Bourbaki

§ 6. Morita equivalence of modules and algebras

In this section, kk denotes a commutative ring.

1. Commutant and Duality

Let A and B be kk-algebras. Recall (III, §4, No. 3, p. 466) that a bimodule over the algebras A and B is an (A(A, B)-bimodule P on which the two kk-module structures deduced from the A- and B-module structures coincide. To avoid any ambiguity, we say that P is an (A,B)k(A,B)_k-bimodule. Let P be an (A,B)k(A,B)_k-bimodule. We denote by P∗P^* the dual HomA(P,A)_A(P,A) of the left A-module underlying P. It is a (B,A)k(B,A)_k-bimodule (II, §1, No. 14, p. 226); for a∈Aa\in A, b∈B,x∈Pb\in B,x\in P, and x∗∈P∗x^*\in P^*, we have

⟨x,bx∗a⟩=⟨xb,x∗⟩a(1) \langle x, bx^*a\rangle =\langle xb, x^*\rangle a \tag{1}

We denote by sAd_sA_d the algebra A viewed as an (A,A)k(A,A)_k-bimodule (loc. cit.) and by Λ:P⊗BP∗→sAd\Lambda : P\otimes_BP^*\rightarrow_sA_d the homomorphism of (A,A)k(A,A)_k-bimodules determined by

Λ(x⊗x∗)=⟨x,x∗⟩(2) \Lambda (x\otimes x^*) =\langle x, x^*\rangle \tag{2}

for x∈Px\in P and x∗∈P∗x^*\in P^*. We denote by P~\widetilde{P} the dual HomB(P,B)_B(P,B) of the right B-module underlying P; it is a (B,A)k(B,A)_k-bimodule. We denote by Λ~:P~⊗AP→sBd\widetilde{\Lambda} :\widetilde{P}\otimes_AP\rightarrow_sB_d the homomorphism of (B,B)k(B,B)_k-bimodules determined by (3) Λ~(x~⊗x)=⟨x~,x⟩\widetilde{\Lambda}(\widetilde{x}\otimes x) =\langle \widetilde{x}, x\rangle

for x∈Px\in P and x~∈P~\widetilde{x}\in \widetilde{P}.

Now, suppose that the mapping b↦bPb\mapsto b_P is a bijection from B to EndA(P)_A(P); it is then an isomorphism from B to the opposite algebra of EndA(P)_A(P). The canonical homomorphism of Z\mathbf{Z}-modules from P∗⊗APP^*\otimes_AP to EndA(P)_A(P) (II, §4, No. 2, p. 271) then determines a homomorphism of Z\mathbf{Z}-modules Θ:P∗⊗AP→\Theta : P^*\otimes_AP\rightarrow B defined by

xΘ(x∗⊗y)=⟨x,x∗⟩y(4) x\Theta (x^*\otimes y) =\langle x, x^*\rangle y \tag{4}

for x,y∈Px, y\in P and x∗∈P∗x^*\in P^*. Because of (1), this homomorphism is (B(B, B)-linear, and we have

Θ(x∗⊗y)y∗=x∗⟨y,y∗⟩(5) \Theta (x^*\otimes y)y^*=x^*\langle y, y^*\rangle \tag{5}

for y∈Py\in P and x∗,y∗∈P∗x^*, y^*\in P^*. From (4) and (5), we deduce the following equalities in the (B,B)k(B,B)_k-bimodule P∗⊗AP:P^*\otimes_AP:

(y∗⊗x)Θ(x∗⊗y)=y∗⊗⟨x,x∗⟩y=y∗⟨x,x∗⟩⊗y=Θ(y∗⊗x)(x∗⊗y) (y^*\otimes x)\Theta (x^*\otimes y) =y^*\otimes \langle x, x^*\rangle y=y^*\langle x, x^*\rangle \otimes y= \Theta (y^*\otimes x)(x^*\otimes y)

for x,y∈Px, y\in P and x∗,y∗∈P∗x^*, y^*\in P^*, and therefore

sΘ(t)=Θ(s)t(6) s\Theta (t) = \Theta (s)t \tag{6}

for s,t∈P∗⊗APs, t\in P^*\otimes_AP. Likewise, for x,yx, y in P and x∗,y∗x^*, y^* in P∗P^*, we deduce the following equalities in the (A,A)k(A,A)_k-bimodule P⊗BP∗P\otimes_BP^* from (4) and (5):

(x⊗x∗)⟨y,y∗⟩=x⊗Θ(x∗⊗y)y∗=xΘ(x∗⊗y)⊗y∗=⟨x,x∗⟩(y⊗y∗) (x\otimes x^*)\langle y, y^*\rangle =x\otimes \Theta (x^*\otimes y)y^*=x\Theta (x^*\otimes y)\otimes y^*=\langle x, x^*\rangle (y\otimes y^*)

and therefore

uΛ(v)=Λ(u)v(7) u\Lambda (v) = \Lambda (u)v \tag{7}

for u,v∈P⊗BP∗u, v\in P\otimes_BP^*.

For any element x∗x^* of P∗P^*, denote the B-linear mapping x↦Θ(x∗⊗x)x\mapsto \Theta (x^*\otimes x) from P to B by σ(x∗)\sigma (x^*). We thus define a mapping σ\sigma from P∗P^* to P~\widetilde{P} that is (B(B, A)-linear and satisfies, by definition,

Θ(x∗⊗y)=⟨σ(x∗),y⟩(8) \Theta (x^*\otimes y) =\langle \sigma (x^*), y\rangle \tag{8}

for x∗∈P∗x^*\in P^* and y∈Py\in P. By the definition of Λ~\widetilde{\Lambda}, we therefore have

Θ=Λ~∘(σ⊗1P)(9) \Theta =\widetilde{\Lambda}\circ (\sigma \otimes 1_P) \tag{9}

Suppose that the mapping a↦aPa\mapsto a_P from A to EndB(P)_B(P) is bijective; it is then an algebra isomorphism. Analogously, we define a homomorphism of (A,A)k(A,A)_k-bimodules Θ~:P⊗BP~→sAd\widetilde{\Theta} : P\otimes_B\widetilde{P}\rightarrow_sA_d by setting

Θ~(x⊗y~)y=x⟨y~,y⟩(10) \widetilde{\Theta}(x\otimes \widetilde{y})y=x\langle \widetilde{y}, y\rangle \tag{10}

for x,y∈Px, y\in P and y~∈P~\widetilde{y}\in \widetilde{P}. We also define a homomorphism of (B,A)k(B,A)_k-bimodules σ~:P~→P∗\widetilde{\sigma}:\widetilde{P}\rightarrow P^* by setting

Θ~(x⊗y~)=⟨x,σ~(y~)⟩(11) \widetilde{\Theta}(x\otimes \widetilde{y}) =\langle x,\widetilde{\sigma}(\widetilde{y})\rangle \tag{11}

for x∈P,y~∈P~x\in P,\widetilde{y}\in \widetilde{P}. We have

Θ~=Λ∘(1P⊗σ~)(12) \widetilde{\Theta} = \Lambda \circ (1_P\otimes \widetilde{\sigma}) \tag{12}

Proposition 1 006R

Suppose that the mappings b↦bPb\mapsto b_P from B to EndA(P)_A(P) and a↦aPa\mapsto a_P from A to EndB(P)_B(P) are bijective. Then σ\sigma and σ~\widetilde{\sigma} are inverse isomorphisms, and we have the relations

Λ=Θ~∘(1P⊗σ)(13) \Lambda =\widetilde{\Theta}\circ (1_P\otimes \sigma ) \tag{13}
Λ~=Θ∘(σ~⊗1P)(14) \widetilde{\Lambda} = \Theta \circ (\widetilde{\sigma}\otimes 1_P) \tag{14}

For x∈P,x∗∈P∗x\in P,x^*\in P^*, and y∈Py\in P, by relations (4), (8), (10), and (11), we have

(15)⟨x,x∗⟩y=xΘ(x∗⊗y)=x⟨σ(x∗),y⟩=Θ~(x⊗σ(x∗))y=⟨x,σ~(σ(x∗))⟩y (15)\langle x, x^*\rangle y=x\Theta (x^*\otimes y) =x\langle \sigma (x^*), y\rangle =\widetilde{\Theta}(x\otimes \sigma (x^*))y=\langle x,\widetilde{\sigma}(\sigma (x^*))\rangle y

Likewise, for x∈P,y~∈P~x\in P,\widetilde{y}\in \widetilde{P}, and y∈Py\in P, we have

x⟨y~,y⟩=Θ~(x⊗y~)y=⟨x,σ~(y~)⟩y=xΘ(σ~(y~)⊗y)=x⟨σ(σ~(y~)),y⟩(16) x\langle \widetilde{y}, y\rangle =\widetilde{\Theta}(x\otimes \widetilde{y})y=\langle x,\widetilde{\sigma}(\widetilde{y})\rangle y=x\Theta (\widetilde{\sigma}(\widetilde{y})\otimes y) =x\langle \sigma (\widetilde{\sigma}(\widetilde{y})), y\rangle \tag{16}

Because of the assumptions, P is faithful as an A-module and as a B-module. From relations (15) and (16), respectively, we deduce σ~∘σ=1P∗\widetilde{\sigma}\circ \sigma = 1_{P^*} and σ∘σ~=\sigma \circ \widetilde{\sigma}= 1P~1_{\widetilde{P}}. Relations (13) and (14) then follow from (12) and (9), respectively.

