Bourbaki

5. Finitely generated projective modules. Invertible fractional ideals

1. Localization with respect to an element

Let AA be a ring and MM an AA-module. For every element f∈Af \in A, we shall write Af=A[f−1],Mf=M[f−1]=M⊗AA[f−1]A_f = A[f^{-1}], M_f = M[f^{-1}] = M \otimes_A A[f^{-1}] (§ 2, nos. 1 and 2); if SfS_f is the set off" for n≥0n \geq 0, then ASf=Sf−1A,MSf=Sf−1MA_{S_f} = S_f^{-1}A, M_{S_f} = S_f^{-1}M. Iff is invertible in AA, AfA_f (resp. MfM_f) is canonically identified with AA (resp. MM); iff is nilpotent, then Af=0A_f = 0 and Mf=0M_f = 0. For every AA-module homomorphism u:M→Nu : M \to N, we write uf=u⊗1:Mf→Nfu_f = u \otimes 1 : M_f \to N_f.

Let gg be another element of AA; AgA_g (resp. MgM_g) is canonically identified with (Af)g/1(A_f)_{g/1} (resp. (Mf)g/1(M_f)_{g/1}), where g/1g/1 is the image of gg in AgA_g, and uf,gu_{f,g} with (uf)g/1(u_f)_{g/1} (§ 2, no. 3, Proposition 7).

Proposition 1

Let ff be an element of a ring AA and ϕ:A→Af\phi : A \to A_f, the canonical mapping. The mapping aϕ:Spec⁡(Af)→Spec⁡(A)^a\phi : \operatorname{Spec}(A_f) \to \operatorname{Spec}(A) is a homeomorphism of Spec⁡(Af)\operatorname{Spec}(A_f) onto the open subspace XfX_f of X=Spec⁡(A)X = \operatorname{Spec}(A) (§ 4, no. 3).

This a particular case of § 4, no. 3, Corollary to Proposition 13.

Proposition 2

Let AA be a ring, u:M→Nu : M \to N an AA-module homomorphism and pp a prime ideal of AA.

(i) Suppose that up:Mp→Npu_p : M_p \to N_p is surjective and that NN is finitely generated. Then there exists f∈A−pf \in A - p such that uf:M→Nu_f : M \to N, is surjective.

(ii) Suppose that upu_p is bijective, that MM is finitely generated and that NN is finitely presented. Then there exists f∈A−pf \in A - p such that ufu_f is bijective.

Let RR and QQ be the kernel and cokernel of uu; if g∈Ag \in A, the kernel and cokernel of ugu_g (resp. upu_p) are RgR_g and QgQ_g, (resp. RpR_p and QpQ_p) (§ 2, no. 4, Theorem 1). Then Qp=0Q_p = 0; as NN is finitely generated, so is QQ and there exists g′∈A−pg' \in A - p such that g′Q=0g'Q = 0 (§ 2, no. 2, Corollary 2 to Proposition 4), whence Qg′=0Q_{g'} = 0. Under the hypotheses of (ii), the sequence 0→Rg′→M→N→00 \to R_{g'} \to M \to N \to 0 is exact, hence Rg′R_{g'} is finitely generated (Chapter I, § 2, no. 8, Lemma 9). Now,

(Rg′)pRg′=Rp=0; (R_{g'})_{pR_{g'}} = R_p = 0;

hence there exists g1∈A,−pAg′g_1 \in A, - pA_{g'} such that g1Rg′=0g_1 R_{g'} = 0 (§ 2, no. 2, Corollary 2 to Proposition 4). Then g1=g′′/g′hg_1 = g''/{g'}^h, where g′′∈A−pg'' \in A - p; as g′/1g'/1 is invertible in Rg′R_{g'}, (g′′/1)Rg′=0(g''/1)R_{g'} = 0, whence Rg′g′′=(Rg′)g′′/1=0R_{g'g''} = (R_{g'})_{g''/1} = 0. If f=g′g′′f = g'g'', f∈A−pf \in A - p, Qf=0Q_f = 0 and Rf=0R_f = 0, so that ufu_f is bijective.

Corollary

If NN is finitely presented and NpN_p is a free ApA_p-module of rank pp, there exists f∈A−pf \in A - p such that NN, is a free AfA_f-module of rank pp.

There exist by hypothesis pp elements xi∈Nx_i \in \mathbf{N} (1≤i≤p1 \leq i \leq p) such that the xi/1x_i/1 form a basis of the free A,−A_,--module NpN_p. Consider the homomorphism u:Ap→Nu : A^p \to \mathbf{N} such that u(ei)=xiu(e_i) = x_i for 1≤i≤p1 \leq i \leq p, (ei)1≤i≤p(e_i)_{1 \leq i \leq p} being the canonical basis of ApA^p. As upu_p is bijective by hypothesis, there exists f∈A−pf \in A - p such that ufu_f is bijective, by virtue of Proposition 2.

Proposition 3

Let* (fi)i∈I(f_i)_{i \in I} be a finite family of elements of a ring AA, generating the ideal AA of AA. The ring B=∏i∈IAiB = \prod_{i \in I} A_i, is then a faithfully flat AA*-module*.

By § 2, no. 4, Theorem 1, each of the AiA_i is a flat AA-module, hence so is BB (Chapter I, § 2, no. 3, Proposition 2). On the other hand, if pp is a prime ideal of AA, there exists an index ii such that fi∉pf_i \notin p and pfi=pAfip_{f_i} = pA_{f_i} is therefore a prime ideal of AfiA_{f_i}. Then pB⊂pAfi×∏j≠iAj≠BpB \subset pA_{f_i} \times \prod_{j \neq i} A_j \neq B since pAfi≠AfipA_{f_i} \neq A_{f_i}; this suffices to imply that BB is a faithfully flat AA-module (Chapter I, § 3, no. 1, Proposition 1).

