Bourbaki

§ 13. Absolutely semisimple algebras

1. Absolutely Semisimple Modules

Definition 1 00F9

Let K be a commutative field and A a K-algebra. We call an A-module M absolutely semisimple if the A(L)A_{(L)}-module M(L)M_{(L)} is semisimple for every extension L of K.

Every absolutely semisimple module is semisimple. Conversely, if the field K is perfect, then every semisimple A-module that is finite-dimensional over K is absolutely semisimple (VIII, p. 222, Corollary 1, a)) because every extension of a perfect field is separable (V, §7, No. 1, p. 36, Proposition 2).

Proposition 1 00FA

Let K be a commutative field and A a K-algebra.

a) Every direct sum of absolutely semisimple A-modules is an absolutely semisimple A-module. Every submodule and quotient module of an absolutely semisimple module is absolutely semisimple.

b) Let M be an A-module and let L be an extension of the field K. The A-module M is absolutely semisimple if and only if the A(L)A_{(L)}-module M(L)M_{(L)} is absolutely semisimple.

Assertion a) follows from the analogous assertion for semisimple modules (VIII, p. 56, Corollaries 1 and 3).

Suppose that the A-module M is absolutely semisimple, and let L′L' be an extension of L. As an A(L′)A_{(L')}-module, L′⊗LM(L)L'\otimes_LM_{(L)} is isomorphic to M(L′)M_{(L')} (II, §5, No. 1, p. 273, Proposition 2); it is therefore a semisimple module. This proves that M(L)M_{(L)} is absolutely semisimple.

Conversely, suppose that M(L)M_{(L)} is absolutely semisimple. Let L′L' be an extension of K. There exists a composite extension (Ω,u,v)(\Omega , u, v) of L and L′L' (V, §2, No. 4, p. 13, Corollary); we identify L and L′L' with subextensions of Ω. The A(Ω)A_{(\Omega )}-module M(Ω)M_{(\Omega )} is isomorphic to (M(L))(Ω)(M_{(L)})_{(\Omega )}; it is therefore semisimple. But M(Ω)M_{(\Omega )} is also isomorphic to (M(L′))(Ω)(M_{(L')})_{(\Omega )}, and Proposition 8, a) of VIII, p. 222 implies that M(L′)M_{(L')} is semisimple. So M is absolutely semisimple.

Proposition 2 00FB

Let K be a commutative field, A a K-algebra, and M an A-module that is finite-dimensional over K. The following properties are equivalent:

(i) The A-module M is absolutely semisimple.

(ii) There exists an extension P of K that is a perfect field such that the A(P)A_{(P)}-module M(P)M_{(P)} is semisimple.

(iii) The A-module M is semisimple, and the center of its commutant is an étale algebra over the field K.

It is clear that (i) implies (ii). Assume that (ii) holds; then the A(P)A_{(P)}-module M(P)M_{(P)} is absolutely semisimple by Corollary 1, a) of VIII, p. 222. By Proposition 1, b), M is then absolutely semisimple. Therefore, (ii) implies (i).

Suppose that M is semisimple. Let Z be the center of the commutant of M; it is a commutative algebra of finite degree over the field K. If L is an extension of K, then it follows from Proposition 8, b) of VIII, p. 222 that the A(L)A_{(L)}-module M(L)M_{(L)} is semisimple if and only if the ring Z(L)Z_{(L)} is reduced. The equivalence of (i) and (iii) therefore follows from Theorem 4 of V, §6, No. 7, p. 34.

2. Algebras over Separably Closed Fields

Lemma 1 00FC

Let D be a field and Z its center. Denote the characteristic exponent of Z by pp. Suppose that for every element aa of D, there exists an integer m⩾0m\geqslant 0 such that apma^{p^m} belongs to Z. Then the field D is commutative.

If p=1p= 1, then D = Z. We therefore assume p>1p >1.

Let us give a proof by contradiction and suppose that D is not commutative. Let aa be an element of D- Z and qq a power of pp such that aqa^q belongs to Z. Denote the identity mapping on D by I and the inner automorphism x↦axa−1x\mapsto axa^{-1} from D to D associated with aa by σ\sigma; we have σq=I\sigma^q= I because aqa^q belongs to Z. We have σ−I≠0\sigma -I\not= 0 because aa does not belong to Z, and we have (σ−I)q=σq−I=0(\sigma -I)^q=\sigma^q-I = 0 because Z has characteristic pp. Let ff be the greatest natural number such that (σ−I)f≠0(\sigma -I)^f\not= 0; we have f⩾1f\geqslant 1. Choose an element cc of D such that (σ−I)f(c)≠0(\sigma -I)^f(c)\not= 0, and set

x=(σ−I)f−1(c),y=(σ−I)(x)=(σ−I)f(c) x= (\sigma -I)^{f-1}(c),y= (\sigma -I)(x) = (\sigma -I)^f(c)

By construction, we have y≠0y\not= 0 and σ(y)=y\sigma (y) =y; if we set z=y−1xz=y^{-1}x, then it follows that

σ(z)=σ(y)−1σ(x)=y−1(y+x)=1+z \sigma (z) =\sigma (y)^{-1}\sigma (x) =y^{-1}(y+x) = 1 +z

and therefore σ(zpj)=1+zpj\sigma (z^{p^j}) = 1 +z^{p^j} for every natural number jj. Choose an integer m⩾0m\geqslant 0 such that zpmz^{p^m} belongs to the center Z of D; we have

zpm=azpma−1=σ(zpm)=1+zpm z^{p^m}=az^{p^m}a^{-1}=\sigma (z^{p^m}) = 1 +z^{p^m}

This contradiction implies Lemma 1.

Proposition 3 00FD

Let K be a separably closed field (V, §7, No. 8, p. 45, Definition 4), and let D be an algebra of finite degree over K that is a field. Then D is commutative.

Denote the characteristic exponent of K by pp. Let aa be an element of D. The ring K[a]K[a] is an algebraic extension of K (V, §3, No. 1, p. 17, Corollary 1). Since the field K is separably closed, it follows from V, §7, No. 7, p. 44, Proposition 13 that the algebra K[a]K[a] is a pp-radical extension of K. Hence there exists an integer m⩾0m\geqslant 0 such that apma^{p^m} belongs to K. By Lemma 1, the field D is consequently commutative.

Corollary 00FE

Let K be a separably closed field and A a semisimple algebra of finite degree over K. Then there exist an integer r⩾0r\geqslant 0, strictly positive integers n1,...,nrn_1, . . . , n_r, and extensions K1,...,KrK_1, . . . ,K_r of K of finite degree such that A is isomorphic to the algebra ∏i=1rMni(Ki)\prod_{i=1}^r\mathbf{M}_{n_i}(K_i).

By the structure theorem for semisimple algebras (VIII, p. 135, Theorem 1), A is isomorphic to an algebra ∏i=1rMni(Di)\prod^r_{i=1}\mathbf{M}_{n_i}(D_i), where rr is an integer ⩾0,n1,...,nr\geqslant 0,n_1, . . . , n_r are strictly positive integers, and D1,...,DrD_1, . . . ,D_r are K-algebras of finite degree that are fields. Since the field K is separably closed, each field DiD_i is commutative by Proposition 3; the corollary follows.