Remark 1 006S

Suppose that the mapping b↦bPb\mapsto b_P from B to EndA(P)_A(P) is bijective. Then

a) the B-module P can be identified with the countermodule of P; it is therefore faithful and balanced;

b) the mapping a↦aPa\mapsto a_P from A to EndB(P)_B(P) is bijective if and only if the A-module P is faithful and balanced.

Remark 2 006T

Under the assumptions of Proposition 1, the A-module P is balanced; since the rings APA_P and AP′A'_P have the same center (VIII, p. 77), there is an isomorphism φ\varphi from the center Z(A) of the ring A to the center Z(B) of the ring B, determined by the relation φ(z)P=zP\varphi (z)_P=z_P for z∈Z(A)z\in Z(A). Moreover, the endomorphisms of the (A,B)k(A,B)_k-bimodule P are the homotheties zPz_P where zz runs through Z(A); the automorphisms of the (A,B)k(A,B)_k-bimodule P are the homotheties zPz_P where zz is invertible in Z(A).

2. Generating Modules and Finitely Generated Projective Modules

Proposition 2 006U

Let A and B be algebras over kk, and let P be an (A,B)k(A,B)_k-bimodule. Suppose that the mapping b↦bPb\mapsto b_P from B to EndA(P)_A(P) is bijective. The following assertions are equivalent:

(i) The A-module P is projective and finitely generated.

(ii) The mapping Θ (VIII, p. 96) is an isomorphism of (B,B)k(B,B)_k-bimodules from P∗⊗APP^*\otimes_AP to sBd_sB_d.

(iii) The image of Θ contains the unit element of B.

If, moreover, the mapping a↦aPa\mapsto a_P from A to EndB(P)_B(P) is bijective, then the assertions above are equivalent to the following statement:

(iv) There exist a (B,A)k(B,A)_k-bimodule Q and a surjective homomorphism of (B,B)k(B,B)_k-bimodules from Q⊗APQ\otimes_AP to sBd_sB_d.

By the corollary of II, §4, No. 2, p. 271, assertion (i) implies (ii). Moreover, (iii) follows from (ii). Let t∈B∗⊗ABt\in B^*\otimes_AB be such that Θ(t)=1\Theta (t) = 1. Let nn be an integer, and let (x1,...,xn)∈Pn(x_1, . . . , x_n)\in P^n and (x1∗,...,xn∗)∈P∗n(x^*_1, . . . , x^*_n)\in P^{*n} be such that t=∑i=1nxi∗⊗xit=\sum^n_{i=1}x^*_i\otimes x_i. For every xx belonging to P, the relation xΘ(t)=xx\Theta (t) =x can be written as

∑i=1n⟨x,xi∗⟩xi=x \sum_{i=1}^n\langle x, x^*_i\rangle x_i=x

It follows that the A-module P is finitely generated by the family (xi)1⩽i⩽n(x_i)_{1\leqslant i\leqslant n}, and we conclude the proof of the implication (iii) ⇒\Rightarrow (i) using Proposition 12 of II, §2, No. 6, p. 238.

We obviously have (ii) ⇒\Rightarrow (iv). Let Q be a (B,A)k(B,A)_k-bimodule and θ\theta a surjective homomorphism of (B,B)k(B,B)_k-bimodules from Q⊗APQ\otimes_AP to sBd_sB_d. In the notation of the previous subsection, there exists a homomorphism of (B,A)k(B,A)_k-bimodules ζ:Q→P~\zeta : Q\rightarrow \widetilde{P} such that θ(y⊗x)=⟨ζ(y),x⟩\theta (y\otimes x) =\langle \zeta (y), x\rangle for x∈Px\in P and y∈Qy\in Q, and we have θ=Λ~∘(ζ⊗1P)\theta =\widetilde{\Lambda}\circ (\zeta \otimes 1_P). Since θ\theta is surjective, the same holds for Λ~\widetilde{\Lambda}. If the mapping a↦aPa\mapsto a_P from A to EndB(P)_B(P) is bijective, then Θ is surjective by relation (14) of Proposition 1 of VIII, p. 97; the implication (iv) ⇒\Rightarrow (iii) follows.

Proposition 3 006V

Let A and B be algebras over kk and P an (A,B)k(A,B)_k-bimodule. The following properties are equivalent:

(i) The A-module P is generating.

(ii) The image of the mapping Λ from P⊗BP∗P\otimes_BP^* to sAd_sA_d (VIII, p. 95) contains the unit element.

(iii) There exist a (B,A)k(B,A)_k-bimodule Q and a surjective homomorphism of (A,A)k(A,A)_k-bimodules from P⊗BQP\otimes_BQ to sAd_sA_d.

If, moreover, the mapping b↦bPb\mapsto b_P from B to EndA(P)_A(P) is bijective, then the assertions are equivalent to the following statement:

(iv) The mapping Λ is an isomorphism from P⊗BP∗P\otimes_BP^* to sAd_sA_d.

(i) ⇔\Leftrightarrow (ii): The image of Λ is the trace ideal τ(P)\tau (P) (VIII, p. 80). The equivalence of (i) and (ii) therefore follows from Theorem 1 of VIII, p. 80.

(ii) ⇒\Rightarrow (iii): It suffices to take Q=P∗Q = P^*.

(iii) ⇒\Rightarrow (ii): Let Q be a (B,A)k(B,A)_k-bimodule and ψ\psi a homomorphism of (A,A)k(A,A)_k-bimodules from P⊗BQP\otimes_BQ to sAd_sA_d. There exists a homomorphism of (B,A)k(B,A)_k-bimodules Ψ from Q to P∗P^* such that ψ(x⊗y)=⟨x,Ψ(y)⟩\psi (x\otimes y) =\langle x,\Psi (y)\rangle for x∈Px\in P and y∈Qy\in Q, and we have the equality ψ=Λ∘(1P⊗Ψ)\psi = \Lambda \circ (1_P\otimes \Psi ). If ψ\psi is surjective, then so is Λ.

It is clear that (iv) implies (ii). Conversely, suppose that property (ii) holds and that the mapping b↦bPb\mapsto b_P from B to EndA(P)_A(P) is bijective. Let ee be an element of P⊗BP∗P\otimes_BP^* such that Λ(e)=1\Lambda (e) = 1. By relation (7) of VIII, p. 96, we have u=Λ(u)eu= \Lambda (u)e for every uu in P⊗BP∗P\otimes_BP^*; the injectivity of Λ follows.

Proposition 4 006W

Let A and B be kk-algebras and P an (A,B)k(A,B)_k-bimodule. Suppose that the mapping b↦bPb\mapsto b_P from B to EndA(P)_A(P) is bijective.

a) If the A-module P is generating, then the right B-module P is projective and finitely generated.

b) If the A-module P is projective and finitely generated, then the right B-module P is generating.

Suppose that the A-module P is generating. It is then faithful and balanced (VIII, p. 82, Theorem 2), and consequently the mapping a↦aPa\mapsto a_P from A to EndB(P)_B(P) is bijective (VIII, p. 97, Remark 1). Moreover, the mapping Λ:P⊗BP∗→sAd\Lambda : P\otimes_BP^*\rightarrow_sA_d is bijective (VIII, p. 98, Proposition 3). We view P as a (Bo,Ao)k(B^o,A^o)_k-bimodule. The mapping Λ induces a bijective mapping P∗⊗BoP→AoP^*\otimes_{B^o}P\rightarrow A^o; by Proposition 2 of VIII, p. 98, (iv) ⇒\Rightarrow (i), the right B-module P is projective and finitely generated.

