Bourbaki

§ 4. Semisimple modules

1. Semisimple Modules

Definition 1 0040

A module is called semisimple if it is the direct sum of a family of simple modules.[^1].

A multimodule is called semisimple if it is the direct sum of a family of simple multimodules (cf. I, §4, No. 4, p. 37, Definition 7).

An A-module M is semisimple if and only if it is semisimple when viewed as a module over its ring of homotheties AMA_M.

Example 1 0041

A module reduced to 0 and a simple module are semisimple modules.

Example 2 0042

If A is a field, then every A-module is semisimple by Theorem 1 of II, §7, No. 1, p. 292. This shows that, in general, a semisimple module decomposes in several ways into a direct sum of simple submodules (see, however, Corollary 2 of VIII, p. 68).

Example 3 0043

Let A be a principal ideal domain, and let P be a system of representatives consisting of irreducible elements of A (VII, §1, No. 3, p. 3). Let M be an A-module, and, for every π∈P\pi \in P, let M(π)M(\pi ) be the set of x∈Mx\in M such that πx=0\pi x= 0. By VII, §2, No. 2, p. 9, the A-module M is semisimple if and only if it is the sum of the submodules M(π)M(\pi ); it is then the direct sum of these submodules. This example will be generalized further on (VIII, p. 65).

Let A1A_1 and A2A_2 be algebras over a commutative ring K. In III, §4, No. 3, p. 466, we introduced the notion of left bimodule over the algebras A1A_1 and A2A_2 and showed that this notion is equivalent to that of a left module over the ring A1⊗KA2A_1\otimes_KA_2. We say that M is a simple (resp. semisimple, finitely generated) bimodule if it is a simple (resp. semisimple, finitely generated) module over the ring A1⊗KA2A_1\otimes_KA_2.

Theorem 1 0044

Let M be a module that is the (not necessarily direct) sum of a family (Si)i∈I(S_i)_{i\in I} of simple submodules, and let N be a submodule of M. There exists a subset J of I such that M is the direct sum of the family consisting of N and the modules SjS_j for jj running through J.

Let S\mathscr{S} be the set of subsets I′I' of I such that the sum of the family consisting of the modules N and SiS_i for ii in I′I' is direct. The set S\mathscr{S} is of finite character: a subset J of I belongs to S\mathscr{S} if and only if the same holds for every finite subset of J. Hence, the set S\mathscr{S} has a maximal element J (Set Theory, III, §4, No. 5, p. 171). Set N′=N+∑j∈JSjN'= N +\sum_{j\in J}S_j. Let ii be in I - J. Since J is maximal in S\mathscr{S}, the set J∪{i}J\cup \{i\} does not belong to S\mathscr{S}, so that Si∩N′≠0S_i\cap N'\not= 0. Since SiS_i is a simple module, we have Si∩N′=SiS_i\cap N'= S_i. We therefore have Si⊂N′S_i\subset N' for every i∈Ii\in I, whence N′=MN'= M. This completes the proof.

Corollary 1 0045

Every module that is the sum of a family of simple modules is semisimple.

It suffices to apply Theorem 1 to the case N = 0.

Corollary 2 0046

A module M is semisimple if and only if every submodule of M is a direct factor.

The condition is necessary by Theorem 1.

Conversely, suppose that every submodule of M admits a supplement. Let M′M' be the sum of the simple submodules of M, and let M′′M'' be a supplement of M′M' in M. Suppose that we have M′≠MM'\not= M and therefore M′′≠0M''\not= 0. Let N be a nonzero monogenous submodule of M′′M''. By Proposition 3 of VIII, p. 49, there exists a maximal submodule P of N. Let Q be a submodule supplementary to P in M. Then N∩QN\cap Q is a submodule of N, supplementary to P in N, hence isomorphic to N/PN/P (II, §1, No. 9, p. 210, Proposition 13). Consequently, N∩QN\cap Q is a simple submodule of M′′M'', contrary to the definition of M′M'.

We therefore have M′=MM'= M, and the module M is semisimple by Corollary 1.

Corollary 3 0047

Let M be a semisimple module and N a submodule of M. The modules N and M/NM/N are semisimple. More precisely, if M is the direct sum of a family (Si)i∈I(S_i)_{i\in I} of simple modules, then there exists a subset J of I such that M/NM/N is isomorphic to ⨁j∈JSj\bigoplus_{j\in J}S_j and N to ⨁i∈I−JSi\bigoplus_{i\in I-J}S_i.

Choose J as in Theorem 1. The module N′=⨁j∈JSjN'=\bigoplus_{j\in J}S_j is supplementary to N in M; it is therefore isomorphic to M/NM/N. Moreover, the submodules N and ⨁i∈I−JSi\bigoplus_{i\in I-J}S_i of M are both supplementary to N′N' and therefore isomorphic to M/N′M/N'.

Corollary 4 0048

Let M be a semisimple module. Then M is simple if and only if the endomorphism ring E of M is a field.

If M is simple, then E is a field by the corollary of Proposition 2 of VIII, p. 47.

If E is a field, then the module M is indecomposable (VIII, p. 31, Proposition 4, a)). Since it is moreover semisimple, it is simple.

Remark 0049

Let K be an algebraically closed commutative field and A a K-algebra. Let M be a semisimple A-module that is a finite-dimensional vector space over the field K. Then M is simple if and only if every endomorphism of the A-module M is of the form x↦αxx\mapsto \alpha x with α\alpha in K: this is necessary by Theorem 1 of VIII, p. 47, and sufficient by Corollary 4 above.