Corollary

Under the hypotheses of Proposition 3, for an* AA*-module* MM to be finitely generated (resp. finitely presented), it is necessary and sufficient that, for every index ii, the AfiA_{f_i}*-module* MfiM_{f_i} be finitely generated (resp. finitely presented).

The condition is obviously necessary (§ 2, no. 4). Conversely, if all the MfiM_{f_i} are finitely generated (resp. finitely presented), M′=∏i∈IMfiM' = \prod_{i \in I} M_{f_i} is a finitely generated (resp. finitely presented) BB-module, for we can obviously suppose that for each ii there is an exact sequence ATi→Afin→Mfi→0AT_i \to A^n_{f_i} \to M_{f_i} \to 0, where mm and nn are independent of ii. Now, M′=M⊗ABM' = M \otimes_A B. The corollary then follows from Proposition 3 and Chapter I, § 3, no. 6, Proposition 11.

Note that the condition on the fif_i means that the open sets XfiX_{f_i} form a covering of Spec(A)\mathrm{Spec}(A) (§ 4, no. 3, Corollary 3 to Proposition 11).

2. Local characterization of finitely generated projective modules

Theorem 1

Let* AA be a ring and PP an AA*-module*. The following properties are equivalent:

(a) PP is a finitely generated projective module. (b) PP is a finitely presented module and, for every maximal ideal mm of AA, PmP_m is a free AmA_m*-module*. (c) PP is a finitely generated module, for all p∈Spec(A)p \in \mathrm{Spec}(A), the ApA_p*-module* PpP_p is free and, if we denote its rank by rpr_p, the function p↦rpp \mapsto r_p is locally constant in the topological space Spec(A)\mathrm{Spec}(A) (that is, every point of Spec(A)\mathrm{Spec}(A) admits a neighbourhood in which this function is constant).

(d) There exists a finite family (fi)i∈I(f_i)_{i \in I} of elements of AA, generating the ideal AA, such that, for all i∈Ii \in I, the A,A,-module PfiP_{f_i} is free with finite rank.

(e) For every maximal ideal mm of AA, there exists f∈A−mf \in A - m such that PfP_f is a free A,A,-module of finite rank.

We show the theorem by proving the following scheme of implications

(a) ⇔ (d)

(b) ⇌ (e)

(c)

(a) ⇒ (b): We know that a finitely generated projective module is finitely presented (Chapter I, § 2, no. 8, Lemma 8 (iii)); if PP is a projective AA-module, Pm=P⊗AAmP_m = P \otimes_A A_m is a projective AmA_m-module (\emph{Algebra}, Chapter II, § 5, no. 1, Corollary to Proposition 4); finally, as AmA_m is a local ring, every finitely presented projective AmA_m-module is free (§ 3, no. 2, Corollary to Proposition 5).

(b) ⇒ (e): This follows from the Corollary to Proposition 2 of no. 1.

(c) ⇒ (e): Let mm be a maximal ideal of AA; write rm=nr_m = n and let (xi)1≤i≤n(x_i)_{1 \leq i \leq n} be a basis of PP. We can assume that the xix_i are canonical images of elements pi∈Pp_i \in P (1≤i≤n)(1 \leq i \leq n) to within multiplication by an invertible element of A,A,. Let (ei)1≤i≤n(e_i)_{1 \leq i \leq n} be the canonical basis of AnA^n and u:An→Pu : A^n \to P the homomorphism such that u(ei)=piu(e_i) = p_i for 1≤i≤n1 \leq i \leq n. As PP is finitely generated, it follows from Proposition 2 of no. 1 that there exists f∈A−mf \in A - m such that ufu_f is surjective. We conclude that uf,gu_{f,g} is also surjective for all g∈A−mg \in A - m and by hypothesis there exists g∈A−mg \in A - m such that rp=nr_p = n for p∈X,p \in X,. Then, replacing ff by fgf g, we may assume that rp=nr_p = n for all p∈X,p \in X,. Then up:Apn→Ppu_p : A_p^n \to P_p is a surjective homomorphism and PpP_p and ApA_p are both free A,A,-modules of the same rank; hence (§ 3, no. 2, Corollary to Proposition 6) upu_p is bijective for all p∈Xfp \in X_f. Let p′p' be a prime ideal of A,A, and let pp be its inverse image in AA under the canonical mapping; if (Afn)p′(A_f^{n})_{p'} and (Pf)p′(P_f)_{p'} are identified with ApnA_p^n and PpP_p under the canonical isomorphisms, (uf)p′(u_f)_{p'} is identified with upu_p and is consequently bijective. We conclude that ufu_f is bijective (§ 3, no. 3, Theorem 1), which establishes (e).

(e) ⇒ (d): Let EE be the set off ∈A\in A such that PfP_f is a finitely generated free AfA_f-module. The hypothesis implies that EE is contained in no maximal ideal of A,A, hence EE generates the ideal AA and there therefore exist a finite family (fi)1≤i≤n(f_i)_{1 \leq i \leq n} of elements of EE and ai∈Aa_i \in A (1≤i≤n)(1 \leq i \leq n) such that 1=∑i=1naifi1 = \sum_{i=1}^n a_i f_i; whence (d).