3. Absolutely Semisimple Algebras

Definition 2 00FF

Let K be a commutative field. We say that a K-algebra A is absolutely semisimple if the ring A(L)A_{(L)} is semisimple for every extension L of K.

An absolutely semisimple algebra is semisimple. The K-algebra A is absolutely semisimple if and only if the A-module AsA_s is absolutely semisimple. By Proposition 1 of VIII, p. 229, we therefore obtain the following result: if L is an extension of K, then the L-algebra A(L)A_{(L)} is absolutely semisimple if and only if the K-algebra A is absolutely semisimple.

Theorem 1 00FG

Let K be a commutative field and A a K-algebra. The following properties are equivalent:

(i) The K-algebra A is absolutely semisimple.

(ii) The algebra A has finite degree over K, and there exists an extension P of K that is a perfect field, for which the P-algebra A(P)A_{(P)} is semisimple.

(iii) The K-algebra A is semisimple and has finite degree over K, and its center is an étale K-algebra.

(iv) There exists a finite family (ni,Di)i∈I(n_i,D_i)_{i\in I}, where nin_i is a strictly positive integer and DiD_i a K-algebra of finite degree that is a field, such that the center ZiZ_i of DiD_i is a separable extension of K for every i∈Ii\in I and A is isomorphic to the product of the matrix rings Mni(Di)\mathbf{M}_{n_i}(D_i).

(v) There exist an extension L of K and a finite family of integers (ni)i∈I(n_i)_{i\in I} such that the L-algebra A(L)A_{(L)} is isomorphic to the algebra ∏i∈IMni(L)\prod_{i\in I}\mathbf{M}_{n_i}(L).

(vi) There exist a Galois extension L of K of finite degree and a finite family of integers (ni)(n_i) such that A(L)A_{(L)} is isomorphic to the product of the matrix algebras Mni(L)\mathbf{M}_{n_i}(L).

We first prove the implications (v) ⇒\Rightarrow (iv) ⇒\Rightarrow (iii) ⇒\Rightarrow (ii) ⇒\Rightarrow (i). Denote the center of A by Z.

If property (v) holds, then the L-algebra A(L)A_{(L)} is semisimple and has finite degree over L, and its center (isomorphic to Z(L)Z_{(L)}, III, p. 41, Corollary) is isomorphic to LrL^r for some integer r⩾0r\geqslant 0. By Corollary 2, a) of VIII, p. 222, the algebra A is semisimple and has finite degree over K. It is therefore isomorphic to a finite product of rings ∏i∈IMni(Di)\prod_{i\in I}\mathbf{M}_{n_i}(D_i) with ni⩾1n_i\geqslant 1 for every i∈Ii\in I, where the field DiD_i is an algebra of finite degree over K. The center ZiZ_i of DiD_i is a commutative field that is an extension of K, and Z is isomorphic to ∏i∈IZi\prod_{i\in I}Z_i. Therefore, Z(L)Z_{(L)} is isomorphic, on the one hand, to ∏i∈I(Zi)(L)\prod_{i\in I}(Z_i)_{(L)} and, on the other hand, to LrL^r. In other words, the algebra ∏i∈IZi\prod_{i\in I}Z_i is étale over the field K (V, §6, No. 3, p. 28, Definition 1), and each of the extensions ZiZ_i is separable over K (V, §6, No. 4, p. 31, Corollary, and V, §7, No. 1, p. 42). So (v) implies (iv).

If property (iv) holds, then it is clear that A is a semisimple algebra and has finite degree over K. Its center Z is isomorphic to the product ∏i∈IZi\prod_{i\in I}Z_i of separable extensions of K of finite degree; it is therefore an étale algebra (loc. cit.). So (iv) implies (iii).

The implications (iii) ⇒\Rightarrow (ii) ⇒\Rightarrow (i) follow from Proposition 2 of VIII, p. 230 applied to the A-module AsA_s.

Lemma 2 00FH

Let L be an algebraically closed field and D a field containing L in its center. If D is distinct from L, then there exists an extension L′L' of L such that the ring D⊗LL′D\otimes_LL' is not left Artinian.

Let xx be an element of D - L; since L is algebraically closed, the extension L′=L(x)L'= L(x) of L is not algebraic and xx is transcendent over L. The ring B=L′⊗LL′B = L'\otimes_LL' is then an integral domain by Proposition 5 of V, §17, No. 4, p. 141.

The element y=x⊗1−1⊗xy=x\otimes 1-1\otimes x of B is not zero, but if φ\varphi is the homomorphism from B to L′L' that sends ξ⊗η\xi \otimes \eta to ξη\xi \eta, then we have φ(y)=0\varphi (y) = 0, so yy is not invertible in B. We view the ring C=D⊗LL′C = D\otimes_LL' as a right module over its subring B; it is a free module because D is a right vector space over its subfield L′L'. Since yy is a nonzero and noninvertible element of the integral domain B, right multiplication by yy in C is a mapping RyR_y that is injective but not bijective. Now, RyR_y is an endomorphism of the left C-module CsC_s; consequently (VIII, p. 28, Corollary 1), the ring C is not left Artinian.

Let us now prove that (i) implies (v). This is a consequence of the following lemma.

Lemma 3 00FI

Let A be an absolutely semisimple K-algebra and L an algebraically closed extension of K. Then the algebra A(L)A_{(L)} is isomorphic to a product of finitely many matrix algebras over L.

The L-algebra A(L)A_{(L)} is semisimple; it is therefore isomorphic to a product of finitely many algebras of the form Mni(Di)\mathbf{M}_{n_i}(D_i), where DiD_i is a field containing L in its center and nin_i an integer ⩾1\geqslant 1 (VIII, p. 137, Remark 1).

Let L′L' be an extension of L. Since the K-algebra A is absolutely semisimple, the ring A(L′)A_{(L')} is semisimple and therefore left Artinian. Now, the ring A(L′)A_{(L')} is isomorphic to L′⊗LA(L)L'\otimes_LA_{(L)}, hence to ∏i∈IMni(L′⊗LDi)\prod_{i\in I}\mathbf{M}_{n_i}(L'\otimes_LD_i); by Proposition 5 of VIII, p. 7, each of the rings Mni(L′⊗LDi)\mathbf{M}_{n_i}(L'\otimes_LD_i) is therefore left Artinian.

Let n⩾1n\geqslant 1 be an integer and B a ring. Let (br)r⩾0(\mathfrak{b}_r)_{r\geqslant 0} be a decreasing sequence of left ideals of B; denote by cr\mathfrak{c}_r the set of square matrices of order nn with entries in br\mathfrak{b}_r. Then (cr)r⩾0(\mathfrak{c}_r)_{r\geqslant 0} is a decreasing sequence of left ideals of Mn(B)\mathbf{M}_n(B). In particular, if the ring Mn(B)\mathbf{M}_n(B) is left Artinian, then so is B.

By the above, for every i∈Ii\in I and every extension L′L' of L, the ring Di⊗LL′D_i\otimes_LL' is left Artinian. By Lemma 2, we have Di=LD_i= L for every i∈Ii\in I, which implies Lemma 3.