Now, suppose that the A-module P is projective and finitely generated. Then the mapping Θ:P∗⊗AP→sBd\Theta : P^*\otimes_AP\rightarrow_sB_d is bijective (loc. cit., (i) ⇒\Rightarrow (ii)). By the implication (iii) ⇒\Rightarrow (i) of Proposition 3 above applied to the (Bo,Ao)k(B^o,A^o)_k-bimodule P, the right B-module P is generating.

Corollary 1 006X

The countermodule of a generating module is projective and finitely generated. The countermodule of a finitely generated projective module is generating.

Let A be a kk-algebra, and let M be an A-module. We denote the opposite kk-algebra of the algebra EndA(M)_A(M) by B. The corollary follows from Proposition 4 applied to the (A,B)k(A,B)_k-bimodule M.

Corollary 2 006Y

Let A and B be kk-algebras and P an (A,B)k(A,B)_k-bimodule. The following properties are equivalent:

(i) The A-module P is generating, and the mapping b↦bPb\mapsto b_P from B to EndA(P)_A(P) is bijective.

(ii) The right B-module P is projective, finitely generated, faithful, and balanced, and the mapping a↦aPa\mapsto a_P from A to EndB(P)_B(P) is bijective.

The implication (i) ⇒\Rightarrow (ii) follows from Proposition 4, a) and Remark 1 (VIII, p. 97). Assume that (ii) holds; then the A-module P is generating (Proposition 4, b) applied to the (Bo,Ao)k(B^o,A^o)_k-bimodule P). Since the B-module P is faithful and balanced, the second assertion of (i) is also satisfied (VIII, p. 97, Remark 1).

3. Invertible Bimodules and Morita Equivalence

Definition 1 006Z

Let A and B be kk-algebras and P an (A,B)k(A,B)_k-bimodule. We say that P is invertible if there exists a (B,A)k(B,A)_k-bimodule Q such that P⊗BQP\otimes_BQ is isomorphic to sAd_sA_d and Q⊗APQ\otimes_AP to sBd_sB_d. Such a bimodule Q is called an inverse of P.

Let A and B be kk-algebras, and let P be an invertible (A,B)k(A,B)_k-bimodule. Let C be a kk-algebra and P′P' an invertible (B,C)k(B,C)_k-bimodule. Finally, let Q and Q′Q' be inverse bimodules of P and P′P', respectively. By the associativity of the tensor product (II, §3, No. 8, p. 258, Proposition 8) and Proposition 4 of II, §3, No. 4, p. 249, the (C,A)k(C,A)_k-bimodule Q′⊗BQQ'\otimes_BQ is an inverse of the (A,C)k(A,C)_k-bimodule P⊗BP′P\otimes_BP', so that P⊗BP′P\otimes_BP' is an invertible (A,C)k(A,C)_k-bimodule.

Hence, the relation

“A and B are kk-algebras, and there exists an invertible (A,B)k(A,B)_k-bimodule” is an equivalence relation.

Definition 2 0070

Two kk-algebras A and B are called Morita equivalent if there exists an invertible (A,B)k(A,B)_k-bimodule. Rings A and B are called Morita equivalent if the Z\mathbf{Z}-algebras A and B are Morita equivalent.

Two isomorphic kk-algebras are Morita equivalent. If two kk-algebras are Morita equivalent, then their opposite algebras are Morita equivalent.

Let P be an invertible (A,B)k(A, B)_k-bimodule and Q an inverse of P. Then Q is an invertible (B,A)k(B, A)_k-bimodule, and it has P as inverse. Moreover, viewed as a (B∘,A∘)k(B^\circ, A^\circ)_k-bimodule, P is invertible and has the (A∘,B∘)k(A^\circ, B^\circ)_k-bimodule Q as inverse.

Lemma 1 0071

— *Let A and B be k-algebras, P be an invertible (A,B)k(A, B)_k-bimodule, M and N be B-modules, and u:M→Nu : M \to N be a B-linear mapping. If the mapping 1P⊗u:P⊗BM→P⊗BN1_P \otimes u : P \otimes_B M \to P \otimes_B N is zero (resp. bijective), then so is u.

Let Q be a bimodule inverse to P and θ:Q⊗AP→sBd\theta : Q \otimes_A P \to {}_s B_d an isomorphism of (B,B)k(B, B)_k-bimodules. The lemma follows from the commutativity of the diagram

Q⊗AP⊗BM→1Q⊗1P⊗uQ⊗AP⊗BN↓θ⊗1M↓θ⊗1MM→uN. \begin{array}{ccc} Q \otimes_A P \otimes_B M & \xrightarrow{1_Q \otimes 1_P \otimes u} & Q \otimes_A P \otimes_B N \\ \downarrow \theta \otimes 1_M & & \downarrow \theta \otimes 1_M \\ M & \xrightarrow{u} & N. \end{array}

Theorem 1. — *Let A and B be k-algebras and P an (A,B)k(A, B)_k-bimodule. Denote the (B,A)k(B, A)_k-bimodule Hom⁡A(P,As)\operatorname{Hom}_A(P, A_s) by P∗P^*. The following properties are equivalent:

(i) The (A,B)k(A, B)_k-bimodule P is invertible.

(ii) The A-module P is projective, finitely generated, and generating, and the mapping b↦bPb \mapsto b_P from B to End⁡A(P)∘\operatorname{End}_A(P)^{\circ} is a k-algebra isomorphism.

(iii) The right B-module P is projective, finitely generated, and generating, and the mapping a↦aPa \mapsto a_P from A to End⁡B(P)\operatorname{End}_B(P) is a k-algebra isomorphism.

If these properties hold, then the homomorphisms

Θ:P∗⊗P→sBdandΛ:P⊗P∗→sAd \Theta : P^* \otimes P \to {}_s B_d \quad \text{and} \quad \Lambda : P \otimes P^* \to {}_s A_d

are isomorphisms, so that the (B,A)k(B, A)_k-bimodule P∗P^* is an inverse of P.

If property (ii) holds, then P is an invertible (A,B)k(A, B)_k-bimodule with inverse P∗P^* (VIII, p. 98, Proposition 2 and p. 98, Proposition 3). This proves that (ii) implies (i) and the last assertion.

Suppose that the (A,B)k(A, B)_k-bimodule P is invertible. Then the A-module P is generating (VIII, p. 98, Proposition 3, (iii) ⇒\Rightarrow (i)). It is therefore faithful and balanced (VIII, p. 82, Theorem 2) and, consequently, the mapping a↦apa \mapsto a_p from A to End⁡B(P)\operatorname{End}_B(P) is bijective.

Let us now prove that the mapping b↦bPb \mapsto b_P from B to End⁡A(P)\operatorname{End}_A(P) is bijective. Let Q be a (B,A)k(B, A)_k-bimodule, inverse to P. Let u∈End⁡A(P)u \in \operatorname{End}_A(P); then 1Q⊗u1_Q \otimes u is an endomorphism of the left B-module Q⊗APQ \otimes_A P. Since the (B,B)k(B, B)_k-bimodule Q⊗APQ \otimes_A P is isomorphic to sBd_s B_d, there exists a unique element bb of B such that 1Q⊗u1_Q\otimes u is the homothety with ratio bb of the right B-module Q⊗APQ\otimes_AP. Consequently, we have 1Q⊗(u−bP)=01_Q\otimes (u-b_P) = 0. Hence, u=bPu=b_P by Lemma 1; this proves that the mapping b↦bPb\mapsto b_P from B to EndA(P)_A(P) is bijective.

By Proposition 2 of VIII, p. 98, the A-module P is then projective and finitely generated. We have therefore proved the equivalence of (i) and (ii).

By interchanging the roles of A and B, we obtain the equivalence of (i) and (iii), which concludes the proof of the proposition.

Corollary 1

Let A and B be Morita equivalent kk-algebras, and let P be an invertible (A,B)k(A,B)_k-bimodule. There exists an isomorphism φ\varphi from the center Z(A) of A to the center of B determined by the relation φ(z)P=zP\varphi (z)_P=z_P for every z∈Z(A)z\in Z(A). The automorphisms of the (A,B)k(A,B)_k-bimodule P are the homotheties zPz_P where zz is an invertible element of Z(A).

Given Theorem 1, this follows from Remark 2 of VIII, p. 97.