(d) ⇒ (c): It follows from no. 1, Corollary to Proposition 3 that PP is finitely generated. On the other hand, for every prime ideal pp of AA, there exists an index ii such that p∈Xfip \in X_{f_i}; if p′=pfip' = p_{f_i}, then Pp=(Pfi)p′P_p = (P_{f_i})_{p'} (§ 2, no. 5, Proposition

10) and hence by hypothesis PpP_p is free and of the same rank as PfiP_{f_i}, which proves (c).

(d) ⇒ (a): Consider the ring B=∏i∈IAiB = \prod_{i \in I} A_i and the B-module

M=∏i∈IPi=P⊗AB. M = \prod_{i \in I} P_i = P \otimes_A B.
For every index ii, there exists a free AiA_i-module LL, such that PiP_i is a direct factor of LL, and it may be assumed that all the LL, have the same rank; then L=∏i∈ILiL = \prod_{i \in I} L_i is a free B-module of which M is a direct factor, in other words M is a finitely generated projective B-module. As B is a faithfully flat A-module (no. 1, Proposition 3), we conclude that P is a finitely generated projective A-module (Chapter I, §3, no. 6, Proposition 12).

Corollary 1

Suppose that the equivalent properties of the statement of Theorem 1 hold. Let m be an integer > 0 such that, for every family (xi)1≤i≤m(x_i)_{1 \leq i \leq m} of elements of P, there exists a family (ai)1≤i≤m(a_i)_{1 \leq i \leq m} of elements of A, which are not all divisors of zero and for which ∑i=1maixi=0\sum_{i=1}^m a_i x_i = 0. Then, for all p∈Spec⁡(A)p \in \operatorname{Spec}(A), rp≤mr_p \leq m.

Let p be a prime ideal of A; set r=rpr = r_p and let (yj)1≤j≤r(y_j)_{1 \leq j \leq r} be a basis of the free ApA_p-module PpP_p. There exist elements xjx_j (1≤j≤r1 \leq j \leq r) of P and s∈A−ps \in A - p such that yj=xj/sy_j = x_j / s for all jj. Then for every family (aj)1≤j≤r(a_j)_{1 \leq j \leq r} of elements of A such that ∑j=1rajxj=0\sum_{j=1}^r a_j x_j = 0, ∑j=1r(aj/1)yj=0\sum_{j=1}^r (a_j/1) y_j = 0 in PpP_p, whence aj/1=0a_j/1 = 0 for 1≤j≤r1 \leq j \leq r. As A−pA - p does not contain 0, this shows that the aja_j are all divisors of zero in A (§ 2, no. 1, Remark 3), hence of necessity r≤mr \leq m.

Corollary 2

Every finitely presented flat module is projective.

If P is a finitely presented flat A-module and m a maximal ideal of A, the AmA_m-module P, is flat (§ 3, no. 4, Proposition 13) and finitely presented (§ 2, no. 4) and hence free (§ 3, no. 2, Corollary 2 to Proposition 5), Condition (b) of Theorem 1 therefore holds.

Remarks (1) There exist finitely generated flat modules which are not projective (Exercise 7). (2) Corollary 2 to Theorem 1 extends to modules over non-commutative rings (Chapter I, §2, Exercise 15).

3. Ranks of projective modules

Definition 1

Let P be a finitely generated projective A-module. For every prime ideal pp of A, the rank of the free ApA_p-module PpP_p is called the rank of P at pp and is denoted by rg⁡p(P)\operatorname{rg}_p(P).

By Theorem 1 the integer-valued function p↦rgp(P)p \mapsto \mathrm{rg}_p(P) is locally constant on X=Spec(A)X = \mathrm{Spec}(A); it is therefore constant if XX is connected and in particular if the ring AA is an integral domain (§ 4, no. 3, Corollary 2 to Proposition 15).

Definition 2

Let nn be an integer ⩾0\geqslant 0. A projective AA-module PP is said to be of rank nn if it is finitely generated and rgp(P)=n\mathrm{rg}_p(P) = n for every prime ideal pp of AA.

Clearly every finitely generated free AA-module LL is of rank nn in the sense of Definition 2, nn being equal to the dimension (or rank) of LL defined in Algebra, Chapter II, § 7, no. 2.

A projective module of rank 0 is zero (§ 3, no. 3, Corollary 2 to Theorem 1). If AA is not reduced to 0 and a projective AA-module PP is of rank nn, the integer nn is determined uniquely; it is then denoted by rg(P)\mathrm{rg}(P).

Theorem 2

Let PP be an AA-module and nn an integer ⩾0\geqslant 0. The following properties are equivalent:

(a) PP is projective of rank nn.

(b) PP is finitely generated and, for every maximal ideal mm of AA, the AA-*module* PP, is free of rank nn.

(c) PP is finitely generated and, for every prime ideal pp of AA, the AA-*module* PpP_p is free of rank nn.

(d) For every maximal ideal mm of AA, there exists f∈A−mf \in A - m such that the AA-*module* PfP_f is free of rank nn.

By Definition 2 and Theorem 1, (a) and (c) are equivalent; (b) implies (c), as, for every prime ideal pp of AA, there exists a maximal ideal mm containing pp and, writing p′=pmp' = p_m, PpP_p is isomorphic to (Pm)p′(P_m)_{p'} (§ 2, no. 5, Proposition 11); if PmP_m is free of rank nn, so then is Pp′P_{p'}. Property (c) implies (d) by virtue of Theorem 1 and the fact that, iff ∈A−m\in A - m and m′=mfm' = m_f, PmP_m is isomorphic to (Pf)m′(P_f)_{m'} and therefore the ranks of PfP_f and PP are equal. Finally, this last argument and Theorem 1 show that (d) implies (b).