We will use the following lemma to prove the implication (i)⇒\Rightarrow(vi).

Lemma 4 00FJ

Let A and B be algebras over the field K that have finite generating sets, and let K′K' be an extension of K. If the K′K'-algebras A(K′)A_{(K')} and B(K′)B_{(K')} are isomorphic, then there exists a subextension L of K′K', finitely generated over K, such that the L-algebras A(L)A_{(L)} and B(L)B_{(L)} are isomorphic.

Let (ei)i∈I(e_i)_{i\in I} and (fj)j∈J(f_j)_{j\in J} be finite generating sets for the algebras A and B, respectively. Let uu be an isomorphism from A(K′)A_{(K')} to B(K′)B_{(K')} and vv the inverse isomorphism; there exists a subextension L of K′K', finitely generated over K, such that we have u(1⊗ei)∈B(L)u(1\otimes e_i)\in B_{(L)} for every i∈Ii\in I and v(1⊗fj)∈A(L)v(1\otimes f_j)\in A_{(L)} for every j∈Jj\in J. Consequently, uu sends A(L)A_{(L)} into B(L)B_{(L)}, and vv sends B(L)B_{(L)} into A(L)A_{(L)}. The induced mappings u′:A(L)→B(L)u': A_{(L)}\rightarrow B_{(L)} and v′:B(L)→A(L)v': B_{(L)}\rightarrow A_{(L)} are ring homomorphisms; they are inverse bijections.

Let us complete the proof of the implication (i)⇒\Rightarrow(vi). Let K′K' be a separable closure of K (V, §7, No. 8, p. 45). Then A(K′)A_{(K')} is an absolutely semisimple algebra over K′K'. By the implication (i)⇒\Rightarrow(iv), the K′K'-algebra A(K′)A_{(K')} is isomorphic to a product Mn1(D1)×⋯×Mnr(Dr)\mathbf{M}_{n_1}(D_1)\times \cdots \times \mathbf{M}_{n_r}(D_r), where DiD_i is a K′K'-algebra of finite degree that is a field whose center ZiZ_i is a separable extension of K′K'. Since K′K' is separably closed, we have Zi=K′Z_i= K'. By Proposition 3 of VIII, p. 231, the field DiD_i is commutative. We therefore have Di=K′D_i= K'. We denote the K-algebra Mn1(K)×⋯×Mnr(K)\mathbf{M}_{n_1}(K)\times \cdots \times \mathbf{M}_{n_r}(K) by B. The K′K'-algebras A(K′)A_{(K')} and B(K′)B_{(K')} are isomorphic by the above. Every subextension of K′K' that is finitely generated over K is separable and has finite degree over K, hence is contained in a subextension L of K′K' that is Galois and of finite degree over K (V, §10, No. 1, p. 57, Proposition 2). The implication therefore follows from Lemma 4.

The implication (vi)⇒\Rightarrow(v) is immediate.

Corollary 1 00FK

Let K be a commutative field, and let A1A_1 and A2A_2 be K-algebras. Suppose that A1A_1 is absolutely semisimple.

a) If A2A_2 is semisimple, then so is A1⊗KA2A_1\otimes_KA_2.

b) If A2A_2 is absolutely semisimple, then so is A1⊗KA2A_1\otimes_KA_2.

Denote the center of A1A_1 by Z1Z_1 and that of A2A_2 by Z2Z_2. The center Z of A1⊗KA2A_1\otimes_KA_2 is equal to Z1⊗KZ2Z_1\otimes_KZ_2 by the corollary of III, §4, No. 4, p. 468. Suppose that A2A_2 is semisimple; then Z2Z_2 is a reduced algebra (VIII, p. 137, Propositions 2 and 3). By Theorem 1, Z1Z_1 is an étale and therefore separable K-algebra. By Proposition 5 of V, §15, No. 2, p. 120, the ring Z=Z1⊗KZ2Z = Z_1\otimes_KZ_2 is reduced; since A1A_1 has finite degree over K (Theorem 1), it follows from Proposition 7 of VIII, p. 221 that the ring A1⊗KA2A_1\otimes_KA_2 is semisimple.

Now suppose that A2A_2 is absolutely semisimple. Let L be an extension of K. Then the algebra A1(L)A_{1(L)} is absolutely semisimple, and the algebra A2(L)A_{2(L)} is semisimple. Therefore, by a), the algebra A1⊗KA2(L)A_1\otimes_KA_{2(L)} is semisimple.

Corollary 2 00FL

Let K be a separably closed field, and let A be an absolutely semisimple K-algebra. Then there exist an integer r⩾0r\geqslant 0 and strictly positive integers n1,...,nrn_1, . . . , n_r such that the algebra A is isomorphic to the algebra ∏i=1rMni(K)\prod_{i=1}^r\mathbf{M}_{n_i}(K).

By Theorem 1, A is isomorphic to an algebra of the form ∏i=1rMni(Di)\prod^r_{i=1}\mathbf{M}_{n_i}(D_i) for an integer r⩾0r\geqslant 0, integers n1,...,nrn_1, . . . , n_r, and K-algebras of finite degree D1,...,DrD_1, . . . ,D_r that are fields and whose centers are separable extensions of K and therefore equal to K. By Proposition 3 of VIII, p. 231, we have Di=KD_i= K for i∈[1,r]i\in [1, r].

Example 00FM

A commutative K-algebra is absolutely semisimple if and only if it is étale: this follows from the definition (V, §6, No. 3, p. 28, Definition 1) and the equivalence of properties (i) and (v) of Theorem 1.

4. Characterization of Absolutely Semisimple Modules

Proposition 4 00FN

Let K be a commutative field and A a K-algebra.

a) Let M be a semisimple A-module. The A-module M is absolutely semisimple if and only if every simple module belonging to the support of M is.

b) Let S be a simple A-module, and let D be its commutant. The following properties are equivalent:

(i) The A-module S is absolutely semisimple.

(ii) The K-algebra D is absolutely semisimple.

(iii) The K-algebra D is a field, has finite degree over K, and its

center is a separable extension of K.

Assertion a) follows from Proposition 1, a) of VIII, p. 229. Let S and D be as in b), and let L be an extension of K. The A(L)A_{(L)}-module S(L)S_{(L)} is semisimple if and only if the ring D(L)D_{(L)} is (VIII, p. 222, Proposition 8, c)). This proves the equivalence of (i) and (ii), and that of (ii) and (iii) follows from Theorem 1 because D is a field.

Corollary 00RF

Let K be a commutative field, let A1A_1 and A2A_2 be K-algebras, and let M1M_1 be an absolutely semisimple A1A_1-module and M2M_2 a semisimple A2A_2-module. Then M1⊗KM2M_1\otimes_KM_2 is a semisimple module over the ring A1⊗KA2A_1\otimes_KA_2.