2. The homomorphism ⨁i\bigoplus_i HomA(M,Ni)⟶_A(M,N_i)\longrightarrow HomA(M,⨁iNi)_A(M,\bigoplus_iN_i)

Let A be a ring, M an A-module, and (Ni)i∈I(N_i)_{i\in I} a family of A-modules. With any element (ui)(u_i) of ⨁i\bigoplus_i HomA(M,Ni)_A(M,N_i), we associate the element m↦(ui(m))m\mapsto (u_i(m)) of HomA(M,⨁iNi)_A(M,\bigoplus_iN_i). We thus define a canonical homomorphism

φ:⨁i\varphi :\bigoplus_i HomA(M,Ni)⟶_A(M,N_i)\longrightarrow HomA(M,⨁iNi)_A(M,\bigoplus_iN_i).

It is clear that φ\varphi is injective. Let uu be an element of HomA(M,⨁iNi)_A(M,\bigoplus_iN_i). Then uu belongs to the image of φ\varphi if and only if the set of indices ii such that pri∘u≠0_i\circ u\not= 0 is finite. This condition is automatically satisfied when the module M is finitely generated.

Consequently, if the module M is finitely generated, then the homomorphism φ\varphi is bijective.

3. Some Operations on Modules

Let A and B be rings, and let P be an (A(A, B)-bimodule (II, §1, No. 14, p. 225). We will define two procedures, one to go from a left B-module to a left A-module, the other to go from a left A-module to a left B-module.

3.1. The operation T\mathscr{T}. — Let V be a left B-module. Denote the left A-module P⊗BVP\otimes_BV (II, §3, No. 4, p. 247) by T(V)\mathscr{T}(V). The law of action on T(V)\mathscr{T}(V) is given by the formula

a(p⊗v)=(ap)⊗v(1) a(p\otimes v) = (ap)\otimes v \tag{1}

for a∈A,p∈Pa\in A,p\in P, and v∈Vv\in V.

Let V′V' be a left B-module. For every B-linear mapping gg from V to V′V', the mapping 1P⊗g1_P\otimes g from T(V)\mathscr{T}(V) to T(V′)\mathscr{T}(V') is A-linear; we denote it by T(g)\mathscr{T}(g). The mapping g↦T(g)g\mapsto \mathscr{T}(g) from HomB(V,V′)_B(V,V') to HomA(T(V),T(V′))_A(\mathscr{T}(V),\mathscr{T}(V')) is Z\mathbf{Z}-linear, and we have

T(1V)=1T(V),T(g′∘g)=T(g′)∘T(g)(2) \mathscr{T}(1_V) = 1_{\mathscr{T}(V)},\mathscr{T}(g'\circ g) =\mathscr{T}(g')\circ \mathscr{T}(g) \tag{2}

if V,V′,V′′V, V', V'' are left B-modules and g:V→V′,g′:V′→V′′g: V\rightarrow V',g': V'\rightarrow V'' are B-linear mappings. Since the tensor product commutes with direct sums, if V is the direct sum of a family of submodules (Vi)i∈I(V_i)_{i\in I}, then we can identify the A-module T(V)\mathscr{T}(V) with ⨁iT(Vi)\bigoplus_i\mathscr{T}(V_i).

3.2. The operation H\mathscr{H}. — Let M be a left A-module. Denote the left B-module HomA(P,M)_A(P,M) (II, §1, No. 14, p. 225) by H(M)\mathscr{H}(M). The law of action on H(M)\mathscr{H}(M) is given by the formula

(bf)(p)=f(pb)(3) (bf)(p) =f(pb) \tag{3}

for b∈B,f∈b\in B,f\in HomA(P,M)_A(P,M), and p∈Pp\in P.

Let M′M' be a left A-module. For every A-linear mapping gg from M to M′M', the mapping HomA(1P,g)_A(1_P, g) from H(M)\mathscr{H}(M) to H(M′)\mathscr{H}(M') is B-linear; we denote it by H(g)\mathscr{H}(g). The mapping g↦H(g)g\mapsto \mathscr{H}(g) from HomA(M,M′)_A(M,M') to HomB(H(M),H(M′))_B(\mathscr{H}(M),\mathscr{H}(M')) is Z\mathbf{Z}-linear, and we have

H(1M)=1H(M),H(g′∘g)=H(g′)∘H(g)(4) \mathscr{H}(1_M) = 1_{\mathscr{H}(M)},\mathscr{H}(g'\circ g) =\mathscr{H}(g')\circ \mathscr{H}(g) \tag{4}

if M,M′,M′′M, M', M'' are left A-modules and g:M→M′,g′:M′→M′′g: M\rightarrow M',g': M'\rightarrow M'' are A-linear mappings. Suppose, moreover, that P is a finitely generated A-module; if M is the direct sum of a family of submodules (Mi)i(M_i)_i, then we can identify H(M)\mathscr{H}(M) with ⨁iH(Mi)\bigoplus_i\mathscr{H}(M_i) by VIII, p. 57. 3.3. Relations Between T\mathscr{T} and H\mathscr{H}. — By Proposition 1 of II, §4, No. 1, p. 268, for every left A-module M and every left B-module V, there exists a unique group isomorphism

(5) γ:\gamma : HomA(T(V),M)⟶_A(\mathscr{T}(V),M)\longrightarrow HomB(V,H(M))_B(V,\mathscr{H}(M))

characterized by the relation

(γ(h)(v))(p)=h(p⊗v)(6) (\gamma (h)(v))(p) =h(p\otimes v) \tag{6}

for h∈h\in HomA(T(V),M),v∈V_A(\mathscr{T}(V),M),v\in V, and p∈Pp\in P. The isomorphism γ\gamma is called the adjunction isomorphism.