Remark

If* AA is an integral domain, a projective AA-*module admits a well-defined rank* (in the sense of Definition 2), as has been observed above; moreover, this rank coincides with the rank defined in Algebra, Chapter II, § 7, no. 2; it is sufficient to apply Theorem 2 (c) with p=(0)p = (0).

Let EE and FF be two finitely generated projective AA-modules. We know (Algebra, Chapter II, §§ 2 and 3) that E×FE \times F, E⊗AFE \otimes_A F, HomA(E,F)\mathrm{Hom}_A(E, F) and the dual E∗E^* of EE are projective and finitely generated; so is the exterior power ⋀kE\bigwedge^k E for every integer k>0k > 0 (Algebra, Chapter 111). Also, it follows immediately from Definition 1 and § 2, no. 7, Propositions 18 and 19 and no. 8, that, for every prime ideal pp of AA:

(1)

rgp(E×F)=rgp(E)+rgp(F) \mathrm{rg}_p(E \times F) = \mathrm{rg}_p(E) + \mathrm{rg}_p(F)

(2) rgp(E⊗AF)=rgp(E)⋅rgp(F)\mathrm{rg}_p(E \otimes_A F) = \mathrm{rg}_p(E) \cdot \mathrm{rg}_p(F)

(3) rgp(HomA(E,F))=rgp(E)⋅rgp(F)\mathrm{rg}_p(\mathrm{Hom}_A(E, F)) = \mathrm{rg}_p(E) \cdot \mathrm{rg}_p(F)

(4) rgp(E∗)=rgp(E)\mathrm{rg}_p(E^*) = \mathrm{rg}_p(E)

(5) rgp(∧kE)=(rgp(E)k)\mathrm{rg}_p(\wedge^k E) = \binom{\mathrm{rg}_p(E)}{k}.

If the ranks of E and F are defined, so are those of E×F,E⊗AF,E \times F, E \otimes_A F, HomA(E,F),E∗\mathrm{Hom}_A(E, F), E^* and ∧kE\wedge^k E and the above equations also hold with the index pp omitted. Moreover:

Corollary

For a finitely generated projective A-module P to be of rank n, it is necessary and sufficient that AnP\mathbf{A}^n P be of rank 1.

Proposition 4

Let B be a commutative A-algebra and P a projective A-module of rank n. The B-module P(B)=B⊗APP_{(B)} = B \otimes_A P is then projective of rank n.

We know that P(B)P_{(B)} is projective and finitely generated (Algebra, Chapter II, § 5, no. 1, Corollary to Proposition 4). If q is a prime ideal of B and p its inverse image in A, then

(P(B))q=(P⊗AB)⊗BBq=P@ABq=(P⊗AAi)⊗ABq (P_{(B)})_q = (P \otimes_A B) \otimes_B B_q = P @_A B_q = (P \otimes_A A_i) \otimes_A B_q

and, as, by hypothesis, P⊗AAiP \otimes_A A_i is a free A,-module of rank nn, (P(B))q(P_{(B)})_q is a free BqB_q-module of rank nn.

Proposition 5

Let A be a semi-local ring and P a finitely generated projective A-module. If the rank of P is defined, P is a free A-module.

Suppose first that A is isomorphic to a product of fields KiK_i (1≤i≤n1 \leq i \leq n). The KiK_i are then identified with the minimal ideals (Algebra, Chapter VIII, § 3, no. 1) of A and, for all ii, the sum pip_i of the K, of index j≠ij \neq i is a maximal ideal of A, the pip_i (1≤i≤n1 \leq i \leq n) being the only prime ideals of A. Every finitely generated A-module P is therefore the direct sum of its isotypical components PiP_i (1≤i≤n1 \leq i \leq n), PiP_i being isomorphic to a direct sum of a finite number rir_i of A-modules isomorphic to K, (Algebra, Chapter VIII, § 5, no. 1, Proposition 1 and no. 3, Proposition 11); the ring AriA_{r_i} is identified with K, and annihilates the PjP_j of index j≠ij \neq i, hence ri=rgpi(P)r_i = \mathrm{rg}_{p_i}(P); if all the rir_i are equal to the same number rr, P is isomorphic to ArA^r, whence the proposition in this case. In the general case, let R\mathfrak{R} be the Jacobson radical of A and B=A/RB = A / \mathfrak{R}; as B is a product of fields, the projective B-module P(B)P_{(B)} is free by the remarks preceding Proposition 4. Also P is a flat A-module and the proposition then follows from § 3, no. 2, Proposition 5.

4. Projective modules of rank 1

Theorem 3

Let AA be a ring and MM ajnitely generated AA-module.

(i) If there exists an AA-module NN such that M⊗ANM \otimes_A N is isomorphic to AA, the module MM is projective of rank 1.

(ii) Conversely, if MM is projective of rank 1 and M∗M^* is the dual of MM, the canonical homomorphism u:M⊗AM∗→Au : M \otimes_A M^* \to A corresponding to the canonical bilinear form (x,x∗)→⟨x,x∗⟩(x, x^*) \to \langle x, x^* \rangle on M×M∗M \times M^* (Algebra, Chapter II, § 2, no. 3) is bijective.