The module M1M_1 is the direct sum of absolutely semisimple simple A1A_1-modules (Proposition 4). It therefore suffices to prove the assertion in the case when the modules M1M_1 and M2M_2 are simple. Denote their commutants by D1D_1 and D2D_2. The K-algebra D1D_1 is absolutely semisimple (loc. cit.); by Corollary 1 of VIII, p. 234, the K-algebra D1⊗KD2D_1\otimes_KD_2 is semisimple. It then follows from Corollary 1 of VIII, p. 215 that the (A1⊗KA2(A_1\otimes_KA_2)-module M1⊗KM2M_1\otimes_KM_2 is semisimple.

5. Derivations on Semisimple Algebras

In this subsection and the next ones, K is a commutative ring, A a K-algebra, B the K-algebra A⊗KAoA\otimes_KA^o, and ε\varepsilon the K-linear mapping from B to A defined by ε(x⊗y)=xy\varepsilon (x\otimes y) =xy for x,yx, y in A.

Recall (III, §4, No. 3, p. 467) that every (A(A, A)-bimodule P can be viewed as a left B-module, with external law given by (a⊗a′)p=apa′(a\otimes a')p=apa' for a,a′a, a' in A and pp in P. Conversely, every B-module can be viewed as an (A,A)(A,A)-bimodule. We endow A with its canonical (A(A, A)-bimodule structure and with the corresponding B-module structure; we endow B with the (A(A, A)-bimodule structure corresponding to the B-module BsB_s. We therefore have

a(x⊗y)a′=(a⊗a′)(x⊗y)=ax⊗ya′ a(x\otimes y)a'= (a\otimes a')(x\otimes y) =ax\otimes ya'

for a,a′,x,ya, a', x, y in A, where the product ya′ya' is calculated in the algebra A.

The K-linear mapping ε\varepsilon is a homomorphism of (A(A, A)-bimodules.

Proposition 5 00FO

The following properties are equivalent:

(i) The B-module A is projective.

(ii) There exists an element ee of the (A,A)(A,A)-bimodule B satisfying the following two conditions: ε(e)=1\varepsilon (e) = 1 and ae=eaae=ea for every a∈Aa\in A.

The mapping ε:B→A\varepsilon : B\rightarrow A is surjective because we have ε(a⊗1)=a\varepsilon (a\otimes 1) =a for every a∈Aa\in A; it is a homomorphism of (A(A, A)-modules, hence a B-linear mapping. If the B-module A is projective, then there exists a section ss of ε\varepsilon (II, §2, No. 2, p. 231, Proposition 4); it is a homomorphism of (A(A, A)-bimodules from A to B. If we set e=se=s(1), then we have ε(e)=ε(s(1))=1\varepsilon (e) =\varepsilon (s(1)) = 1 and ae=s(a)=eaae=s(a) =ea for every a∈Aa\in A. So (i) implies (ii).

Conversely, let ee be an element of B satisfying the conditions in item (ii). Define a mapping ss from A to B by the formula

s(a)=ae=ea(1) s(a) =ae=ea \tag{1}

It is a homomorphism of (A(A, A)-modules, and we have ε∘s=1A\varepsilon \circ s= 1_A; in other words, ss is a B-linear section of the surjective mapping ε\varepsilon. Consequently, the B-module A is isomorphic to the direct factor submodule s(A)s(A) of BsB_s (II, §1, No. 9, p. 211, Proposition 15) and is therefore projective (II, §2, No. 2, p. 231, Proposition 4). This proves that (ii) implies (i).

Remark 1 00FP

Let e=∑i=1rai⊗ai′e=\sum^r_{i=1}a_i\otimes a'_i be an element of B. The conditions in item (ii) of Proposition 5 translate to the formulas

∑i=1raiai′=1(2) \sum_{i=1}^ra_ia'_i= 1 \tag{2}

(3) ∑i=1raai⊗ai′=∑i=1rai⊗ai′a\sum_{i=1}^raa_i\otimes a'_i=\sum_{i=1}^ra_i\otimes a'_ia for every a∈Aa\in A.

When they are satisfied, ee is an idempotent in B. Indeed, we then have the relations

e2=∑i=1raieai′=∑i=1reaiai′=e e^2=\sum_{i=1}^ra_iea'_i=\sum_{i=1}^rea_ia'_i=e

Remark 2 00FQ

Let K be a commutative field, let A be a K-algebra, and let M be an A-module. The group EndK(M)_K(M) is endowed with an (A(A, A)-bimodule structure defined by

aua′(x)=au(a′x) aua'(x) =au(a'x)

for every a,a′∈Aa, a'\in A, every u∈u\in EndK_K(M), and every x∈Mx\in M. We endow it with the associated B-module structure. Let e=∑i=1rai⊗ai′e=\sum^r_{i=1}a_i\otimes a'_i be an element of B satisfying the conditions in part (ii) of Proposition 5, so also relations (2) and (3). If p∈p\in EndK(M)_K(M) is a projector whose image N is an A-submodule of M, then epep is an A-linear projector with the same image.

Indeed, the image of epep is contained in N. If xx belongs to N, then so does ai′xa'_ix, so that we have p(ai′x)=ai′xp(a'_ix) =a'_ix and

ep(x)=∑i=1raiai′x=x ep(x) =\sum_{i=1}^ra_ia'_ix=x

by formula (2). We deduce from formula (3) that aep(x)=ep(ax)aep(x) =ep(ax) for every a∈Aa\in A and every x∈Mx\in M, which proves that epep is A-linear.

Theorem 2 00RG

Let K be a commutative field and A a K-algebra. The following properties are equivalent:

(i) The K-algebra A is absolutely semisimple.

(ii) The K-algebra B=A⊗KAoB = A\otimes_KA^o is semisimple.

(iii) The B-module A is projective.

(iv) There exists an element ee of the (A,A)(A,A)-bimodule B satisfying ε(e)=1\varepsilon (e) = 1 and ae=eaae=ea for every a∈Aa\in A.

Suppose that the algebra A is absolutely semisimple, hence semisimple. Then the algebra AoA^o is semisimple (VIII, p. 137, Proposition 2), and it follows from Corollary 1 of VIII, p. 234 that B=A⊗KAoB = A\otimes_KA^o is a semisimple K-algebra. Therefore, (i) implies (ii).

Since every module over a semisimple ring is projective (VIII, p. 138, Proposition 4), (ii) implies (iii).

The equivalence of (iii) and (iv) follows from Proposition 5. To complete the proof, let us show that (iv) implies (i). Let e=∑i=1rai⊗ai′e=\sum^r_{i=1}a_i\otimes a'_i be an element of B satisfying the conditions in item (ii) of Proposition 5. Let L be an extension of the field K; we must prove that the ring A(L)A_{(L)} is semisimple or, equivalently, that every A(L)A_{(L)}-module is semisimple (VIII, p. 138, Proposition 4). Let M be an A(L)A_{(L)}-module, and let N be a submodule of M; we view M as a left (A(A, L)-bimodule and N as a sub-bimodule (III, §4, No. 3, p. 466). Since L is a field, there exists an L-linear projector uu in M with image N. Since the homotheties aMa_M associated with the elements aa of A are L-linear, there exists a unique group homomorphism from A⊗KAoA\otimes_KA^o to EndL(M)_L(M) that sends an element a⊗a′a\otimes a' to the L-linear mapping x↦au(a′x)x\mapsto au(a'x). We denote the image of ee by this homomorphism by vv; it follows from Remark 2 that vv is an A(L)A_{(L)}-linear projector with image N. The kernel of vv is an A(L)A_{(L)}-submodule of M, supplementary to N. By Corollary 2 of VIII, p. 56, the A(L)A_{(L)}-module M is semisimple.