Let M be a left A-module. The A-module T(H(M))\mathscr{T}(\mathscr{H}(M)) is simply the A-module P⊗BP\otimes_B HomA(P,M)_A(P,M). By applying the above to the B-module H\mathscr{H} (M), we see that the mapping

αM=γ−1\alpha_M=\gamma^{-1}(IdH(M)_{\mathscr{H}(M)}) :T(H(M))⟶M:\mathscr{T}(\mathscr{H}(M))\longrightarrow M

is the unique mapping satisfying

αM(p⊗f)=f(p)(7) \alpha_M(p\otimes f) =f(p) \tag{7}

for p∈Pp\in P and f∈f\in HomA(P,M)_A(P,M). We say that αM\alpha_M is the canonical A-linear mapping from T(H(M))\mathscr{T}(\mathscr{H}(M)) to M. For every A-linear mapping g:M→M′g: M\rightarrow M', we have a commutative diagram

T(H(M))αM\mathscr{T}(\mathscr{H}(M))^{\alpha_M} // M

(I) T(H(g))g\mathscr{T}_{(\mathscr{H}(g))}g

T(H(M′))αM′\mathscr{T}(\mathscr{H}(M'))^{\alpha_{M'}} // M′M'.

The inverse

γ−1:\gamma^{-1}: HomB(V,H(M))⟶_B(V,\mathscr{H}(M))\longrightarrow HomA(T(V),M)_A(\mathscr{T}(V),M)

of the adjunction isomorphism coincides with the mapping h↦αM∘T(h)h\mapsto \alpha_M\circ \mathscr{T}(h). Indeed, by (6) and (7), we have the relations

γ−1(h)(p⊗v)=(h(v))(p)=αM(p⊗h(v))=αM∘T(h)(p⊗v) \gamma^{-1}(h)(p\otimes v) = (h(v))(p) =\alpha_M(p\otimes h(v)) =\alpha_M\circ \mathscr{T}(h)(p\otimes v)

for all h∈h\in HomB(V,H_B(V,\mathscr{H} (M)), v∈Vv\in V, and p∈Pp\in P.

Let V be a B-module. The B-module H(T(V))\mathscr{H}(\mathscr{T}(V)) is simply the B-module HomA(P,P⊗BV)_A(P,P\otimes_BV). By applying (5) to the A-module T\mathscr{T} (V), we see that the B-linear mapping βV=γ\beta_V=\gamma(IdT(V)_{\mathscr{T}(V)}) from V to H(T(V))\mathscr{H}(\mathscr{T}(V)) is characterized by the relation

βV(v)(p)=p⊗v(8) \beta_V(v)(p) =p\otimes v \tag{8}

for p∈Pp\in P and v∈Vv\in V. We call βV\beta_V the canonical B-linear mapping from V to H(T(V))\mathscr{H}(\mathscr{T}(V)). For every B-module V′V' and every B-linear mapping g:V→V′g: V\rightarrow V', we have a commutative diagram

V βV^{\beta_V} //H(T(V))/\mathscr{H}(\mathscr{T}(V))

(II) gHTg\mathscr{H}_{\mathscr{T}}

( (g)(g))

V′βV′{V'}^{\beta_{V'}} //H(T(V′))/\mathscr{H}(\mathscr{T}(V')).

Note that the adjunction morphism (5) coincides with the mapping that sends uu to H(u)∘βV\mathscr{H}(u)\circ \beta_V. Indeed, from relations (6) and (8), we deduce the equalities

(γ(u)(v))(p)=u(p⊗v)=u∘(βV(v))(p) (\gamma (u)(v))(p) =u(p\otimes v) =u\circ (\beta_V(v))(p)

for all u∈u\in HomA(T(V),M),v∈V_A(\mathscr{T}(V),M),v\in V, and p∈Pp\in P.

Remark 1 004A

Let V and V′V' be B-modules. The adjunction isomorphism

γ:\gamma : HomA(T(V),T(V′))⟶_A(\mathscr{T}(V),\mathscr{T}(V'))\longrightarrow HomB(V,H(T(V′)))_B(V,\mathscr{H}(\mathscr{T}(V')))

satisfies the relation γ(T(f))=βV′∘f\gamma (\mathscr{T}(f)) =\beta_{V'}\circ f for every f∈f\in HomB(V,V′)_B(V,V') because

(γ(T(f))(v))(p)=T(f)(p⊗v)=p⊗f(v)=(βV′∘f)(v)(p) (\gamma (\mathscr{T}(f))(v))(p) =\mathscr{T}(f)(p\otimes v) =p\otimes f(v) = (\beta_{V'}\circ f)(v)(p)

Let M and M′M' be A-modules; the inverse of the adjunction isomorphism

γ−1:\gamma^{-1}: HomB(H(M),H(M′))⟶_B(\mathscr{H}(M),\mathscr{H}(M'))\longrightarrow HomA(T(H(M)),M′)_A(\mathscr{T}(\mathscr{H}(M)),M')

satisfies the relation γ−1(H(u))=u∘αM\gamma^{-1}(\mathscr{H}(u)) =u\circ \alpha_M for every u∈u\in HomB(M,M′)_B(M,M'). Indeed, we have the relations

γ−1(H(u))(p⊗v)=(H(u)(v))(p)=u(v(p))=u∘αM(p⊗v) \gamma^{-1}(\mathscr{H}(u))(p\otimes v) = (\mathscr{H}(u)(v))(p) =u(v(p)) =u\circ \alpha_M(p\otimes v)

for all u∈u\in HomB(M,M′),v∈H_B(M,M'),v\in \mathscr{H} (M), and p∈Pp\in P.