(i) It is required to prove that, for every maximal ideal mm of AA, the AmA_m-module MM, is free of rank 1 (Theorem 2 (b)); we are free to replace AA by AmA_m and hence may assume that AA is a local ring (§2§ 2, no. 7, Proposition 18). Let k=A/mk = A/m. The isomorphism v:M⊗AN→Av : M \otimes_A N \to A defines an isomorphism

v⊗1k:(M/mM)⊗k(N/mN)→k v \otimes 1_k : (M/mM) \otimes_k (N/mN) \to k

as the rank over kk of (M/mM)⊗k(N/mN)(M/mM) \otimes_k (N/mN) is the product of the ranks of M/mMM/mM and N/mNN/mN, these latter are necessarily equal to 1, in other words M/mMM/mM is monogenous. It follows that MM is monogenous (§3§ 3, no. 2, Corollary 2 to Proposition 4); on the other hand, the annihilator of MM also annihilates M⊗ANM \otimes_A N and hence is zero, which proves that MM is isomorphic to AA.

(ii) It is sufficient to prove that, for every maximal ideal mm of AA, umu_m is an isomorphism (§3§ 3, no. 3, Theorem 1). As MM is finitely presented (Chapter I, §2§ 2, no. 8, Lemma 8), (M∗)m(M^*)_m is canonically identified with the dual (Mm)∗(M_m)^* (§2§ 2, no. 7, Proposition 19) and, as MmM_m is free of rank 1 like its dual (Mm)∗(M_m)^*, clearly the canonical homomorphism um:(Mm)⊗Am(Mm)∗→Amu_m : (M_m) \otimes_{A_m} (M_m)^* \to A_m is bijective, which completes the proof.

Remark (1). If MM is projective of rank 1 and NN is such that M⊗ANM \otimes_A N is isomorphic to AA, then NN is isomorphic to M∗M^*: there are isomorphisms

N→N⊗A→N⊗M⊗M∗+A⊗M∗→M∗. N \to N \otimes A \to N \otimes M \otimes M^* + A \otimes M^* \to M^*.

Proposition 6

Let MM and NN be projective AA-modules of rank 1. Then M⊗ANM \otimes_A N, Hom⁡A(M,N)\operatorname{Hom}_A(M, N) and the dual M∗M^* of MM are projective of rank 1.

This follows immediately from formulae (2), (3) and (4).

Let us now note that every finitely generated AA-module is isomorphic to a quotient module of L=A(N)L = \mathbf{A}^{(\mathbf{N})}; we may therefore speak of the set F(A)F(A) of classes of finitely generated AA-modules with respect to isomorphism (Set Theory, Chapter I, §6, no. 9); we denote by P(A)P(A) the subset of F(A)F(A) consisting of the classes of projective AA-modules of rank 1 and by cl⁡(M)\operatorname{cl}(M) the image in P(A)P(A) of a projective AA-module MM of rank 1. It is immediate that, for two projective AA-modules MM,

N of rank 1, cl(M \otimes_A N) depends only on cl(M) and cl(N); as definition we set

(6)

cl(M)+cl(N)=cl(M⊗AN) \mathrm{cl}(M) + \mathrm{cl}(N) = \mathrm{cl}(M \otimes AN)

and an internal law of composition is thus defined on P(A)P(A).

Proposition 7

The set P(A)P(A) of classes of projective A-modules of rank 1, with the law of composition (6), is a commutative group. If M is a projective A-module of rank 1 and M∗M^* is its dual, then

(7)

cl(M∗)=−cl(M)andcl(A)=0. \mathrm{cl}(M^*) = -\mathrm{cl}(M) \quad \text{and} \quad \mathrm{cl}(A) = 0.

The associativity and commutativity of the tensor product show that the law of composition (6) is associative and commutative; the isomorphism between A⊗AMA \otimes_A M and M prove that cl(A)\mathrm{cl}(A) is the identity element under this law and, by virtue of Theorem 3, cl(M)+cl(M∗)=cl(A)\mathrm{cl}(M) + \mathrm{cl}(M^*) = \mathrm{cl}(A), whence the proposition.

Let B be a commutative A-algebra and M a projective A-module of rank 1; then M(B)=B⊗AMM_{(B)} = B \otimes_A M is a projective B-module of rank 1 (no. 3, Proposition 4). Then there exists a mapping called canonical ϕ:P(A)→P(B)\phi : P(A) \to P(B) such that

(8)

ϕ(cl(M))=cl(M(B)). \phi(\mathrm{cl}(M)) = \mathrm{cl}(M_{(B)}).

The equation M(B)⊗BN(B)=(M⊗AN)(B)M_{(B)} \otimes_B N_{(B)} = (M \otimes_A N)_{(B)} for two A-modules M, N proves that the mapping ϕ\phi is a commutative group homomorphism.

Remark (2). Condition (e) of Theorem 1 (equivalent to the fact that P is projective and finitely generated) may also be expressed by saying that the sheaf of modules P~\tilde{P} over X=Spec(A)X = \mathrm{Spec}(A) associated() with P is locally free and of finite type* and may therefore be interpreted as the sheaf of sections of a vector bundle over X. Conversely, every vector bundle over X arises from a finitely generated projective module, which is determined to within a unique isomorphism; the projective modules of rank nn thus correspond to the vector bundles all of whose fibres have dimension nn. In particular, the vector bundles of rank 1 correspond to the projective modules of rank 1. If we denote by OX\mathscr{O}_X the structure sheaf A~\tilde{A} and by OX∗\mathscr{O}_X^* the sheaf of units of OX\mathscr{O}_X (whose sections over an open set U of X are the invertible elements of the ring of sections of OX\mathscr{O}_X over U), it follows that the group P(A)P(A) is isomorphic to the first cohomology group H1(X,OX∗)H^1(X, \mathscr{O}_X^*).*

(* See A. Grothendieck, Éléments de géométrie algébrique, I (§ 1) (Publ. Math. I.H.E.S., no. 4, 1960).

