM is projective (resp. graded free, resp. flat).

Corollary 2

Corollary 3 (Hilbert’s Syzygy Theorem)

Assume that A0A_0 is a commutative field and that AA is generated as an A0A_0-algebra by nn homogeneous elements of degrees > 0 and algebraically independent. For every graded aa-module MM bounded below (resp. of finite type), there exists an exact sequence of graded aa-modules and of degree 00 homogeneous maps*

0→Ln→Ln−1→…→L0→M→0, 0 \to L_n \to L_{n-1} \to \ldots \to L_0 \to M \to 0,
where the LiL_i are graded free and bounded below (resp. graded free and of finite type).

Indeed, dh(A)=ndh(A) = n by Theorem 1 of X, p. 143, and Corollary 2 is applied.

Remark

Corollary 2 also applies in the following cases: a) A0A_0 is principal and A=A0[X1,...,Xn−1]A = A_0[X_1, ..., X_{n-1}]; b) A0A_0 is a regular local noetherian ring of dimension rr and A=A0[X1,...,Xn,r]A = A_0[X_1, ..., X_n, r].

Corollary 4

Assume that A0A_0 is semisimple. Let MM be a graded aa-module bounded below and let nn be an integer ≥0\geq 0. In order that dpA(M)≤n\mathrm{dp}_A(M) \leq n, it is necessary and sufficient that*

Tor⁡n+1A(A0,M)=0. \operatorname{Tor}_{n+1}^A(A_0, M) = 0.
If dpA(M)≤n\mathrm{dp}_A(M) \leq n, then Tor⁡n+1A(A0,M)=0\operatorname{Tor}_{n+1}^A(A_0, M) = 0 (X, p. 135, Lemma 1). Conversely, let ⋯→L1→L0→M→0\dots \to L_1 \to L_0 \to M \to 0 be an exact sequence of graded aa-modules bounded below such that L0,…,Ln−1L_0, \dots, L_{n-1} are graded free (X, p. 56, Prop. 11); according to Cor. 3 of X, p. 131, the equality Tor⁡n+1A(A0,M)=0\operatorname{Tor}_{n+1}^A(A_0, M) = 0 implies Tor⁡1A(A0,K)=0\operatorname{Tor}_1^A(A_0, K) = 0; since K/J0KK/J_0 K is a projective A0A_0-module because A0A_0 is semisimple (X, p. 140, Prop. 6), KK is projective by Prop. 8 (X, p. 144), and dpA(M)≤n\mathrm{dp}_A(M) \leq n.

Corollary 5

Suppose the ring A0A_0 semisimple. If Tor⁡n+1A(A0,A0)=0\operatorname{Tor}_{n+1}^A(A_0, A_0) = 0, one has dpA(M)≤n\mathrm{dp}_A(M) \leq n for every graded AA-module bounded below.

Denote by A∘A^\circ the graded ring opposite to AA: we have (A∘)0=(A0)∘(A^\circ)_0 = (A_0)^\circ, so (A∘)0(A^\circ)_0 is semisimple (VIII, § 5, No. 1, remark 3). Since Tor⁡n+1A∘(A0∘,A0∘)=0\operatorname{Tor}_{n+1}^{A^\circ}(A_0^\circ, A_0^\circ) = 0, one has dpA(A0s∘)≤n\mathrm{dp}_A(A_{0s}^\circ) \leq n by cor. 4: this implies Tor⁡n+1A∘(M0,A0∘)=0\operatorname{Tor}_{n+1}^{A^\circ}(M_0, A_0^\circ) = 0 for every AA-module MM, hence Tor⁡n+1A(A0,M)=0\operatorname{Tor}_{n+1}^A(A_0, M) = 0 and we apply cor. 4.

Exercises

See the exercises for § 8.

Machine translation of the English, by laguna-s-2.1-free, hy3-free, against glossary version 34. Not checked by a person.