Remark 2 004B

Let M be a left A-module. The B-linear mappings

βH(M):H(M)→H(T(H(M)))\beta_{\mathscr{H}(M)}:\mathscr{H}(M)\rightarrow \mathscr{H}(\mathscr{T}(\mathscr{H}(M))) and H(αM):H(T(H(M)))→H(M)\mathscr{H}(\alpha_M) :\mathscr{H}(\mathscr{T}(\mathscr{H}(M)))\rightarrow \mathscr{H}(M) satisfy the relation H(αM)∘βH(M)=1H(M)\mathscr{H}(\alpha_M)\circ \beta_{\mathscr{H}(M)}= 1_{\mathscr{H}(M)}. They are not bijective in general.

Let V be a left B-module. The A-linear mappings

T(βV):T(V)→T(H(T(V)))\mathscr{T}(\beta_V) :\mathscr{T}(V)\rightarrow \mathscr{T}(\mathscr{H}(\mathscr{T}(V))) and αT(V):T(H(T(V)))→T(V)\alpha_{\mathscr{T}(V)}:\mathscr{T}(\mathscr{H}(\mathscr{T}(V)))\rightarrow \mathscr{T}(V) satisfy the relation αT(V)∘T(βV)=1T(V)\alpha_{\mathscr{T}(V)}\circ \mathscr{T}(\beta_V) = 1_{\mathscr{T}(V)}. They are not bijective in general.

Remark 3 004C

Suppose that P is finitely generated as an A-module. Let M be the direct sum of a family (Mi)i∈I(M_i)_{i\in I} of A-modules. The A-modules T(H(M))\mathscr{T}(\mathscr{H}(M)) and ⨁iT(H(Mi))\bigoplus_i\mathscr{T}(\mathscr{H}(M_i)) are canonically isomorphic. When we identify them, αM\alpha_M is identified with ⨁iαMi\bigoplus_i\alpha_{M_i}. Likewise, let V be the direct sum of a family (Vj)j∈J(V_j)_{j\in J} of B-modules. The B-module H(T(V))\mathscr{H}(\mathscr{T}(V)) is identified with ⨁jH(T(Vj))\bigoplus_j\mathscr{H}(\mathscr{T}(V_j)), and the linear mapping βV\beta_V with ⨁jβVj\bigoplus_j\beta_{V_j}.

4. Isotypical Modules

Let A be a ring and S a simple left A-module. Let D be the opposite ring of the endomorphism ring of S; it is a field. Endowed with the actions of A and D, S is an (A(A, D)-bimodule.

Proposition 1 004D

Let M be an A-module. The following properties are equivalent:

(i) There exists a set I such that M is isomorphic to S(I)S^{(I)}.

(ii) The module M is the direct sum of a family of submodules isomorphic to S.

(iii) The module M is the sum of a family of submodules isomorphic to S.

(iv) There exists a left vector space V over the field D such that the A-module M is isomorphic to S⊗DVS\otimes_DV.

The equivalence of (i) and (ii) is immediate, and that of (ii) and (iii) follows from Theorem 1 of VIII, p. 56, applied to the case N = 0. Every left vector space over D is isomorphic to a vector space of the form Ds(I)D_s^{(I)}, where I is a set (II, §7, No. 1, p. 292, Theorem 1). Since the tensor product commutes with direct sums, (i) is equivalent to (iv).

Definition 2 004E

An A-module M is isotypical of type S if it has the equivalent properties of Proposition 1. The module M is called isotypical if there exists a simple A-module T such that M is isotypical of type T.

Every isotypical module is semisimple.

Proposition 2 004F

If a module is the sum of isotypical submodules of type S, then it is isotypical of type S. The submodules and the quotient modules of an isotypical module of type S are isotypical of type S.

The first assertion follows from the definitions, the second from Corollary 3 of VIII, p. 56.

Remark 004G

Every nonzero isotypical module of type S has a quotient module and a submodule isomorphic to S; consequently, if M and M′M' are nonzero isotypical A-modules of type S, then the group HomA(M,M′)_A(M,M') is not reduced to 0.

Proposition 3 004H

a) Let M be an isotypical A-module of type S. The A-linear mapping αM:S⊗D\alpha_M: S\otimes_DHomA(S,M)→M_A(S,M)\rightarrow M characterized by αM(s⊗f)=f(s)\alpha_M(s\otimes f) =f(s) (VIII, p. 59) is bijective.

b) Let V be a left vector space over the field D. The D-linear mapping βV:V→\beta_V: V\rightarrow HomA(S,S⊗DV)_A(S,S\otimes_DV) defined by βV(v)(s)=s⊗v\beta_V(v)(s) =s\otimes v (VIII, p. 59) is bijective.

Denote the left D-vector space HomA(S,M)_A(S,M) by H(M)\mathscr{H}(M). The A-module M is, by assumption, the direct sum of a family of submodules isomorphic to S. The A-module S is monogenous; to prove that the mapping αM\alpha_M is bijective, it therefore suffices to consider the case M = S (VIII, p. 60, Remark 3). Now, H(S)\mathscr{H}(S) is simply the D-vector space DsD_s, and αS\alpha_S is simply the isomorphism ι:S⊗DDs→S\iota : S\otimes_DD_s\rightarrow S defined by ι(s⊗d)=sd\iota (s\otimes d) =sd.

Likewise, to prove b), it suffices to consider the case V=DsV = D_s. Since the mapping αS\alpha_S is bijective, the mapping βDs=βH(S)\beta_{D_s}=\beta_{\mathscr{H}(S)} is too (VIII, p. 60, Remark 2).

5. Description of an Isotypical Module

As in the previous subsection, A denotes a ring, S a simple left A-module, and D the field EndA(S)o_A(S)^o. We view S as an (A(A, D)-bimodule.