5. Non-degenerate submodules

In this no. and the two following, AA denotes a ring, SS a multiplicative subset of AA consisting of elements which are not divisors of zero in AA, and BB the ring S−1AS^{-1}A; AA is canonically identified with a subring of BB (§2§ 2, no. 1, Remark 3). The elements of SS are therefore invertible in BB.

One of the most important special cases for applications is that where AA is an integral domain and SS is the set of elements ≠0\neq 0 of AA; BB is then the field of fractions of AA.

Definition 3

Let MM be a sub-A-module of BB. MM is called non-degenerate if B.M=BB.M = B.

If BB is a field, this condition simply means that MM is not reduced to 0.

Proposition 8

Let MM be a sub-A-module of BB. The following conditions are equivalent: (a) MM is non-degenerate. (b) MM meets SS. (c) If j:M→Bj : M \to B is the canonical injection, the homomorphism u=S−1j:S−1M→Bu = S^{-1}j : S^{-1}M \to B is bijective.

(a) implies (b), for if B.M=BB.M = B, there exists a∈A,s∈Sa \in A, s \in S and x∈Mx \in M such that (a/s)x=1(a/s)x = 1, hence ax=sax = s belongs to S∩MS \cap M. To see that (b) implies (c), note that uu is already injective (§2§ 2, no. 4, Theorem 1); moreover, if x∈M∩Sx \in M \cap S, the image under uu of x/x∈S−1Mx/x \in S^{-1}M in BB is equal to 1 and uu is therefore surjective. Finally, (c) clearly implies (a).

Corollary

If MM and NN are two non-degenerate sub-A-modules of BB, the A-modules M+N,M.NM + N, M.N and M∩NM \cap N are non-degenerate.

The assertion is trivial for M+NM + N; on the other hand if s∈S∩Ms \in S \cap M and t∈S∩Nt \in S \cap N, then st∈S∩(M.N)st \in S \cap (M.N) and st∈S∩(M∩N)st \in S \cap (M \cap N).

Given two sub-A-modules MM and NN of BB, let us denote by N:MN : M the sub-A-module of BB consisting of those b∈Bb \in B such that bM⊂NbM \subset N (Chapter I, §2§ 2, no. 10, Remark). If every b∈N:Mb \in N : M is mapped to the homomorphism hb:x↦bxh_b : x \mapsto bx of MM to NN, a canonical homomorphism b↦hbb \mapsto h_b is obtained from N:MN : M to Hom⁡A(M,N)\operatorname{Hom}_A(M, N).

Proposition 9

Let M,NM, N be two sub-A-modules of BB. If MM is non-degenerate, the canonical homomorphism from N:MN : M to Hom⁡A(M,N)\operatorname{Hom}_A(M, N) is bijective.

Let s∈S∩Ms \in S \cap M. If b∈N:Mb \in N : M is such that bx=0bx = 0 for all x∈Mx \in M, then bs=0bs = 0, whence b=0b = 0 since ss is not a divisor of 0 in BB. On the other hand, let f∈HomA(M,N)f \in \mathrm{Hom}_A(M, N) and set b=f(s)/sb = f(s)/s; for all x∈Mx \in M, there exists t∈St \in S such that tx∈Atx \in A. Then

f(x)=s−1t−1f(stx)=s−1t−1txf(s)=bx, f(x) = s^{-1}t^{-1}f(stx) = s^{-1}t^{-1}txf(s) = bx,
whence b∈N:Mb \in N : M and f=hbf = h_b, which proves the proposition.

Remark

In particular, A:MA : M is canonically identified with the dual M∗M^* of MM, the canonical bilinear form on M×M∗M \times M^* being identified with the restriction to M×(A:M)M \times (A : M) of the multiplication B×B→BB \times B \to B.

6. Invertible submodules

(We preserve the notation & no. 5.)

Definition 4

A sub*-A-module MM of BB is called invertible if there exists a sub-A-module NN of BB such that M.N=AM.N = A.

Example

If bb is invertible element of BB, the AA-module AbAb is invertible, as is seen by taking N=Ab−1N = Ab^{-1}.

Proposition 10

Let MM be an invertible sub*-A*-module & BB. Then: (i) There exists s∈Ss \in S such that As⊂M⊂As−1As \subset M \subset As^{-1} (and in particular MM is non-degenerate). (ii) A:MA : M is the only sub-A*-module* NN of BB such that M.N=AM.N = A.

If M.N=AM.N = A, then B.M=B.(B.M)⊃B.(M.N)=B.A=BB.M = B.(B.M) \supset B.(M.N) = B.A = B, hence MM is non-degenerate. Similarly NN is non-degenerate. If t∈S∩Mt \in S \cap M and u∈S∩Nu \in S \cap N (no. 5, Proposition 8 ), the element s=tus = tu belongs to S∩M∩NS \cap M \cap N, whence Ms⊂M.N=AMs \subset M.N = A and therefore As⊂M⊂As−1As \subset M \subset As^{-1}. On the other hand, obviously N⊂A:MN \subset A : M, whence

A=M.N⊂M.(A:M)⊂A A = M.N \subset M.(A : M) \subset A
and M.(A:M)=AM.(A : M) = A; multiplying the two sides by NN, we deduce A:M=NA : M = N, which completes the proof.