Corollary 2

Let A and B be Morita equivalent kk-algebras, and let P be an invertible (A,B)k(A,B)_k-bimodule. Every (B,A)k(B,A)_k-bimodule inverse to P is isomorphic to the dual P∗=P^*= HomA(P,A)_A(P,A) of P. More precisely, let Q be a (B,A)k(B,A)_k-bimodule that is an inverse of P, and let λ:P⊗BQ→sAd\lambda : P\otimes_BQ\rightarrow_sA_d be an isomorphism of (A,A)k(A,A)_k-bimodules. There exists a unique mapping τ:Q→P∗\tau : Q\rightarrow P^* determined by the relation ⟨p,τ(q)⟩=λ(p⊗q)\langle p, \tau (q)\rangle =\lambda (p\otimes q) for p∈Pp\in P and q∈Qq\in Q, and τ\tau is an isomorphism of (B,A)k(B,A)_k-bimodules.

The existence and uniqueness of the mapping τ\tau are clear. It is a homomorphism of (B,A)k(B,A)_k-bimodules, and we have λ=Λ∘(1P⊗τ)\lambda = \Lambda \circ (1_P\otimes \tau ). Since λ\lambda and Λ are isomorphisms of (A,A)k(A,A)_k-bimodules (VIII, p. 101, Theorem 1), so is 1P⊗τ1_P\otimes \tau. By Lemma 1, the mapping τ\tau is bijective.

Remark 0075

Under the assumptions of the corollary, let qq be an element of Q such that we have λ(p⊗q)=0\lambda (p\otimes q) = 0 for every p∈Pp\in P. We then have τ(q)=0\tau (q) = 0, that is, q=0q= 0. Likewise, if pp is an element of P such that we have λ(p⊗q)=0\lambda (p\otimes q) = 0 for every q∈Qq\in Q, then p=0p= 0.

Example 1 0076

Let B be a kk-algebra, nn an integer ⩾1\geqslant 1, and A the kk-algebra Mn(B)\mathbf{M}_n(B). The right B-module P=BdnP = B^n_d is projective, finitely generated, and generating, and A can be identified with the endomorphism algebra of P (II, §10, No. 7, p. 349). By Theorem 1, the (A,B)k(A,B)_k-bimodule P is invertible. The algebras B and Mn(B)\mathbf{M}_n(B) are therefore Morita equivalent.

Example 2 00SH

Let A be a commutative kk-algebra and P an A-module. We view P as an (A,A)k(A,A)_k-bimodule whose two laws of action are equal. If the (A,A)k(A,A)_k-bimodule P is invertible, then the A-module P is finitely generated (Theorem 1). By Theorem 3 of Comm. Alg., II, §5, No. 4, p. 114, the following properties are equivalent:

(i) The (A,A)k(A,A)_k-bimodule P is invertible.

(ii) There exists an A-module Q such that P⊗AQP\otimes_AQ is isomorphic to A.

(iii) The A-module P is projective, finitely generated, and of rank 1.∗1.*

4. The Morita Correspondence of Modules

In this subsection, the letters A and B denote Morita equivalent kk-algebras and P an invertible (A,B)k(A,B)_k-bimodule. Choose a (B,A)k(B,A)_k-bimodule Q inverse to P and isomorphisms

λ:P⊗BQ→sAd\lambda : P\otimes_BQ\rightarrow_sA_d and θ:Q⊗AP→sBd\theta : Q\otimes_AP\rightarrow_sB_d.

For any left B-module V, denote by θV\theta_V the homomorphism of B-modules θ⊗1V:Q⊗AP⊗BV→V\theta \otimes 1_V: Q\otimes_AP\otimes_BV\rightarrow V; it is an isomorphism since θ\theta is an isomorphism. Likewise, for every left A-module M, denote by λM\lambda_M the homomorphism of A-modules λ⊗1M:P⊗BQ⊗AM→M\lambda \otimes 1_M: P\otimes_BQ\otimes_AM\rightarrow M; it is an isomorphism since λ\lambda is an isomorphism.

Theorem 2 (Morita) 00S2

a) Let V and W be left B-modules. The mapping g↦1P⊗gg\mapsto 1_P\otimes g is a bijection from HomB(V,W)_B(V,W) to HomA(P⊗BV,P⊗BW)_A(P\otimes_BV,P\otimes_BW). The inverse bijection sends an element hh of HomA(P⊗BV,P⊗BW)_A(P\otimes_BV,P\otimes_BW) to the element θW∘(1Q⊗h)∘θV−1\theta_W\circ (1_Q\otimes h)\circ \theta_V^{-1} of HomB(V,W)_B(V,W).

b) Every left A-module M is isomorphic to a module of the form P⊗BVP\otimes_BV, where V is a left B-module.

Let V and W be left B-modules. By Lemma 1 of VIII, p. 101, the mapping φ:g↦1P⊗g\varphi :g\mapsto 1_P\otimes g from HomB(V,W)_B(V,W) to HomA(P⊗BV,P⊗BW)_A(P\otimes_BV,P\otimes_BW) is injective. By interchanging the roles of P and Q (and of A and B), we see that the mapping ψ:h↦1Q⊗h\psi :h\mapsto 1_Q\otimes h from HomA(P⊗BV,P⊗BW)_A(P\otimes_BV,P\otimes_BW) to HomA(Q⊗AP⊗BV,Q⊗AP⊗BW)_A(Q\otimes_AP\otimes_BV,Q\otimes_AP\otimes_BW) is also injective. Now, the composition ψ∘φ\psi \circ \varphi is the mapping g↦θW−1∘g∘θVg\mapsto \theta^{-1}_W\circ g\circ \theta_V. It is bijective, hence so is ψ\psi. Consequently, φ\varphi is bijective, and its inverse is the mapping h↦θW∘(1Q⊗h)∘θV−1h\mapsto \theta_W\circ (1_Q\otimes h)\circ \theta^{-1}_V.

Assertion b) follows from the fact that λM\lambda_M is an isomorphism from P⊗BP\otimes_B ⊗A⊗M\otimes_A\otimes M to M.

Let V be a left B-module and W a submodule of V. Since the B-module P is projective (VIII, p. 101, Theorem 1), the canonical mapping from P⊗BWP\otimes_BW to P⊗BVP\otimes_BV is injective. We identify P⊗BWP\otimes_BW with its image in P⊗BVP\otimes_BV through this mapping. We adopt the analogous conventions when P and B are replaced by Q and A.

Proposition 5 0077

Let V be a left B-module. The mapping W↦P⊗BWW\mapsto P\otimes_BW is an isomorphism from the set of B-submodules of V, ordered by inclusion, to the set of A-submodules of P⊗BVP\otimes_BV, ordered by inclusion. The inverse isomorphism sends an A-submodule N of P⊗BVP\otimes_BV to the image under θV\theta_V of the B-submodule Q⊗ANQ\otimes_AN of Q⊗AP⊗BVQ\otimes_AP\otimes_BV.

Denote by DB(V)D_B(V) the set of B-submodules of V, ordered by inclusion, and define the sets DA(P⊗BV)D_A(P\otimes_BV) and DB(Q⊗AP⊗BV)D_B(Q\otimes_AP\otimes_BV) likewise. Let φ:DB(V)→\varphi : D_B(V)\rightarrow DA(P⊗BV)D_A(P\otimes_BV) be the mapping W↦P⊗BWW\mapsto P\otimes_BW and ψ\psi the mapping from DA(P⊗BV)D_A(P\otimes_BV) to DB(Q⊗AP⊗BV)D_B(Q\otimes_AP\otimes_BV) given by N↦Q⊗ANN\mapsto Q\otimes_AN. These are increasing mappings, and the composition ψ∘φ\psi \circ \varphi is the mapping W↦θV−1W\mapsto \theta^{-1}_V(W), which is bijective. Consequently, φ\varphi is injective, and ψ\psi is surjective. By replacing B with A and V with P⊗BVP\otimes_BV, we see that ψ\psi is also injective. Hence, φ\varphi and ψ\psi are bijective, and the inverse mapping of φ\varphi is indeed that described in the proposition.

Example 1 0078

Let us apply Proposition 5 of VIII, p. 104, to the specific case V=BsV = B_s.

a) The mapping J↦J\mapsto PJ is an isomorphism from the ordered set D(Bs)D(B_s) of left ideals of B to the ordered set D(P) of A-submodules of P. The inverse mapping sends an A-submodule M of P to the left ideal J(M) of B consisting of the elements bb of B such that M contains PbPb.

b) The mapping K↦K\mapsto KP is an isomorphism from the ordered set D(Ad)D(A_d) of right ideals of A to the ordered set D(P) of B-submodules of P. The inverse mapping sends a B-submodule V of P to the right ideal K(V) of A consisting of the elements aa of A such that V contains aPaP.