Remark 3 00FR

We know (VIII, p. 234, Corollary 1) that the tensor product of two absolutely semisimple algebras over a commutative field is absolutely semisimple. Consequently, if the algebra A is absolutely semisimple, then so is the algebra B=A⊗KAoB = A\otimes_KA^o.

6. Cohomology of Algebras

In this subsection, K is a commutative ring, A a K-algebra, B the K-algebra A⊗KAoA\otimes_KA^o, and ε\varepsilon the K-linear mapping from B to A defined by ε(x⊗y)=\varepsilon (x\otimes y) = xyxy for x,yx, y in A. For n∈Nn\in \mathbf{N}, we denote the tensor product over K of n+2n+ 2 copies of the K-module A by BnB_n. We view it as an (A(A, A)-bimodule (and also as a B-module); we endow it with the structure of a left A-module deduced from the left A-module structure of the first factor of the tensor product and with the structure of a right A-module deduced from the right A-module structure of the last factor. In particular, B0B_0 is just the (A(A, A)-bimodule B.

For every integer n⩾1n\geqslant 1, we define homomorphisms of bimodules dnid_n^i for i∈[0,n]i\in [0, n] and dnd_n from BnB_n to Bn−1B_{n-1} by the formulas

dni(x0⊗⋯⊗xn+1)=x0⊗⋯⊗xixi+1⊗⋯⊗xn+1(4) d_n^i(x_0\otimes \cdots \otimes x_{n+1}) =x_0\otimes \cdots \otimes x_ix_{i+1}\otimes \cdots \otimes x_{n+1} \tag{4}

for i∈[0,n]i\in [0, n] and

dn=∑i=0n(−1)idni(5) d_n=\sum_{i=0}^n(-1)^id_n^i \tag{5}

We denote the mapping ε:B0→A\varepsilon : B_0\rightarrow A by d0=d00d_0=d^0_0.

Let nn be an integer ⩾1\geqslant 1. For 0⩽i<j⩽n0\leqslant i < j\leqslant n, we have

dn−1i∘dnj=dn−1j−1∘dni(6) d_{n-1}^i\circ d_n^j=d_{n-1}^{j-1}\circ d_n^i \tag{6}

and from this we deduce

dn−1∘dn=∑0⩽i<j⩽n(−1)i+jdn−1i∘dnj+∑0⩽j⩽i⩽n−1(−1)i+jdn−1i∘dnj d_{n-1}\circ d_n=\sum_{0\leqslant i<j\leqslant n}(-1)^{i+j}d_{n-1}^i\circ d_n^j+\sum_{0\leqslant j\leqslant i\leqslant n-1}(-1)^{i+j}d_{n-1}^i\circ d_n^j
=∑0⩽i<j⩽n(−1)i+jdn−1j−1∘dni+∑0⩽j⩽i⩽n−1(−1)i+jdn−1i∘dnj =\sum_{0\leqslant i<j\leqslant n}(-1)^{i+j}d_{n-1}^{j-1}\circ d_n^i+\sum_{0\leqslant j\leqslant i\leqslant n-1}(-1)^{i+j}d_{n-1}^i\circ d_n^j

and consequently

dn−1∘dn=0(7) d_{n-1}\circ d_n= 0 \tag{7}

Let P be an (A(A, A)-bimodule. For any integer n⩾0n\geqslant 0, we denote the K-module of K-multilinear mappings from AnA^n to P by Cn(A,P)C^n(A,P). The mapping αn:Cn(A,P)→\alpha^n: C^n(A,P)\rightarrow HomB(Bn,P)_B(B_n,P) that sends f∈Cn(A,P)f\in C^n(A,P) to the homomorphism αn(f)\alpha^n(f) characterized by

αn(f)(x0⊗⋯⊗xn+1)=x0f(x1,...,xn)xn+1(8) \alpha^n(f) (x_0\otimes \cdots \otimes x_{n+1}) =x_0f(x_1, . . . , x_n)x_{n+1} \tag{8}

is an isomorphism of K-modules.

We denote by ∂n\partial^n (for n≥0n \geq 0) the unique K-linear mapping from Cn(A,P)C^n(A, P) to Cn+1(A,P)C^{n+1}(A, P) that makes the following diagram commutative:

Cn(A,P)→∂nCn+1(A,P)∣↓αn∣∣↓αn+1∣Hom⁡B(Bn,P)→Hom⁡(dn+1,1P)Hom⁡B(Bn+1,P). \begin{array}{ccc} C^n(A, P) & \xrightarrow{\partial^n} & C^{n+1}(A, P) \\ | \downarrow \alpha^n | & & | \downarrow \alpha^{n+1} | \\ \operatorname{Hom}_B(B_n, P) & \xrightarrow{\operatorname{Hom}(d_{n+1}, 1_P)} & \operatorname{Hom}_B(B_{n+1}, P). \end{array}

By definition, we therefore have

(9)(αn+1∘∂n)(f)=αn(f)∘dn+1 (9) \quad (\alpha^{n+1} \circ \partial^n)(f) = \alpha^n(f) \circ d_{n+1}

for every f∈Cn(A,P)f \in C^n(A, P). In other words, we have

∂n(f)(x0,…,xn)=αn(f)(dn+1(1⊗x0⊗⋯⊗xn⊗1)) \partial^n(f)(x_0, \ldots, x_n) = \alpha^n(f)(d_{n+1}(1 \otimes x_0 \otimes \cdots \otimes x_n \otimes 1))

for x0,…,xnx_0, \ldots, x_n in A and ff in Cn(A,P)C^n(A, P), that is,

(10)∂n(f)(x0,…,xn)=x0f(x1,…,xn)+∑i=0n−1(−1)i+1f(x0,…,xi−1,xixi+1,xi+2,…,xn)=+(−1)n+1f(x0,…,xn−1)xn. \begin{align*} (10) \quad \partial^n(f)(x_0, \ldots, x_n) &= x_0 f(x_1, \ldots, x_n) \\ &+ \sum_{i=0}^{n-1} (-1)^{i+1} f(x_0, \ldots, x_{i-1}, x_i x_{i+1}, x_{i+2}, \ldots, x_n) \\ &\phantom{=} + (-1)^{n+1} f(x_0, \ldots, x_{n-1}) x_n. \end{align*}

By (7) and (9), we have

(11)∂n+1∘∂n=0for every n≥0. (11) \quad \partial^{n+1} \circ \partial^n = 0 \qquad \text{for every } n \geq 0.