Theorem 4

Let MM be a non-degenerate sub*-A*-module* of BB. The following properties are equivalent: (a) MM is invertible. (b) MM is projective. (c) MM is projective of rank 1. (d) MM is a finitely generated AA-module and, for every maximal ideal mm of AA, the AmA_m-module MmM_m is monogenous.

Let us show first the equivalence of properties (a), (b) and (c). If (a) holds and N is sub-A-module of B such that M . N = A, then there is a relation

∑i=1pmini=1(mi∈M,ni∈N for all i). \sum_{i=1}^{p} m_i n_i = 1 \quad (m_i \in M, n_i \in N \text{ for all } i).

For all x∈Mx \in M, set vi(x)=nixv_i(x) = n_i x; the viv_i are linear forms on M and by (9) x=∑i=1nmivi(x)x = \sum_{i=1}^{n} m_i v_i(x) for all x∈Mx \in M; this proves (Algebra, Chapter 11, §2, no. 6, Proposition 12) that M is projective and generated by the mim_i; hence M is a finitely generated projective module.

Let m be a maximal ideal of A; we show that the integer r=rgm(M)r = \mathrm{rg}_m(M) is equal to 1. Let S' be the image of S in Am′\mathbf{A}_{m'}; as the elements of S are not divisors of 0 in A, those of S' are not divisors of 0 in Am′\mathbf{A}_{m'} since Am′\mathbf{A}_{m'} is a flat A-module (§ 2, no. 4, Theorem 1 and Chapter I, §2, no. 4, Proposition 3); then S′−1Am≠0{S'}^{-1}A_m \neq 0 and, as MmM_m is a free AmA_m-module of rank rr, S′−1Mm{S'}^{-1}M_m is a free S′−1Am{S'}^{-1}A_m-module of rank rr. But if T' is the image of A −m-m in S−1AS^{-1}A, S′−1A{S'}^{-1}A, (resp. S′−1Mm{S'}^{-1}M_m) is canonically identified with T′−1(S−1A){T'}^{-1}(S^{-1}A) (resp. T′−1(S−1M){T'}^{-1}(S^{-1}M)) (§ 2, no. 3, Proposition 7). Now S−1M=BS^{-1}M = B (Proposition 8 (c)) and hence T′−1(S−1M){T'}^{-1}(S^{-1}M) is a free A-module of rank 1 over T′−1(S−1A){T'}^{-1}(S^{-1}A), which proves that r=1r = 1 and shows the implication (a) ⇒\Rightarrow (c).

The implication (c) ⇒\Rightarrow (b) is trivial. Let us show that (b) ⇒\Rightarrow (a). There exists by hypothesis a family (not necessarily finite) (fλ)λ∈L(f_\lambda)_{\lambda \in L} of linear forms on M and a family (mλ)λ∈L(m_\lambda)_{\lambda \in L} of elements of M such that, for all x∈Mx \in M, the family (fλ(x))(f_\lambda(x)) has finite support and x=∑λ∈Lmλfλ(x)x = \sum_{\lambda \in L} m_\lambda f_\lambda(x) (Algebra, Chapter II, §2, no. 6, Proposition 12). Since M is non-degenerate, fλ(x)=nλxf_\lambda(x) = n_\lambda x for some nλ∈An_\lambda \in A: M by virtue of Proposition 9 of no. 5. Taking x as an element of M ∩S\cap S (no. 5, Proposition 8), it is seen that of necessity nλ=0n_\lambda = 0 except for a finite number of indices and ∑λ∈Lmλnλ=1\sum_{\lambda \in L} m_\lambda n_\lambda = 1. This obviously implies M . (A: M) = A, whence (a).

By virtue of Definition 2 of no. 3, (c) implies (d). Let us show the converse. As M is non-degenerate, its annihilator is zero (Proposition 8 (b)), then so is the annihilator of MmM_m (§ 2, no. 4, formula (9)). As M, is assumed to be a monogenous &-module, it is therefore free of rank 1 and it then follows from no. 3, Theorem 2 that M is projective of rank 1.

Corollary

Every invertible sub-A-module of B is flat and finitely presented. This follows from Theorem 4 (c).

Proposition 11

Let M, N be two sub-A-modules of B. Suppose that M is invertible. Then: (i) The canonical homomorphism M⊗AN→M.NM \otimes_A N \to M . N is bijective. (ii) N:M=N.(A:M)N : M = N . (A : M) and N=(N:M).MN = (N : M) . M.

Let jj be the canonical injection N→BN \to B. Since MM is a flat AA-module (Corollary to Theorem 4), 1⊗j:M⊗AN→M⊗AB1 \otimes j : M \otimes_A N \to M \otimes_A B is injective. But, as B=S−1AB = S^{-1}A, the BB-module M⊗ABM \otimes_A B is equal to S−1MS^{-1}M and hence is identified with BB since MM is non-degenerate (no. 5, Proposition 8). If this identification is made, the image of 1⊗j1 \otimes j is M.NM . N, whence (i).

Let us set M′=A:MM' = A : M. Then obviously M′.N⊂N:MM'.N \subset N : M and M.(N:M)⊂NM.(N : M) \subset N. On the other hand, since M.M′=AM.M' = A (Proposition 10),

N:M=M′.M.(N:M)⊂M′.N N : M = M'.M.(N : M) \subset M'.N

and N=M.M′.N⊂M.(N:M)N = M.M'.N \subset M.(N : M), whence (ii).