Indeed, the A-module P⊗BBsP\otimes_BB_s can be canonically identified with P. If J is a left ideal of B, then the canonical image of P⊗BJP\otimes_BJ in P⊗BBsP\otimes_BB_s corresponds to PJ through this identification. Consequently, the mapping J↦J\mapsto PJ is an isomorphism of ordered sets from D(Bs)D(B_s) to D(P). Let J∈D(Bs)J\in D(B_s). Denote the set of elements bb of B such that PJ contains PbPb by J′J'. It is a left ideal of B that contains J, and we have PJ′⊂'\subset PJ. Since the mapping J↦J\mapsto PJ is an isomorphism of ordered sets, we necessarily have PJ′='= PJ and J=J′J = J'. This proves a).

Assertion b) follows from assertion a) applied to the invertible (Bo,Ao)k(B^o,A^o)_k-bimodule P.

Note that the ring A is a field if and only if the B-module P is simple.

Example 2 0079

Denote by BA,BB\mathscr{B}_A,\mathscr{B}_B, and BP\mathscr{B}_P the sets of two-sided ideals of A, two-sided ideals of B, and (A,B)k(A,B)_k-sub-bimodules of P, respectively.

a) The mapping b↦Pb\mathfrak{b}\mapsto P\mathfrak{b} is an isomorphism of ordered sets from BB\mathscr{B}_B to BP\mathscr{B}_P; the inverse isomorphism sends an (A,B)k(A,B)_k-sub-bimodule P′P' of P to the two-sided ideal of B consisting of the elements bb such that Pb⊂P′Pb\subset P'.

b) The mapping a↦aP\mathfrak{a}\mapsto \mathfrak{a}P is an isomorphism of ordered sets from BA\mathscr{B}_A to BP\mathscr{B}_P; the inverse isomorphism sends an (A,B)k(A,B)_k-sub-bimodule P′P' of P to the two-sided ideal of A consisting of the elements aa such that aP⊂P′aP\subset P'.

Indeed, let J be a left ideal of B and P′=P'= PJ. Then P′P' is an A-submodule of P and, by Example 1, the ideal J consists of the elements bb of B such that Pb⊂P′Pb\subset P'. Moreover, P′P' is an (A,B)k(A,B)_k-sub-bimodule of P if and only if J is a two-sided ideal of B. Thus a) follows from loc. cit.

Assertion b) follows from assertion a) applied to the invertible (Bo,Ao)k(B^o,A^o)_k-bimodule P.

Proposition 6 007A

Denote by BA\mathscr{B}_A and BB\mathscr{B}_B the sets of two-sided ideals of A and two-sided ideals of B.

a) There exists an isomorphism of ordered sets ff from BA\mathscr{B}_A to BB\mathscr{B}_B determined by the following property: if a\mathfrak{a} is a two-sided ideal of A and b\mathfrak{b} a two-sided ideal of B, then the relation f(a)=bf(\mathfrak{a}) =\mathfrak{b} is equivalent to aP=Pb\mathfrak{a}P = P\mathfrak{b}.

b) Suppose that the ring A is commutative, so that A can be identified with the center of B (VIII, p. 102, Corollary 1). The isomorphism f:BA→f:\mathscr{B}_A\rightarrow BB\mathscr{B}_B sends an ideal a\mathfrak{a} of A to the two-sided ideal BaB\mathfrak{a} of B, and we have a=A∩Ba\mathfrak{a}= A\cap B\mathfrak{a}.

Assertion a) follows from Example 2.

Now, suppose that A is commutative, and identify A with the center of B. Let a\mathfrak{a} be an ideal of A. Then BaB\mathfrak{a} is a two-sided ideal of B; we have PBa=aP\mathfrak{a}=\mathfrak{a}P and therefore f(a)=Baf(\mathfrak{a}) = B\mathfrak{a}. Let a′\mathfrak{a}' be the ideal A∩BaA\cap B\mathfrak{a} of A; it is contained in BaB\mathfrak{a} and contains a\mathfrak{a}, so Ba′B\mathfrak{a}' is equal to BaB\mathfrak{a}. Since ff is bijective, it follows that a′=a\mathfrak{a}'=\mathfrak{a}.

Example 3 007B

Let V be a left B-module. Then the correspondence given above sends the annihilator of the B-module V to the annihilator of the A-module P⊗BVP\otimes_BV. Indeed, denote the annihilator of the A-module P⊗BVP\otimes_BV by a\mathfrak{a} and that of the B-module V by b\mathfrak{b}. Let W be the (A,B)k(A,B)_k-sub-bimodule of P consisting of the elements such that p⊗v=0p\otimes v= 0 for every element vv of V. We have the inclusion Pb⊂WP\mathfrak{b}\subset W; conversely, for every p∈Wp\in W and q∈Qq\in Q, we have that θ(q⊗p)\theta (q\otimes p) belongs to b\mathfrak{b}. Hence, the element b\mathfrak{b} of D(Bs)D(B_s) corresponds to the element W of D(P). Likewise, a∈D(Ad)\mathfrak{a}\in D(A_d) corresponds to W.

Example 4 007C

For any two-sided ideal a\mathfrak{a} of A, denote the subset of Mn(A)\mathbf{M}_n(A) consisting of the matrices with entries in a\mathfrak{a} by Mn(a)\mathbf{M}_n(\mathfrak{a}). It is a two-sided ideal of Mn(A)\mathbf{M}_n(A). We have Mn(a)An=an=Ana\mathbf{M}_n(\mathfrak{a})A^n=\mathfrak{a}^n= A^n\mathfrak{a}. It follows from Proposition 6 that every two-sided ideal of Mn(A)\mathbf{M}_n(A) is of the form Mn(a)\mathbf{M}_n(\mathfrak{a}), where a\mathfrak{a} is a two-sided ideal of A.

Remark 007D

We keep the assumptions and notation above and suppose that the (B,A)k(B,A)_k-bimodule Q is the dual P∗P^* of the A-module P and that the isomorphisms λ\lambda and θ\theta are the canonical mappings Λ:P⊗BP∗→sAd\Lambda : P\otimes_BP^*\rightarrow_sA_d and Θ : P∗⊗AP→sBdP^*\otimes_AP\rightarrow_sB_d (VIII, p. 101, Theorem 1). Since the A-module P is projective and finitely generated, we have a canonical isomorphism ϑM:P∗⊗AM→\vartheta_M: P^*\otimes_AM\rightarrow HomA(P,M)_A(P,M) for every A-module M (II, §4, No. 2, p. 271, Corollary). We leave it to the reader to reformulate the results of Subsections 3 and 4 by replacing the construction M↦Q⊗AMM\mapsto Q\otimes_AM with the construction M↦M\mapsto HomA(P,M)_A(P,M).

5. Ordered Sets of Submodules

In this subsection, A and B are kk-algebras, M is a left A-module, and V is a left B-module. We denote by D(M) (resp. D(V)) the set of submodules of M (resp. of V), ordered by inclusion. We assume given an isomorphism of ordered sets φ:D(V)→D(M)\varphi : D(V)\rightarrow D(M).

By Morita’s theorem (VIII, p. 103, Theorem 2), we obtain such an isomorphism in the following situation: P is an invertible (A,B)k(A,B)_k-bimodule, M is the A-module P⊗BVP\otimes_BV, and for every submodule W of V,φ(W)V,\varphi (W) is the canonical image of P⊗BWP\otimes_BW in M.

Some properties of the module M, or of its submodules, can be expressed in terms of the ordered set D(M): they are listed in Tables I and II.

The module M is the direct sum of a family (Mi)i∈I(M_i)_{i\in I} of submodules if and only if we have M=∑i∈IMiM =\sum_{i\in I}M_i and Mi∩∑j≠iMj=0M_i\cap \sum_{j\not=i}M_j= 0 for every i∈Ii\in I. This remark and an examination of Table I give the following result.