We denote the K-module Ker⁡∂0\operatorname{Ker} \partial^0 by H0(A,P)H^0(A, P) and, for n≥1n \geq 1, the K-module Ker⁡∂n/Im⁡∂n−1\operatorname{Ker} \partial^n / \operatorname{Im} \partial^{n-1} by Hn(A,P)H^n(A, P). We identify the K-module C0(A,P)C^0(A, P) with P, and we have C1(A,P)=Hom⁡K(A,P)C^1(A, P) = \operatorname{Hom}_K(A, P). The mappings ∂n\partial^n for n≤2n \leq 2 are given by the formulas

(12)∂0(p)(a)=ap−pafor every p∈P,(13)∂1(f)(a,a′)=af(a′)−f(aa′)+f(a)a′for f∈C1(A,P),(14)∂2(f)(a,a′,a′′)=af(a′,a′′)−f(aa′,a′′)+f(a,a′a′′)−f(a,a′)a′′ \begin{align*} (12) \quad \partial^0(p)(a) &= ap - pa \qquad \text{for every } p \in P, \\ (13) \quad \partial^1(f)(a, a') &= af(a') - f(aa') + f(a)a' \qquad \text{for } f \in C^1(A, P), \\ (14) \quad \partial^2(f)(a, a', a'') &= af(a', a'') - f(aa', a'') + f(a, a'a'') - f(a, a')a'' \end{align*}

for f∈C2(A,P)f \in C^2(A, P).

So H0(A,P)H^0(A, P) is the K-submodule of P consisting of the elements pp such that ap=paap = pa for every a∈Aa \in A, and H1(A,P)H^1(A, P) is the quotient of the K-module Der⁡K(A,P)\operatorname{Der}_K(A, P) of K-derivations from A to P (III, §10, No. 2, p. 553) by the K-submodule consisting of the derivations of the form a↦ap−paa \mapsto ap - pa with p∈Pp \in P (called inner derivations).

7. Cohomology of Absolutely Semisimple Algebras

Proposition 6 00FS

Let K be a commutative ring and A a K-algebra. Let e=∑i=1rai⊗ai′e=\sum^r_{i=1}a_i\otimes a'_i be an element of B=A⊗KAoB = A\otimes_KA^o satisfying the conditions in item (ii) of Proposition 5 of VIII, p. 236. For any integer n⩾1n\geqslant 1 and any element ff of Cn(A,P)C^n(A,P), we denote by γn(f)\gamma^n(f) the element of Cn−1(A,P)C^{n-1}(A,P) defined by the formula

γn(f)(x1,...,xn−1)=∑i=1raif(ai′,x1...,xn−1)(15) \gamma^n(f)(x_1, . . . , x_{n-1}) =\sum_{i=1}^ra_if(a'_i, x_1. . . , x_{n-1}) \tag{15}

We then have

∂n−1(γn(f))+γn+1(∂n(f))=f(16) \partial^{n-1}(\gamma^n(f)) +\gamma^{n+1}(\partial^n(f)) =f \tag{16}

for every integer n⩾1n\geqslant 1 and every f∈Cn(A,P)f\in C^n(A,P).

Remark 1 00SF

∗*The morphisms ∂n:Cn(A,P)→Cn+1(A,P)\partial^n: C^n(A,P)\rightarrow C^{n+1}(A,P) define a complex (C(A,P),∂)(C(A,P), \partial ) of K-modules (X, §2, no1^o1, p. 24). The mapping γn\gamma_n therefore defines a homotopy from this complex to itself linking 0 to IdC(A,P)_{C(A,P)} (X, §2, no4^o4, p. 32, définition 44).∗.*

We keep the notation of No.6. For every integer n⩾0n\geqslant 0, we define a mapping hn:Bn→Bn+1h_n: B_n\rightarrow B_{n+1} by the formula

hn(x)=dn+21(e⊗x)=∑i=1rai⊗ai′x h_n(x) =d^1_{n+2}(e\otimes x) =\sum_{i=1}^ra_i\otimes a'_ix

It is a homomorphism of (A(A, A)-bimodules (formula (3)).

Lemma 5 00FT

We have the relation

dn+1∘hn+hn−1∘dn=1Bn(17) d_{n+1}\circ h_n+h_{n-1}\circ d_n= 1_{B_n} \tag{17}

for every n⩾1n\geqslant 1.

Let x∈Bnx\in B_n; we have

(dn+1∘hn)(x)=(dn+1∘dn+21)(e⊗x) (d_{n+1}\circ h_n)(x) = (d_{n+1}\circ d^1_{n+2})(e\otimes x)
=(dn+10∘dn+21)(e⊗x)−∑i=2n+2(−1)i(dn+1i−1∘dn+21)(e⊗x) = (d^0_{n+1}\circ d^1_{n+2})(e\otimes x)-\sum^{n+2}_{i=2}(-1)^i(d_{n+1}^{i-1}\circ d^1_{n+2})(e\otimes x)

so, by formula (6),

(dn+1∘hn)(x)=(dn+10∘dn+20)(e⊗x)−∑i=2n+2(−1)i(dn+11∘dn+2i)(e⊗x) (d_{n+1}\circ h_n)(x) = (d^0_{n+1}\circ d^0_{n+2})(e\otimes x)-\sum^{n+2}_{i=2}(-1)^i(d^1_{n+1}\circ d_{n+2}^i)(e\otimes x)

But we have

(dn+10∘dn+20)(e⊗x)=ε(e)x=x (d^0_{n+1}\circ d^0_{n+2})(e\otimes x) =\varepsilon (e)x=x

by property (ii) of Proposition 5 of VIII, p. 236 and, for i⩾2i\geqslant 2,

dn+2i(e⊗x)=e⊗dni−2(x) d^i_{n+2}(e\otimes x) =e\otimes d^{i-2}_n(x)

which gives

(dn+1∘hn)(x)=x−dn+11(e⊗dn(x))=x−hn−1∘dn(x) (d_{n+1}\circ h_n)(x) =x-d^1_{n+1}(e\otimes d_n(x)) =x-h_{n-1}\circ d_n(x)

and therefore formula (17).

We can finish the proof of Proposition 6 using Lemma 5. Let nn be an integer ⩾1\geqslant 1 and ff an element of Cn(A,P)C^n(A,P). By construction, we have

αn−1(γn(f))=αn(f)∘hn−1(18) \alpha^{n-1}(\gamma^n(f)) =\alpha^n(f)\circ h_{n-1} \tag{18}

and, consequently, by formulas (9) and (18),

αn(∂n−1(γn(f))+γn+1(∂n(f)))=αn−1(γn(f))∘dn+αn+1(∂n(f))∘hn \alpha^n(\partial^{n-1}(\gamma^n(f)) +\gamma^{n+1}(\partial^n(f))) =\alpha^{n-1}(\gamma^n(f))\circ d_n+\alpha^{n+1}(\partial^n(f))\circ h_n
=αn(f)∘hn−1∘dn+αn(f)∘dn+1∘hn =\alpha^n(f)\circ h_{n-1}\circ d_n+\alpha^n(f)\circ d_{n+1}\circ h_n
=αn(f) =\alpha^n(f)

where the last equality follows from (17). Since αn\alpha^n is bijective, the proposition follows.

Theorem 3 00FU

Let K be a commutative ring, A a K-algebra, and P an (A,A)(A,A)-bimodule. Suppose that the (A⊗KAo)(A\otimes_KA^o)-module A is projective. We then have Hn(A,P)=0H^n(A,P) = 0 for every integer n⩾1n\geqslant 1.