Definition 3 004I

Let M be an isotypical A-module of type S. A description of M with respect to S is a pair (V,α)(V, \alpha ), where V is a left vector space over the field D and α:S⊗DV→M\alpha : S\otimes_DV\rightarrow M is an isomorphism of A-modules.

Every isotypical A-module M of type S has a canonical description: it is the pair (HomA(S,M),αM_A(S,M), \alpha_M), where αM:S⊗D\alpha_M: S\otimes_D HomA(S,M)→M_A(S,M)\rightarrow M is the isomorphism of A-modules characterized by αM(s⊗f)=f(s)\alpha_M(s\otimes f) =f(s) (VIII, p. 62, Proposition 3, a)).

Theorem 2 004J

Let M be an isotypical A-module of type S and (V,α)(V, \alpha ) a description of M. Denote by DD(V)\mathscr{D}_D(V) the set, ordered by inclusion, of D-linear subspaces of V, and by DA(M)\mathscr{D}_A(M) that of A-submodules of M. For every W∈W\in DD(V)\mathscr{D}_D(V), identify the A-module S⊗DWS\otimes_DW with its canonical image in S⊗DVS\otimes_DV.

a) The mapping W↦α(S⊗DW)W\mapsto \alpha (S\otimes_DW) is an isomorphism of ordered sets from DD(V)\mathscr{D}_D(V) to DA(M)\mathscr{D}_A(M).

b) The inverse isomorphism sends a submodule N of M to the linear subspace of V consisting of the elements vv such that α(s⊗v)\alpha (s\otimes v) belongs to N for every s∈Ss\in S.

For W∈DDW\in \mathscr{D}_D(V), set φ(W)=α(S⊗DW)\varphi (W) =\alpha (S\otimes_DW). For N∈DAN\in \mathscr{D}_A(M), denote by ψ(N)\psi (N) the set of elements v∈Vv\in V such that α(s⊗v)∈N\alpha (s\otimes v)\in N for every s∈Ss\in S. This defines two mappings φ:DD(V)→DA(M)\varphi :\mathscr{D}_D(V)\rightarrow \mathscr{D}_A(M) and ψ:DA(M)→DD(V)\psi :\mathscr{D}_A(M)\rightarrow \mathscr{D}_D(V). They are clearly increasing.

Let N be a submodule of M. It is isotypical of type S (VIII, p. 61, Proposition 2). Set W=ψ(N)W =\psi (N). By Proposition 3, b) of VIII, p. 62, the A-linear mappings h:S→Mh: S\rightarrow M are simply the mappings s↦α(s⊗v)s\mapsto \alpha (s\otimes v), where vv runs through V. Those with image contained in N are the mappings s↦α(s⊗w)s\mapsto \alpha (s\otimes w) where ww runs through W; their images generate N because N is isotypical of type S. We therefore have α(S⊗DW)=N\alpha (S\otimes_DW) = N, that is, φ(ψ(N))=N\varphi (\psi (N)) = N. This proves that φ∘ψ\varphi \circ \psi is the identity mapping on DA(M)\mathscr{D}_A(M). In particular, φ\varphi is surjective and ψ\psi is injective.

To finish the proof, it suffices to prove that the mapping φ\varphi is injective. Let W and W′W' be linear subspaces of V such that φ(W)=φ(W′)\varphi (W) =\varphi (W'). The vector spaces S⊗DWS\otimes_DW and S⊗DW′S\otimes_DW' coincide when viewed as linear subspaces of S⊗DVS\otimes_DV. Choose a nonzero linear form ff on the D-vector space S, and let g:S⊗DV→Vg: S\otimes_DV\rightarrow V be the group homomorphism defined by g(s⊗v)=f(s)vg(s\otimes v) =f(s)v. We have W=g(S⊗DW)=g(S⊗DW′)=W′W =g(S\otimes_DW) =g(S\otimes_DW') = W', so φ\varphi is injective.

Remark 1 004K

Let M be an isotypical A-module of type S and (V,α)(V, \alpha ) a description of M. Then M has finite length if and only if V is finite-dimensional, and in this case, we have the relation

longA(M)=_A(M) = dimD(V)_D(V).

Corollary 1 004L

Let M be an isotypical A-module of type S. For every A-submodule N of M, identify HomA(S,N)_A(S,N) with the linear subspace of HomA(S,M)_A(S,M) consisting of the mappings with image contained in N.

a) The mapping N↦N\mapsto HomA(S,N)_A(S,N) is an isomorphism of ordered sets from DA(M)\mathscr{D}_A(M) to DD\mathscr{D}_D(HomA(S,M)_A(S,M)).

b) The inverse bijection sends a linear subspace W of HomA(S,M)_A(S,M) to the submodule ∑h∈Wh(S)\sum_{h\in W}h(S) of M.

This is a reformulation of Theorem 2 when we take (V,α)(V, \alpha ) to be the canonical description of M.

Corollary 2 004M

Let V be a left vector space over D and F\mathscr{F} a set of endomorphisms of V. An A-submodule of S⊗DVS\otimes_DV is stable under all endomorphisms 1S⊗u1_S\otimes u, where uu runs through F\mathscr{F}, if and only if it is of the form S⊗DWS\otimes_DW, where W is a linear subspace of V that is stable under all endomorphisms belonging to F\mathscr{F}.

Indeed, by Theorem 2, every A-submodule N of S⊗DVS\otimes_DV is equal to S⊗DWS\otimes_DW, where W is the linear subspace of V consisting of the elements vv such that s⊗vs\otimes v belongs to N for every s∈Ss\in S.