Proposition 7 007E

a) We have φ(0)=0\varphi (0) = 0 and φ(V)=M\varphi (V) = M.

b) Let (Vi)i∈I(V_i)_{i\in I} be a family of submodules of V. We have

φ(∑i∈IVi)=∑i∈Iφ(Vi),φ(⋂i∈IVi)=⋂i∈Iφ(Vi) \varphi (\sum_{i\in I}V_i)=\sum_{i\in I}\varphi (V_i),\varphi (\bigcap_{i\in I}V_i)=\bigcap_{i\in I}\varphi (V_i)

Submodules of M Ordered set D(M)

Zero submodule Smallest element of D(M)

Submodule M Greatest element of D(M)

⋂i∈IMi\bigcap_{i\in I}M_i Greatest lower bound infi∈IMi_{i\in I}M_i

∑i∈IMi\sum_{i\in I}M_i Least upper bound supi∈IMi_{i\in I}M_i

Supplementary submodules inf(M′,M′′',M'') =0= 0, sup(M′,M′′',M'') =M= M

Simple submodule of M Minimal element of D(M) −{0}-\{0\}

Maximal submodule of M Maximal element of D(M) −{M}-\{M\}

Socle S(M)\mathscr{S}(M) of M Least upper bound in D(M) of the set of

minimal elements of D(M)−{0}-\{0\}

∗*Radical R(M)\mathfrak{R}(M) of M Greatest lower bound in D(M) of the set

∗ *

(VIII, p. 151) of maximal elements of D(M)−{M}-\{M\}

Table I.

Properties of the module M Properties of D(M)

M is Noetherian. The ordered set D(M) is Noetherian (Set

Theory, III, §6, No. 5, p. 190).

M is Artinian. The set D(M), ordered by ⊃\supset, is

Noetherian.

M is indecomposable. We have M ≠0\not= 0, and there are no

two nonzero elements M′M' and M′′M'' of

D(M) satisfying inf(M′,M′′',M'') =0= 0,

sup(M′,M′′',M'') =M= M.

M is finitely generated. For every family (Mi)i∈I(M_i)_{i\in I} in D(M) with

upper bound M, there exists a finite

subset J of I such that M = supj∈IMj_{j\in I}M_j. M is simple. Card(D(M)) = 2.

M is semisimple. M is the least upper bound, in D(M),

of the set of minimal elements of

D(M)−{0}-\{0\}.

Table II.

c) The B-module V is the direct sum of the family (Vi)i∈I(V_i)_{i\in I} of submodules if and only if M is the direct sum of the family (φ(Vi))i∈I(\varphi (V_i))_{i\in I}.

Let V′V' and V′′V'' be submodules of V such that V′V' is contained in V′′V''; set M′=φ(V′)M'=\varphi (V') and M′′=φ(V′′)M''=\varphi (V''), so that M′′M'' contains M′M'. Denote by [V′,V′′][V',V''] the interval in D(V) consisting of the submodules W of V such that we have V′⊂W⊂V′′V'\subset W\subset V'', and define the interval [M′,M′′][M',M''] in D(M) likewise. The mapping W↦W/V′W\mapsto W/V' is an isomorphism of ordered sets from [V′,V′′][V',V''] to D(V′′/V′)D(V''/V'); we define an isomorphism of ordered sets from [M′,M′′][M',M''] to D(M′′/M′)D(M''/M') likewise. Since φ\varphi sends the interval [V′,V′′][V',V''] to [M′,M′′][M',M''], it defines an isomorphism φ\varphi of ordered sets from D(V′′/V′)D(V''/V') to D(M′′/M′)D(M''/M'). We deduce the following proposition from this and Tables I and II.

Proposition 8 007F

a) Let V′V' and V′′V'' be submodules of V such that V′′V'' contains V′V'. The B-module V′′/V′V''/V' is simple if and only if the A-module φ(V′′)/φ(V′)\varphi (V'')/\varphi (V') is simple.

b) If V′V' is a simple submodule, maximal submodule, or direct factor of V, then φ(V′)\varphi (V') is, respectively, a simple submodule, maximal submodule, or direct factor of M.

c) The isomorphism φ\varphi transforms the socle S(V)\mathscr{S}(V) of V into the socle S(M)\mathscr{S}(M) of M∗M*and the radical (VIII, p. 151, Definition 1)R(V)\mathfrak{R}(V) of V into the radical R(M)\mathfrak{R}(M) of M.∗.*

d) Let (Vi)0⩽i⩽n(V_i)_{0\leqslant i\leqslant n} be a finite sequence of submodules of V. It is a Jordan– Hölder series of V if and only if (φ(Vi)0⩽i⩽n)(\varphi (V_i)_{0\leqslant i\leqslant n}) is a Jordan–Hölder series of M.

Lemma 2 007G

Let H and H′H' be submodules of V such that H∩H′=0H\cap H'= 0. The B-modules H and H′H' are isomorphic if and only if the A-modules φ(H)\varphi (H) and φ(H′)\varphi (H') are isomorphic.

We identify H+H′H+H' with the product H×H′H\times H'. The graph of an isomorphism from H to H′H' is a submodule H′′H'' of V satisfying

H∩H′′=H′∩H′′=0,H+H′=H+H′′=H′+H′′(17) H\cap H''= H'\cap H''= 0,H + H'= H + H''= H'+ H'' \tag{17}

conversely, every submodule that has these properties is the graph of an isomorphism from H to H′H'. By Proposition 7, the relation H∩H′=0H\cap H'= 0 is equivalent to φ(H)∩φ(H′)=0\varphi (H)\cap \varphi (H') = 0, and relations (17) are equivalent to the relations

φ(H)∩φ(H′′)=φ(H′)∩φ(H′′)=0 \varphi (H)\cap \varphi (H'') =\varphi (H')\cap \varphi (H'') = 0
φ(H)+φ(H′)=φ(H)+φ(H′′)=φ(H′)+φ(H′′) \varphi (H) +\varphi (H') =\varphi (H) +\varphi (H'') =\varphi (H') +\varphi (H'')

the lemma follows.

Proposition 9 00R6

Let S be a simple submodule of V and T the simple submodule φ(S)\varphi (S) of M. If VSV_S denotes the isotypical component of V of type S and MTM_T the isotypical component of M of type T, then we have φ(VS)=MT\varphi (V_S) = M_T.

Every simple submodule S′S' of V distinct from S satisfies S′∩S=0S'\cap S = 0. It is therefore isomorphic to S if and only if φ(S′)\varphi (S') is isomorphic to T (Lemma 2). Now, VSV_S is the sum of the simple submodules of V isomorphic to S, and MTM_T is the sum of the simple submodules of M isomorphic to T. Proposition 9 therefore follows immediately from Propositions 7 and 8.

Proposition 10 007H

a) The B-module V is Artinian (resp. Noetherian, indecomposable, simple, finitely generated) if and only if M is.

b) The B-module V has finite length if and only if the A-module M has finite length, and we then have longB(V)=_B(V) = longA(M)_A(M).

c) The B-module V is semisimple (resp. isotypical) if and only if the A-module M is semisimple (resp. isotypical). If this is the case, then we have longB(V)=_B(V) = longA(M)_A(M).

Assertion a) follows from the second property listed in Table II.

Assertion b) follows from Proposition 8, d).

The module V is semisimple if and only if it is equal to its socle S(V)\mathscr{S}(V); it is isotypical if and only if there exists a simple submodule S of V such that V=VSV = V_S. Assertion c) therefore follows from Propositions 7, c), 8, c), and 9 (VIII, p. 106 and 109).

6. Other Properties Preserved by the Morita Correspondence

Let A and B be Morita equivalent kk-algebras and P an invertible (A,B)k(A,B)_k-bimodule.

Proposition 11 007I

Let

(E)V′−→fV−→gV′′ (\mathscr{E})V'-\overset{f}{\rightarrow}V-\overset{g}{\rightarrow}V''

be a diagram of B-modules and B-linear mappings, and let

(P⊗E)P⊗BV′⟶1P⊗fP⊗BV⟶1P⊗gP⊗BV′′ (P\otimes \mathscr{E})P\otimes_BV'\longrightarrow^{1_P\otimes f}P\otimes_BV\longrightarrow^{1_P\otimes g}P\otimes_BV''

be the corresponding diagram of A-modules. Then (E)(\mathscr{E}) is an exact sequence if and only if (P⊗E)(P\otimes \mathscr{E}) is an exact sequence.

Suppose that the sequence (E)(\mathscr{E}) is exact. Since the right B-module P is projective, the sequence (P⊗E)(P\otimes \mathscr{E}) is exact (II, §3, No. 6, p. 251, Proposition 5 and II, §3, No. 7, p. 257, Corollary 6).