We must prove that for every integer n⩾1n\geqslant 1, every element ff of Cn(A,P)C^n(A,P) with ∂n(f)=0\partial^n(f) = 0 is of the form ∂n−1(g)\partial^{n-1}(g) for an element gg of Cn−1(A,P)C^{n-1}(A,P). By Proposition 5, this is an immediate consequence of Proposition 6.

Corollary 00FV

Every K-derivation from A to P is inner.

This is a translation of the equality H1(A,P)=0H^1(A,P) = 0.

Remark 2 00FW

The assumptions of Theorem 3 are, in particular, satisfied when K is a field and A an absolutely semisimple K-algebra (VIII, p. 238, Theorem 2).

Remark 3 00FX

Suppose that the K-module A is projective. Theorem 3 can also be proved as follows. The complex (⨁n⩾0Bn,d)(\bigoplus_{n\geqslant 0}B_n, d) and the homomorphism ε:B0→A\varepsilon : B_0\rightarrow A define a projective resolution of the B-module A; therefore, for every n⩾0n\geqslant 0, the K-module Hn(A,P)H^n(A,P) is isomorphic to ExtBn(A,P)^n_B(A,P) (X, §6, no1^o1, p. 100, théorème 1). If the B-module A is projective, then the K-modules ExtBn(A,P)^n_B(A,P) are zero for n⩾1n\geqslant 1 (X, §5, no3^o3, p. 88, corollaire de la proposition 5), which implies that Hn(A,P)H^n(A,P) is zero. Conversely, if H1(A,P)H^1(A,P) is zero for every (A(A, A)-bimodule P, then the B-module A is projective (X, §5, no5^o5, p. 93, proposition 1010).∗.*

8. The Splitting of Artinian Algebras

In this subsection, K is a commutative ring and A a K-algebra. Let r\mathfrak{r} be the radical of A. We denote the quotient algebra A/rA/\mathfrak{r} by A‾\overline{A} and the canonical mapping from A to A‾\overline{A} by π\pi. We are interested in the subalgebras S of A such that A=S⊕rA = S\oplus \mathfrak{r}.

We denote by Σ the set of K-linear sections ss of π\pi satisfying s(αβ)=s(\alpha \beta ) = s(α)s(β)s(\alpha )s(\beta ) for α,β\alpha , \beta in A. Note that such a section necessarily satisfies s(1)=1s(1) = 1 (in other words, ss is a ring homomorphism): we indeed have s(1)2=ss(1)^2=s(1), and s(1)s(1) is invertible because it belongs to 1+r1 +\mathfrak{r} (VIII, p. 156, Theorem 1). If ss is an element of Σ, then the image S of ss is a subalgebra of A, and we have A=S⊕rA = S\oplus \mathfrak{r}. Conversely, if S is a subalgebra of A such that A=S⊕rA = S\oplus \mathfrak{r}, then the restriction of π\pi to S is bijective, and the inverse bijection defines an element of Σ with image S.

By Jacobson’s theorem (loc. cit.), every element of 1+r1 +\mathfrak{r} is invertible in A. We call an inner automorphism of A of the form a↦xax−1a\mapsto xax^{-1} with x∈1+rx\in 1 +\mathfrak{r} a special automorphism.

Proposition 7 00FY

Suppose that the (A⊗KA‾o)(A\otimes_K\overline{A}^o)-module A‾\overline{A} is projective.

a) Let S1S_1 and S2S_2 be subalgebras of A satisfying A=S1⊕r=S2⊕rA = S_1\oplus \mathfrak{r}= S_2\oplus \mathfrak{r}. There exists a special automorphism of A transforming S1S_1 into S2S_2.

b) Suppose that π\pi has a K-linear section and that the radical r\mathfrak{r} of A is nilpotent. Then there exists a subalgebra S of A satisfying A=S⊕rA = S\oplus \mathfrak{r}.

Let S1S_1 and S2S_2 be as in a). Let s1s_1 and s2s_2 be the elements of the set Σ corresponding to the subalgebras S1S_1 and S2S_2. Let ε\varepsilon be the K-linear mapping from A⊗KAA\otimes_KA to A given by ε(a⊗b)=ab\varepsilon (a\otimes b) =ab. By Proposition 5 of VIII, p. 236 and Remark 1 of VIII, p. 237, there exists an element e=∑i=1rαi⊗αi′e=\sum^r_{i=1}\alpha_i\otimes \alpha '_i of A‾⊗KA\overline{A}\otimes_KA satisfying ∑i=1rαiαi′=1\sum^r_{i=1}\alpha_i\alpha '_i= 1 and ∑i=1rααi⊗αi′=∑i=1rαi⊗αi′α\sum^r_{i=1}\alpha \alpha_i\otimes \alpha '_i=\sum^r_{i=1}\alpha_i\otimes \alpha '_i\alpha for every α∈A\alpha \in A. Set x=∑i=1rs1(αi)s2(αi′)x=\sum^r_{i=1}s_1(\alpha_i)s_2(\alpha '_i). We have π(x)=∑i=1rαiαi′=1\pi (x) =\sum^r_{i=1}\alpha_i\alpha '_i= 1 and therefore x∈1+rx\in 1 +\mathfrak{r}. Let α\alpha be an element of A. We have

s1(α)x=∑i=1rs1(ααi)s2(αi′)=(ε∘(s1⊗s2))(∑i=1rααi⊗αi′) s_1(\alpha )x=\sum_{i=1}^rs_1(\alpha \alpha_i)s_2(\alpha '_i) = (\varepsilon \circ (s_1\otimes s_2))(\sum_{i=1}^r\alpha \alpha_i\otimes \alpha '_i)
=(ε∘(s1⊗s2))(∑i=1rαi⊗αi′α)=∑i=1rs1(αi)s2(αi′α)=xs2(α) = (\varepsilon \circ (s_1\otimes s_2))(\sum_{i=1}^r\alpha_i\otimes \alpha '_i\alpha )=\sum_{i=1}^rs_1(\alpha_i)s_2(\alpha '_i\alpha ) =xs_2(\alpha )

The equality x−1S1x=S2x^{-1}S_1x= S_2 follows, giving assertion a).