Theorem 3 004N

Let M and M′M' be isotypical A-modules of type S. Let (V,α)(V, \alpha ) and (V′,α′)(V', \alpha ') be descriptions of M and M′M', respectively. For any D-linear mapping f:V→V′f: V\rightarrow V', denote by f~:M→M′\widetilde{f}: M\rightarrow M' the unique A-linear mapping that makes the following diagram commute:

S⊗DVαS\otimes_DV^{\alpha} // M

1S⊗ff~(9) 1_{_S\otimes f}\widetilde{f} \tag{9}

S⊗DV′α′S\otimes_D{V'}^{\alpha'} // M′M'.

The mapping f↦f~f\mapsto \widetilde{f} from HomD(V,V′)_D(V,V') to HomA(M,M′)_A(M,M') is a group isomorphism.

It suffices to prove that the Z\mathbf{Z}-linear mapping u↦1S⊗uu\mapsto 1_S\otimes u from HomD(V,V′)_D(V,V') to HomA(S⊗DV,S⊗DV′)_A(S\otimes_DV,S\otimes_DV') is bijective. In the notation of No.3 applied to the (A(A, D)-bimodule S, this corresponds to showing that the mapping

T:\mathscr{T}: HomD(V,V′)⟶_D(V,V')\longrightarrow HomA(T(V),T(V′))_A(\mathscr{T}(V),\mathscr{T}(V'))

is bijective. But by Remark 1 of VIII, p. 60, since the adjunction isomorphism (VIII, p. 59) is bijective, this corresponds to showing that the mapping that sends uu to βV′∘u\beta_{V'}\circ u is bijective, which follows from the fact that the D-linear mapping βV′\beta_{V'} is bijective (VIII, p. 62, Proposition 3, b)).

We keep the notation of Theorem 3. Let M′′M'' be an isotypical A-module of type S, and let (V′′,α′′)(V'', \alpha '') be a description of M′′M''. For every f∈f\in HomD(V,V′)_D(V,V') and every g∈g\in HomD(V′,V′′)_D(V',V''), we have g]∘f=g~∘f~g]\circ f=\widetilde{g}\circ \widetilde{f}. In particular, for M=M′M = M', V=V′V = V', and α=α′\alpha =\alpha ', the mapping f↦f~f\mapsto \widetilde{f} from EndD(V)_D(V) to EndA(M)_A(M) is a ring isomorphism.

Remark 2 004O

Let M be an isotypical A-module of type S, and let (V,α)(V, \alpha ) be a description of M. Let B be a subring of the ring EndA(M)o_A(M)^o. The ring isomorphism from EndD(V)o_D(V)^o to EndA(M)o_A(M)^o endows V with the structure of a (D(D, B)-bimodule, so that α\alpha is an isomorphism of (A(A, B)-bimodules. There exists an isomorphism from the set of (D(D, B)-sub-bimodules of V, ordered by inclusion, to the set of (A(A, B)-sub-bimodules of M (VIII, p. 62, Theorem 2 and VIII, p. 63, Corollary 2).

Corollary 004P

Let M and M′M' be isotypical A-modules of type S. The mapping u↦u\mapsto Hom(1S,u_S, u) from HomA(M,M′)_A(M,M') to HomD_D(HomA(S,M)_A(S,M), HomA(S,M′)_A(S,M')) is a group isomorphism. When M is equal to M′M', it is a ring isomorphism from EndA(M)_A(M) to EndD_D(HomA(S,M)_A(S,M)).

Because of the commutativity of diagram (I) of VIII, p. 59, this corollary follows from Theorem 3, applied to the canonical descriptions of M and M′M'.

6. Isotypical Components of a Module

Definition 4 004Q

Let A be a ring, M an A-module, and S a simple A-module. The isotypical component of type S of M, denoted by MSM_S, is the sum of the submodules of M isomorphic to S.

It is clear that MSM_S is the greatest submodule of M that is isotypical of type S. Because every submodule of MSM_S is isotypical of type S (VIII, p. 61, Proposition 2), we have NS=MS∩NN_S= M_S\cap N for every submodule N of M.

If S′S' is a simple A-module isomorphic to S, then we clearly have MS=M_S= MS′M_{S'}, so MSM_S depends only on the class of S (VIII, p. 51).

Let M be an A-module. There exists a greatest semisimple submodule of M, called the socle of M; it is the sum of the simple submodules of M and also the sum of the isotypical components of M. In particular, M is semisimple if and only if it is equal to its socle.

Proposition 4 004R

Let A be a ring. Denote the set of classes of simple A-modules by S\mathscr{S}. Let M be a semisimple A-module.

a) The module M is the direct sum of the family (Mλ)λ∈S(M_{\lambda})_{\lambda\in\mathscr{S}} of its isotypical components.

b) Suppose that M is the direct sum of a family (Ni)i∈I(N_i)_{i\in I} of simple submodules. For every λ∈S\lambda \in \mathscr{S}, let I(λ)I(\lambda ) be the set of indices i∈Ii\in I such that NiN_i is of class λ\lambda. We have Mλ=⨁i∈I(λ)NiM_{\lambda}=\bigoplus_{i\in I(\lambda)}N_i.

c) If M is finitely generated, then the set of λ∈S\lambda \in \mathscr{S} such that Mλ≠0M_{\lambda}\not= 0 is finite.

d) For every submodule N of M and every λ∈S\lambda \in \mathscr{S}, we have Nλ=N∩MλN_{\lambda}= N\cap M_{\lambda} and (M/N)λ=(Mλ+N)/N(M/N)_{\lambda}= (M_{\lambda}+ N)/N.