Conversely, suppose that the sequence (P⊗E)(P\otimes \mathscr{E}) is exact. Let Q be a (B,A)k(B,A)_k-bimodule, inverse to P, and θ:Q⊗AP→sBd\theta : Q\otimes_AP\rightarrow_sB_d an isomorphism. Consider the commutative diagram

Q⊗AP⊗BV′1Q⊗1P⊗f/Q\otimes_AP\otimes_B{V'}^{1_Q\otimes 1_P\otimes f}// Q⊗AP⊗BV1Q⊗1P⊗g/Q\otimes_AP\otimes_BV^{1_Q\otimes 1_P\otimes g}// Q⊗AP⊗BV′′Q\otimes_AP\otimes_BV''

θ⊗1V′θ⊗1Vθ⊗1V′′\theta \otimes 1_{V'}\theta \otimes 1_V\theta \otimes 1_{V''}

V′fV'f // V gg // V′′V''.

Since Q is a projective A-module and the sequence (P⊗E)(P\otimes \mathscr{E}) is exact, the first line of this diagram is an exact sequence. Since the vertical arrows are isomorphisms, the second line is also exact.

Corollary 007J

Let f:V→Wf: V\rightarrow W be a B-linear mapping. Then ff is injective (resp. surjective) if and only if 1P⊗f1_P\otimes f is.

Proposition 12 007K

Let V be a left B-module. The B-module V is projective (resp. generating, faithful, ∗*injective, finitely presented ) if and only if the

∗ *

A-module P⊗BVP\otimes_BV is.

a) Suppose that V is projective. There exists a set I such that V is isomorphic to a direct factor submodule of Bs(I)B^{(I)}_s . The A-module P⊗BVP\otimes_BV is then isomorphic to a direct factor submodule of P(I)P^{(I)}; since P is a projective A-module, the same holds for P⊗BVP\otimes_BV.

b) Suppose that the B-module V is generating. Let M be an A-module. There exists a B-module W such that M is isomorphic to P⊗BWP\otimes_BW. By Theorem 1 of VIII, p. 80, there exist a set I and a surjection φ:V(I)→W\varphi : V^{(I)}\rightarrow W. By the corollary, the mapping 1P⊗φ1_P\otimes \varphi from P⊗(V(I)P\otimes (V^{(I}) to P⊗AWP\otimes_AW is surjective, which gives a surjection (P⊗V)(I)→M(P\otimes V)^{(I)}\rightarrow M. By Theorem 1 of VIII, p. 80, P⊗VP\otimes V is a generating A-module.

c) The B-module V is faithful if and only if its annihilator is reduced to 0. Assertion c) therefore follows from Example 3 of VIII, p. 105.

∗*d) Suppose that V is injective. By the remark of VIII, p. 106, the A-module P⊗BVP\otimes_BV is isomorphic to HomB(Q,V)_B(Q,V), where Q is a (B,A)k(B,A)_k-bimodule inverse to A. Since the A-module Q is projective, hence flat (X, §1, no3^o3, p. 9, exemple 1), the A-module HomB(Q,V)_B(Q,V) is injective by X, §1, no8^o8, p. 18, proposition 11.

e) Suppose that V admits a finite presentation L1→L0→V→0L_1\rightarrow L_0\rightarrow V\rightarrow 0 (X, §1, no4^o4, p. 10). By taking the tensor product with P, we deduce an exact sequence of A-modules N1′−→uN0′→P⊗BV→0N'_1-\overset{u}{\rightarrow}N'_0\rightarrow P\otimes_BV\rightarrow 0, where N1′N'_1 and N0′N'_0 are projective and finitely generated (Proposition 11 and a)). Let N0′′N''_0 be a finitely generated A-module such that the module N0=N0′⊕N0′′N_0= N'_0\oplus N''_0 is free and finitely generated, and let u′:N1′⊕N0′′→N0u': N'_1\oplus N''_0\rightarrow N_0 be the homomorphism (u,1N0′′)(u,1_{N''_0}); then P⊗BVP\otimes_BV can be identified with the cokernel of u′u'. Let N1N_1 be a finitely generated free A-module and p:N1→N1′⊕N0′′p: N_1\rightarrow N'_1\oplus N''_0 a surjective homomorphism; the sequence N1⟶u′∘pN0→P⊗BV→0N_1\longrightarrow^{u'\circ p}N_0\rightarrow P\otimes_BV\rightarrow 0 is a finite presentation of the A-module P⊗BV.∗P\otimes_BV.*

f) Suppose that the A-module P⊗BVP\otimes_BV is projective (resp. generating, faithful, ∗*injective, finitely presented ). By applying the above (interchanging

∗ *

the roles of A and B and of P and Q), we see that the B-module Q⊗AP⊗BVQ\otimes_AP\otimes_BV also has this property. The same is therefore true for the B-module V, which is isomorphic to it.

Corollary 007L

The ring A is left Artinian (resp. left Noetherian) if and only if the ring B is.

Because of the isomorphism between the ordered set of left ideals of B and the set of A-submodules of P, the ring B is left Artinian (resp. left Noetherian) if and only if the A-module P is Artinian (resp. Noetherian). However, by Theorem 1 of VIII, p. 101, the A-module P is generating and finitely generated; in particular, AsA_s is isomorphic to a direct factor of PnP^n for an integer n⩾1n\geqslant 1. Consequently, P is Artinian (resp. Noetherian) if and only if A is left Artinian (resp. left Noetherian).

7. Morita Equivalence of Algebras

Proposition 13 007M

a) If two kk-algebras are Morita equivalent, then their centers are isomorphic kk-algebras.

b) Two commutative kk-algebras are Morita equivalent if and only if they are isomorphic.

c) Two kk-algebras that are fields are Morita equivalent if and only if they are isomorphic.

d) For i=1,2i= 1,2, let AiA_i and BiB_i be Morita equivalent kk-algebras and PiP_i an invertible (Ai,Bi)k(A_i,B_i)_k-bimodule. Set A=A1⊗kA2,B=B1⊗kB2A = A_1\otimes_kA_2,B = B_1\otimes_kB_2, and P=P1⊗kP2P = P_1\otimes_kP_2. The kk-algebras A and B are Morita equivalent, and P is an invertible (A,B)k(A,B)_k-bimodule.

e) If A and B are Morita equivalent kk-algebras and k′k' is a commutative kk-algebra, then the k′k'-algebras A(k′)A_{(k')} and B(k′)B_{(k')} are Morita equivalent.

Assertion a) follows from Corollary 1 of VIII, p. 102, and implies b).

Let K and L be kk-algebras that are fields, and let P be an invertible (K,L)k(K,L)_k-bimodule. The right L-vector space P is a simple module (VIII, p. 104), hence of dimension 1, so the kk-algebras EndL(P)_L(P) and L are isomorphic. By VIII, p. 101, Theorem 1, the mapping a↦aPa\mapsto a_P from K to EndL(P)_L(P) is an isomorphism. Therefore, the fields K and L are isomorphic over kk; assertion c) follows.

Under the assumptions of d), let Qi(i=1,2)Q_i(i= 1,2) be a (Bi,Ai)k(B_i,A_i)_k-bimodule inverse to PiP_i. Denote the (B,A)k(B,A)_k-bimodule Q1⊗kQ2Q_1\otimes_kQ_2 by Q. Consider the canonical kk-linear isomorphism (P1⊗kP2)⊗k(Q1⊗kQ2)→(P1⊗kQ1)⊗k(P_1\otimes_kP_2)\otimes_k(Q_1\otimes_kQ_2)\rightarrow (P_1\otimes_kQ_1)\otimes_k (P2⊗kQ2)(P_2\otimes_kQ_2); when passing to the quotient, it defines a morphism

(P1⊗kP2)⊗B(Q1⊗kQ2)→(P1⊗B1Q1)⊗k(P2⊗B2Q2) (P_1\otimes_kP_2)\otimes_B(Q_1\otimes_kQ_2)\rightarrow (P_1\otimes_{B_1}Q_1)\otimes_k(P_2\otimes_{B_2}Q_2)

that is (A(A, A)-linear. Conversely, the inverse isomorphism (P1⊗kQ1)⊗k(P2⊗k(P_1\otimes_kQ_1)\otimes_k(P_2\otimes_k Q2)→(P1⊗kP2)⊗k(Q1⊗kQ2)Q_2)\rightarrow (P_1\otimes_kP_2)\otimes_k(Q_1\otimes_kQ_2) defines an (A(A, A)-linear morphism (P1⊗B1(P_1\otimes_{B_1} Q1)⊗k(P2⊗B2Q2)→(P1⊗kP2)⊗B(Q1⊗kQ2)Q_1)\otimes_k(P_2\otimes_{B_2}Q_2)\rightarrow (P_1\otimes_kP_2)\otimes_B(Q_1\otimes_kQ_2). These two morphisms are each other’s inverses and are thus isomorphisms. Since the (Ai,Ai(A_i,A_i)-bimodule Pi⊗BiQiP_i\otimes_{B_i}Q_i is isomorphic to AiA_i, we obtain an (A(A, A)-linear isomorphism P⊗BP\otimes_B Q→AQ\rightarrow A. We likewise obtain a (B(B, B)-linear isomorphism Q⊗AP→BQ\otimes_AP\rightarrow B, which completes the proof of d).