Remark

The proof of (i) in Proposition 11 uses only the fact that MM is flat and non-degenerate.

7. The group of classes of invertible modules

(We preserve the notation of nos. 5 and 6.)

Under multiplication, the sub-A-modules of B form a commutative monoid m\mathfrak{m}, with A as identity element. Then the invertible modules are the invertible elements of m\mathfrak{m} and therefore form a commutative group z\mathfrak{z}. We have seen (no. 6, Proposition 10) that the inverse of M∈zM \in \mathfrak{z} is A:MA : M.

Let A∗A^* (resp. B∗B^*) be the multiplicative group of invertible elements of A (resp. B) and let uu denote the canonical injection A+BA + B. For all b∈B∗b \in B^*, θ(b)=bA\theta(b) = bA is an invertible sub-A-module. The mapping θ:B∗→g\theta : B^* \to \mathfrak{g} is a homomorphism whose kernel is u(A∗)u(A^*); its cokernel will be denoted by c\mathfrak{c} or c(A)\mathfrak{c}(A). The group c\mathfrak{c} is called the group of classes of invertible sub-A-modules of B. The following exact sequence has been constructed

(10)(1)⟶A∗→uB∗→θg→ρc⟶(1) \text{(10)} \quad (1) \longrightarrow A^* \xrightarrow{u} B^* \xrightarrow{\theta} \mathfrak{g} \xrightarrow{\rho} \mathfrak{c} \longrightarrow (1)

where (1) denotes the group consisting only of the identity element and ρ\rho is the canonical mapping g→c=g/θ(B∗)\mathfrak{g} \to \mathfrak{c} = \mathfrak{g}/\theta(B^*).

As every invertible sub-A-module M of B is projective of rank 1 (no. 6, Theorem 4), the element cl(M)∈P(A)cl(M) \in P(A) is defined (no. 4).

Proposition 12

The mapping* cl:g→P(A)cl : \mathfrak{g} \to P(A) defines, by taking quotients, an isomorphism from c=g/θ(B∗)\mathfrak{c} = \mathfrak{g}/\theta(B^*) *onto the kernel of the canonical homomorphism

ϕ:P(A)→P(B) \phi : P(A) \to P(B)

(no. 4).

In other words, there is an exact sequence

(11)(1)⟶A∗→uB∗→θg→clP(A)→ϕP(B). \text{(11)} \quad (1) \longrightarrow A^* \xrightarrow{u} B^* \xrightarrow{\theta} \mathfrak{g} \xrightarrow{cl} P(A) \xrightarrow{\phi} P(B).

It follows from Proposition 11 of no. 6 and the definition of addition in P(A)P(A) that cl(M.N)=cl(M)+cl(N)cl(M.N) = cl(M) + cl(N) for M,NM, N in 3\mathbf{3}, which shows that clcl is a homomorphism. If M∈JM \in \mathfrak{J} is isomorphic to AA, there exists b∈Bb \in B such that M=AbM = Ab and, as MM is invertible, there exists b′∈Bb' \in B such that b′b=1b'b = 1, in other words bb is invertible in BB; the converse is immediate. Hence the kernel of clcl in J\mathfrak{J} is θ(B∗)\theta(B^*).

Let us now determine the image of clcl. If M∈JM \in \mathfrak{J}, then M⊗AB=S−1M=BM \otimes_A B = S^{-1}M = B (no. 5, Proposition 8 (c)), whence cl(M)∈Ker(ϕ)cl(M) \in \mathrm{Ker}(\phi). Conversely, let PP be a projective AA-module of rank 1 such that P(B)=P⊗ABP_{(B)} = P \otimes_A B is BB-isomorphic to BB. As PP is a flat AA-module, the injection u:A→Bu : A \to B defines an injection u⊗1:P→P(B)=Bu \otimes 1 : P \to P_{(B)} = B and PP is thus identified with a sub-AA-module of BB; by virtue of Proposition 8 (c) of no. 5, PP is non-degenerate and Theorem 4 of no. 6 shows that PP is invertible. The kernel of ϕ\phi is therefore equal to the image of cl:J→P(A)cl : \mathfrak{J} \to P(A).

Corollary 1

For two invertible sub-A-modules of B to have the same image in C\mathfrak{C}, it is necessary and sufficient that they be isomorphic.

Corollary 2

If the ring BB is semi-local, the group C\mathfrak{C} of classes of invertible sub-A-modules of B is canonically identified with the group P(A)P(A) of classes of projective A-modules of rank 1.

In this case P(B)=0P(B) = 0 (no. 3, Proposition 5).

Remark

The hypothesis of Corollary 2 is fulfilled in the two following cases:

(1) AA is an integral domain and SS is the set of elements ≠0\neq 0 of AA, BB then being the field of fractions of AA. The invertible sub-A-modules of BB are also called in this case invertible fractional ideals; those which are monogenous free AA-modules AbAb (b#0b \# 0 in BB) are just the fractional principal ideals defined in Algebra, Chapter VI, § 1, no. 5.

(2)* The ring AA is Noetherian and SS is the set of elements of AA which are not divisors of 0 such that BB is the total ring of fractions of AA. In this case S=A−⋃piS = A - \bigcup p_i, where the pip_i are the elements (finite in number) of Ass(A)\mathrm{Ass}(A) (Chapter IV, § 1), hence BB is semi-local (§ 3, no. 5, Proposition 17).

Exercises

See the exercises for § 5.

Transcribed from the printed volume. Not translated.