Since M is semisimple, it is the sum of the family (Mλ)λ∈S(M_{\lambda})_{\lambda\in\mathscr{S}}; let us prove that this sum is direct. Let λ∈S\lambda \in \mathscr{S}. Denote the sum of the family (Mμ)μ∈S−{λ}(M_\mu)_{\mu\in\mathscr{S}-\{\lambda\}} by Mλ′M'_{\lambda}. The module Mλ′M'_{\lambda} is the direct sum of a family of simple modules not isomorphic to λ\lambda (VIII, p. 56, Theorem 1). By Corollary 3 of Theorem 1 of VIII, p. 56, Mλ′M'_{\lambda} does not contain any simple submodules of class λ\lambda. We consequently have Mλ∩Mλ′=0M_{\lambda}\cap M'_{\lambda}= 0. Assertion a) is therefore proved. By construction, we have Mλ⊃⨁i∈I(λ)NiM_{\lambda}\supset \bigoplus_{i\in I(\lambda)}N_i, so assertion b) follows from Remark 1 of II, §1, No. 8, p. 208.

Assertion c) follows from a) and Proposition 23 of II, §1, No. 12, p. 221.

Let N be a submodule of M and λ∈S\lambda \in \mathscr{S}. The isotypical component NλN_{\lambda} of type λ\lambda of N is contained in MλM_{\lambda} and Mλ∩N⊂NλM_{\lambda}\cap N\subset N_{\lambda}. The intersection N∩MλN\cap M_{\lambda} is therefore the isotypical component of N of type λ\lambda.

For every λ∈S\lambda \in \mathscr{S}, the module Mλ+N/NM_{\lambda}+N/N is isomorphic to Mλ/(Mλ∩N)M_{\lambda}/(M_{\lambda}\cap N). It is therefore isotypical of type λ\lambda and contained in (M/N)λ(M/N)_{\lambda}. The last assertion then follows from a) and II, §1, No. 8, p. 208, Remark 1.

Corollary 004S

Let A be a ring and S\mathscr{S} the set of classes of simple A-modules. Let M be a semisimple A-module and N a submodule of M. Then we have N=⨁λ∈SN∩MλN =\bigoplus_{\lambda\in\mathscr{S}}N\cap M_{\lambda} and M/N=⨁λ∈S(Mλ+N)/NM/N =\bigoplus_{\lambda\in\mathscr{S}}(M_{\lambda}+ N)/N.

Since N and M/NM/N are semisimple (VIII, p. 56, Corollary 3), the corollary follows from Proposition 4, d).

The support of a semisimple A-module M is the set of classes λ\lambda of simple A-modules such that the isotypical component of M of type λ\lambda is nonzero. The support of a finitely generated semisimple A-module is finite.

Proposition 5 004T

Let A be a ring, and let S\mathscr{S} be the set of classes of simple A-modules. Let M and N be A-modules.

a) Let f:M→Nf: M\rightarrow N be a homomorphism. For every λ∈S,f\lambda \in \mathscr{S},f induces a homomorphism fλf_{\lambda} from MλM_{\lambda} to NλN_{\lambda}; if M is semisimple and ff surjective, then each of the homomorphisms fλf_{\lambda} is surjective.

b) Suppose that M is semisimple. The mapping f↦(fλ)λ∈Sf\mapsto (f_{\lambda})_{\lambda\in\mathscr{S}} is a group isomorphism from HomA(M,N)_A(M,N) to ∏λ∈S\prod_{\lambda\in\mathscr{S}} HomA(Mλ,Nλ)_A(M_{\lambda},N_{\lambda}). When M is equal to N, the mapping is a ring isomorphism from EndA(M)_A(M) to ∏λ∈S\prod_{\lambda\in\mathscr{S}} EndA(Mλ)_A(M_{\lambda}).

For every λ∈S\lambda \in \mathscr{S}, the submodule f(Mλ)f(M_{\lambda}) of N is isomorphic to a quotient of an isotypical module of type λ\lambda; it is therefore isotypical of type λ\lambda and, consequently, contained in NλN_{\lambda}.

Suppose that M is semisimple and ff surjective. Then ff induces an isomorphism from M/M/ Ker(ff) to N that sends (Mλ(M_{\lambda}+Ker(ff))// Ker(ff) to f(Mλ)f(M_{\lambda}). By Proposition 4 of VIII, p. 65, we have Nλ=f(Mλ)N_{\lambda}=f(M_{\lambda}), which completes the proof of a).

The mapping considered in b) is clearly a group homomorphism, and it is a ring homomorphism when M is equal to N. Let (fλ)λ∈S(f_{\lambda})_{\lambda\in\mathscr{S}} be an element of ∏λ∈S\prod_{\lambda\in\mathscr{S}} Hom(Mλ,Nλ_{\lambda},N_{\lambda}). The unique element of its inverse image under the mapping in b) is the homomorphism f:M→Nf: M\rightarrow N defined by

f((∑λ∈Sxλ)=∑λ∈Sfλ(xλ) f((\sum_{\lambda\in\mathscr{S}}x_{\lambda})=\sum_{\lambda\in\mathscr{S}}f_{\lambda}(x_{\lambda})

for every (xλ)λ∈S∈⨁λMλ(x_{\lambda})_{\lambda\in\mathscr{S}}\in \bigoplus_{\lambda}M_{\lambda}.

Remark 004U

Let A and B be rings. Let M be an (A(A, B)-bimodule. It follows from Proposition 5 that the isotypical components of the A-module M are sub-bimodules of M. This holds, in particular, when M is an A-module and B is the opposite ring of EndA(M)_A(M).

Example 004V

Let us consider the case when the ring A is commutative. The mapping that sends a maximal ideal m\mathfrak{m} onto cl(A/m/\mathfrak{m}) is a bijection from the set of maximal ideals of A to the set S\mathscr{S} of classes of simple A-modules (VIII, p. 51). The inverse bijection sends λ\lambda to its annihilator mλ\mathfrak{m}_{\lambda}.