Under the assumptions of b), suppose furthermore that r2=0\mathfrak{r}^2= 0. In this case, the (A(A, A)-bimodule r\mathfrak{r} is annihilated by r\mathfrak{r}, and we therefore view it as an (A(A, A)-bimodule. Choose a K-linear section σ\sigma of π\pi. We have

(19) αx=σ(α)x\alpha x=\sigma (\alpha )x and xα=xσ(α)x\alpha =x\sigma (\alpha )

for α∈A‾\alpha \in \overline{A} and x∈rx\in \mathfrak{r}. Set

φ(α,β)=σ(αβ)−σ(α)σ(β)(20) \varphi (\alpha , \beta ) =\sigma (\alpha \beta )-\sigma (\alpha )\sigma (\beta ) \tag{20}

for α,β∈A\alpha , \beta \in A. We have the relation π(φ(α,β))=αβ−αβ=0\pi (\varphi (\alpha , \beta )) =\alpha \beta -\alpha \beta = 0 for α,β∈A\alpha , \beta \in A. Therefore, φ\varphi defines an element of C2(A,r)C^2(A,\mathfrak{r}). Let α,β,γ\alpha , \beta , \gamma be elements of A; in view of (19), we have

∂2φ(α,β,γ)=αφ(β,γ)−φ(αβ,γ)+φ(α,βγ)−φ(α,β)γ \partial^2\varphi (\alpha , \beta , \gamma ) =\alpha \varphi (\beta , \gamma )-\varphi (\alpha \beta , \gamma ) +\varphi (\alpha , \beta \gamma )-\varphi (\alpha , \beta )\gamma
=σ(α)φ(β,γ)−φ(αβ,γ)+φ(α,βγ)−φ(α,β)σ(γ) =\sigma (\alpha )\varphi (\beta , \gamma )-\varphi (\alpha \beta , \gamma ) +\varphi (\alpha , \beta \gamma )-\varphi (\alpha , \beta )\sigma (\gamma )
=σ(α)(σ(βγ)−σ(β)σ(γ))−σ(αβγ)+σ(αβ)σ(γ)+σ(αβγ) =\sigma (\alpha )(\sigma (\beta \gamma )-\sigma (\beta )\sigma (\gamma ))-\sigma (\alpha \beta \gamma ) +\sigma (\alpha \beta )\sigma (\gamma ) +\sigma (\alpha \beta \gamma )
−σ(α)σ(βγ)−(σ(αβ)−σ(α)σ(β))σ(γ) -\sigma (\alpha )\sigma (\beta \gamma )-(\sigma (\alpha \beta )-\sigma (\alpha )\sigma (\beta ))\sigma (\gamma )
=0 = 0

By Theorem 3 of VIII, p. 242, the K-module H2(A,r)H^2(A,\mathfrak{r}) is reduced to zero. Hence there exists an element ψ\psi of C1(A,r)C^1(A,\mathfrak{r}) such that ∂1ψ=φ\partial^1\psi =\varphi, in other words, such that we have

(21) φ(α,β)=αψ(β)−ψ(αβ)+ψ(α)β\varphi (\alpha , \beta ) =\alpha \psi (\beta )-\psi (\alpha \beta ) +\psi (\alpha )\beta for α,β\alpha , \beta in A‾\overline{A}.

We have ψ(α)ψ(β)=0\psi (\alpha )\psi (\beta ) = 0 because r2\mathfrak{r}^2 is zero; from (19) and (20), we then deduce (22) (σ+ψ)(αβ)=(σ+ψ)(α)(σ+ψ)(β)(\sigma +\psi )(\alpha \beta ) = (\sigma +\psi )(\alpha )(\sigma +\psi )(\beta ),

so that the K-linear section σ+ψ\sigma +\psi of π\pi belongs to Σ. Its image is a subalgebra S of A such that A=S+rA = S +\mathfrak{r}.

Let us now prove the existence of S in the general case. We reason by induction on the least integer p⩾1p\geqslant 1 such that rp=0\mathfrak{r}^p= 0; the case p=1p= 1 is trivial. Suppose p⩾2p\geqslant 2, and set A′=A/rp−1A'= A/\mathfrak{r}^{p-1}. The radical r′\mathfrak{r}' of A′A' is equal to r/rp−1\mathfrak{r}/\mathfrak{r}^{p-1} (Proposition 5 of VIII, p. 155), so satisfies r′p−1=0{\mathfrak{r}'}^{p-1}= 0, and the algebra A′/r′A'/\mathfrak{r}' is isomorphic to A‾=A/r\overline{A}= A/\mathfrak{r} and is therefore absolutely semisimple. By the induction hypothesis, there exists a subalgebra S′S' of A′A' such that A′=S′⊕r′A'= S'\oplus \mathfrak{r}'. Then S′S' is of the form A′′/rp−1A''/\mathfrak{r}^{p-1}, where A′′A'' is a subalgebra of A containing rp−1\mathfrak{r}^{p-1}, and we have

A=A′′+r,rp−1=A′′∩r(23) A = A''+\mathfrak{r},\mathfrak{r}^{p-1}= A''\cap \mathfrak{r} \tag{23}

The algebra A′′/rp−1A''/\mathfrak{r}^{p-1} is isomorphic to A′/r′A'/\mathfrak{r}'; we have (rp−1)2=0(\mathfrak{r}^{p-1})^2= 0, so rp−1\mathfrak{r}^{p-1} is the radical of A′′A''. By the case we just treated, there exists a subalgebra S of A′′A'' such that A′′=S⊕rp−1A''= S\oplus \mathfrak{r}^{p-1}; we deduce the relation A=S⊕rA = S\oplus \mathfrak{r} from (23).

Corollary 1 (Wedderburn’s theorem) 00S8

Let K be a commutative field, A a K-algebra, and r\mathfrak{r} the radical of A. Suppose that the K-algebra A/rA/\mathfrak{r} is absolutely semisimple.

a) Let S1S_1 and S2S_2 be subalgebras of A satisfying A=S1⊕r=S2⊕rA = S_1\oplus \mathfrak{r}= S_2\oplus \mathfrak{r}. There exists a special automorphism of A transforming S1S_1 into S2S_2.

b) If r\mathfrak{r} is nilpotent, then there exists a subalgebra S of A satisfying A = S⊕rS\oplus \mathfrak{r}.

This follows from Proposition 7 and Theorem 2 of VIII, p. 238.

Corollary 2 00FZ

Let A be a commutative algebra of finite degree over a perfect field K, and let r\mathfrak{r} be its radical. There exists a unique subalgebra S of A such that A=S⊕rA = S\oplus \mathfrak{r}. Moreover, S is isomorphic to a product of finitely many extensions of K of finite degree.

The K-algebra A/rA/\mathfrak{r} is semisimple (VIII, p. 173, Proposition 1) and has finite degree; it is absolutely semisimple because the field K is perfect (VIII, p. 232, Theorem 1). Since the ideal r\mathfrak{r} is nilpotent and A is commutative, the existence and uniqueness of S then follow from Corollary 1. Since S is semisimple and commutative and has finite degree, the last assertion is a consequence of Proposition 3 of VIII, p. 137.

Remark 1 00G0

The assumption that A/rA/\mathfrak{r} is absolutely semisimple is essential in Corollary 1 (VIII, p. 246, Exercise 4).

Remark 2 00G1

Suppose that A is an Artinian algebra over the field K. If A is commutative, then we can show (VIII, p. 180, Exercise 9) that A is isomorphic to a product of algebras A1×⋯×AnA_1\times \cdots \times A_n such that Ai/R(Ai)A_i/\mathfrak{R}(A_i) is a field for every ii. Conversely, if A is not commutative, then A may not be isomorphic to a product of algebras A1×⋯×AnA_1\times \cdots \times A_n such that Ai/R(Ai)A_i/\mathfrak{R}(A_i) is a simple ring for every ii (VIII, p. 247, Exercise 5).

Exercises

See the exercises for § 13.

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