Under the assumptions of e), let P be an invertible (A,B)k(A,B)_k-bimodule; then P(k′)P_{(k')} is an invertible (A(k′),B(k′))k′(A_{(k')},B_{(k')})_{k'}-bimodule.

Let A be a kk-algebra, and let ee be an idempotent in A. The set eAeeAe, endowed with the addition, multiplication, and action of kk induced by those of A, is a kk-algebra with unit element ee.

Proposition 14 007N

Let A and B be kk-algebras. Then A and B are Morita equivalent if and only if there exist an integer n⩾1n\geqslant 1 and a square matrix e=(eij)e= (e_{ij}) in Mn(B)\mathbf{M}_n(B) satisfying the following conditions:

(i) We have e2=ee^2=e.

(ii) The two-sided ideal of B generated by the elements eije_{ij} is equal to B.

(iii) The kk-algebra A is isomorphic to eMn(B)ee\mathbf{M}_n(B)e.

If conditions (i) and (ii) are satisfied, then the (eMn(B)e,B)k(e\mathbf{M}_n(B)e,B)_k-bimodule eBdneB^n_d is invertible.

In view of Theorem 1 (VIII, p. 101), the kk-algebra A is Morita equivalent to B if and only if it is isomorphic to the endomorphism algebra of a projective, finitely generated, and generating right B-module. The proposition therefore follows from Lemmas 3 and 4 below.

Lemma 3 007O

A right B-module P is projective, finitely generated, and generating if and only if there exist an integer n⩾0n\geqslant 0 and an idempotent e=(eij)e= (e_{ij}) in Mn(B)\mathbf{M}_n(B) with the following properties:

(i) The B-module P is isomorphic to eBdneB^n_d.

(ii) The two-sided ideal of B generated by the elements eije_{ij} is equal to B.

Let P be a right B-module. Then P is projective and finitely generated if and only if it is isomorphic to a direct factor submodule of a B-module of the form BdnB^n_d, where nn is an integer ⩾0\geqslant 0 (II, §2, No. 2, p. 232, Corollary 1). If we identify the kk-algebras Mn(B)\mathbf{M}_n(B) and End(Bdn^n_d), then this means that there exists an idempotent ee in Mn(B)\mathbf{M}_n(B) such that P is isomorphic to eBdneB^n_d.

The B-module P is generating if and only if its trace ideal τ(P)\tau (P) is equal to B, that is, τ(eBdn)=B\tau (eB^n_d) = B (VIII, p. 80, Theorem 1). Let x1,...,xnx_1, . . . , x_n be the elements of BdnB^n_d corresponding to the columns of the matrix ee, and let xi∗x^*_i (for 1⩽i⩽n1\leqslant i\leqslant n) be the linear form (b1,...,bn)↦bi(b_1, . . . , b_n)\mapsto b_i on eBdneB^n_d. The family (x1,...,xn)(x_1, . . . , x_n) generates the B-module eBdneB^n_d, and the family (x1∗,...,xn∗)(x^*_1, . . . , x^*_n) generates its dual. Now, we have ⟨xi∗,xj⟩=eij\langle x^*_i, x_j\rangle =e_{ij}, so τ(eBdn)\tau (eB^n_d) is the two-sided ideal of B generated by the eije_{ij}. This proves Lemma 3.

Lemma 4 007P

Let V be a B-module and E the endomorphism kk-algebra of V. Let ee be a projector of V and P the image of ee. The mapping that sends v∈eEev\in eEe to the endomorphism x↦v(x)x\mapsto v(x) of the B-module P is an isomorphism of kk-algebras from eEeeEe to EndB(P)_B(P).

Denote the mapping described in the statement by φ:eEe→\varphi :eEe\rightarrow EndB(P)_B(P); it is a homomorphism of kk-algebras. Let u∈u\in EndB(P)_B(P). Denote by vv the endomorphism of V defined by v(x)=u(e(x))v(x) =u(e(x)) for x∈Vx\in V. We have (eve)(x)=(eve)(x) = u(x)u(x) for x∈Px\in P, that is, φ(eve)=u\varphi (eve) =u. Consequently, φ\varphi is surjective. Let ww be an element of the kernel of φ\varphi; the restrictions of ww to the kernel and to the image of ee are zero, so ww is zero, which proves that φ\varphi is injective.

Example 1 007Q

Let A be a kk-algebra and ee an idempotent in A such that AeA=AAeA = A. The kk-algebra eAeeAe can be identified with the endomorphism kk-algebra of the submodule eAdeA_d of AdA_d (Lemma 4). Since AeA=AAeA = A, it follows from Proposition 14 that eAdeA_d is an invertible (eAe,A)k(eAe,A)_k-bimodule, so that the kk-algebras eAeeAe and A are Morita equivalent. Moreover, Morita’s theorem (VIII, p. 103) implies the following results:

a) Let M and N be left A-modules. Every eAeeAe-linear mapping from eMeM to eNeN extends uniquely to an A-linear mapping from M to N.

b) Every left module over the kk-algebra eAeeAe is isomorphic to a module of the form eMeM, where M is a left A-module.

Example 2 007R

Let A be a kk-algebra and n⩾1n\geqslant 1 an integer. We identify the matrix algebra Mn(A)\mathbf{M}_n(A) with the endomorphism algebra of the right A-module AdnA^n_d. We have seen that A and Mn(A)\mathbf{M}_n(A) are Morita equivalent. For every left A-module M, identify Adn⊗AMA^n_d\otimes_AM with MnM^n. The algebra Mn(A)\mathbf{M}_n(A) then has a left action on MnM^n, and we have

(a⋅m)i=∑j=1naijmj (a\cdot m)_i=\sum_{j=1}^na_{ij}m_j

for a=(aij)a= (a_{ij}) in Mn(A)\mathbf{M}_n(A) and m=(mi)m= (m_i) in MnM^n. Morita’s theorem implies the following results:

a) Every left module over the algebra Mn(A)\mathbf{M}_n(A) is isomorphic to a module of the form MnM^n, where M is a left A-module.

b) Let M be a left A-module. The mapping N↦NnN\mapsto N^n is a bijection from the set of A-submodules of M to the set of Mn\mathbf{M}_n(A)-submodules of MnM^n.

c) Let M and N be left A-modules. For an A-linear mapping g:M→Ng: M\rightarrow N, let gng_n be the mapping (mi)↦(g(mi))(m_i)\mapsto (g(m_i)) from MnM^n to NnN^n. Then the mapping g↦gng\mapsto g_n is a bijection from HomA(M,N)_A(M,N) to HomMn(A)(Mn,Nn)_{\mathbf{M}_n(A)}(M^n,N^n).

d) Let M be a left A-module. The module MnM^n over the ring Mn(A)\mathbf{M}_n(A) is indecomposable (resp. semisimple, simple, Artinian, Noetherian, finitely generated) if and only if the A-module M is.

Example 3 007S

Let A be a principal ideal domain and L a nonzero, finitely generated, free A-module. Let B be the endomorphism ring of L; then L is an invertible (A,B)Z(A,B)_{\mathbf{Z}}-bimodule, and the rings A and B are Morita equivalent. By Morita’s theorem, Proposition 10, a) (VIII, p. 109), and the structure theorem for finitely generated A-modules (VII, §4, No. 4, p. 19, Theorem 2), every finitely generated B-module is isomorphic to ⊕i=1m(L/aiL)\oplus^m_{i=1}(L/\mathfrak{a}_iL), where mm is a natural number and the ai\mathfrak{a}_i are ideals of A satisfying a1⊂a2⊂⋯⊂am\mathfrak{a}_1\subset \mathfrak{a}_2\subset \cdots \subset \mathfrak{a}_m and am≠A\mathfrak{a}_m\not= A; the integer mm and the ideals ai\mathfrak{a}_i are uniquely determined. By Proposition 6 of VIII, p. 105, every two-sided ideal of B is of the form dBdB, where dd is an element of A.

Exercises

See the exercises for § 6.

Transcribed from the printed edition, 2023, Springer Nature. Not translated.