Let M be an A-module. For every λ∈S\lambda \in \mathscr{S}, the isotypical component MλM_{\lambda} of type λ\lambda of M consists of the elements annihilated by mλ\mathfrak{m}_{\lambda}, and we can view MλM_{\lambda} as a vector space over the field A/mλA/\mathfrak{m}_{\lambda}. If M is semisimple and N is another A-module, then we can deduce a group isomorphism from HomA(M,N)_A(M,N) to ∏λ∈S\prod_{\lambda\in\mathscr{S}} HomA/mλ(Mλ,Nλ)_{A/\mathfrak{m}_{\lambda}}(M_{\lambda},N_{\lambda}) from Proposition 5.

Corollary 1 004W

Let M be a semisimple A-module and N a submodule of M. The following properties are equivalent:

(i) There exists a unique submodule supplementary to N in M.

(ii) We have HomA(M/N,N)=0_A(M/N,N) = 0.

(iii) There exists a subset Λ of S\mathscr{S} such that N=⨁λ∈ΛMλN =\bigoplus_{\lambda\in\Lambda}M_{\lambda}.

Choose a submodule N′N' supplementary to N in M (VIII, p. 56, Corollary 2). If we identify M with N′×NN'\times N, then the submodules of M supplementary to N are the graphs of the A-linear mappings from N′N' to N. Since N′N' is isomorphic to M/NM/N, we have proved the equivalence of properties (i) and (ii). By Proposition 5, b), the group HomA(N′,N)_A(N',N) is isomorphic to the group ∏λ∈S\prod_{\lambda\in\mathscr{S}} HomA(Nλ′,Nλ)_A(N'_{\lambda},N_{\lambda}). It is zero if and only if for every λ∈S\lambda \in \mathscr{S}, we have Nλ=0N_{\lambda}= 0 or Nλ′=0N'_{\lambda}= 0 (VIII, p. 61, Remark), that is, Nλ=0N_{\lambda}= 0 or Nλ=MλN_{\lambda}= M_{\lambda}. This proves the equivalence of properties (ii) and (iii).

Corollary 2 00R0

Let M be an A-module. The following two conditions are equivalent:

(i) Every submodule of M admits a unique supplementary submodule.

(ii) The module M is the direct sum of a family (Si)i∈I(S_i)_{i\in I} of simple, pairwise nonisomorphic modules.

Suppose that M satisfies these conditions. Then, for every submodule N of M, there exists a unique subset J of I such that we have N=⨁j∈JSjN =\bigoplus_{j\in J}S_j and every simple submodule of M is equal to one of the SiS_i.

Conditions (i) and (ii) both imply that M is semisimple.

Suppose that condition (i) is satisfied. Let λ∈S\lambda \in \mathscr{S}. By the equivalence of (i) and (iii) in Corollary 1, every submodule of MλM_{\lambda} is zero or equal to MλM_{\lambda}. Consequently, MλM_{\lambda} is zero or simple, and (ii) follows from the fact that M is the direct sum of the family (Mλ)λ∈S(M_{\lambda})_{\lambda\in\mathscr{S}}.

Conversely, if condition (ii) is satisfied, then MλM_{\lambda} is zero or simple for every λ∈S\lambda \in \mathscr{S}. If N is a submodule of M, then N∩MλN\cap M_{\lambda} is equal to 0 or MλM_{\lambda} for every λ∈S\lambda \in \mathscr{S}. Since we have N=⨁λ∈S(N∩Mλ)N =\bigoplus_{\lambda\in\mathscr{S}}(N\cap M_{\lambda}) (VIII, p. 66, Corollary), the submodule N has property (iii) of Corollary 1 and therefore admits a unique supplementary submodule in M. This proves that (ii) implies (i), as well as the last assertions of the corollary.

Let M be an A-module and S a simple A-module. Denote the opposite ring of the field EndA(S)_A(S) by D, and view S as an (A(A, D)-bimodule. Then HomA(S,M)_A(S,M) is a left vector space over D, and HomA(M,S)_A(M,S) is a right vector space over D. The dual of the left D-vector space HomA(S,M)_A(S,M) is a right vector space over D (II, §2, No. 3, p. 232, Definition 2). For every u∈u\in HomA(M,S)_A(M,S), the mapping h(u):v↦u∘vh(u) :v\mapsto u\circ v from HomA(S,M)_A(S,M) to HomA(S,S)=D_A(S,S) = D is a linear form on the left D-vector space HomA(S,M)_A(S,M).

Proposition 6 004X

Keep the notation above, and suppose that the A-module M is semisimple. The mapping u↦h(u)u\mapsto h(u) from the right D-vector space HomA(M,S)_A(M,S) to the dual of the left D-vector space HomA(S,M)_A(S,M) is D-linear and bijective.

Let u∈u\in HomA(M,S),v∈_A(M,S),v\in HomA(S,M)_A(S,M), and d∈Dd\in D. We have

h(ud)(v)=h(d∘u)(v)=d∘u∘v=d∘(h(u)(v))=h(u)(v)d h(ud)(v) =h(d\circ u)(v) =d\circ u\circ v=d\circ (h(u)(v)) =h(u)(v)d

This proves that the mapping hh is D-linear. It is simply the mapping given by u↦u\mapsto Hom(1S,u_S, u) from HomA(M,S)_A(M,S) to HomD_D(HomA(S,M)_A(S,M), HomA(S,S)_A(S,S)). To prove that it is bijective, by Proposition 5, b) of VIII, p. 66, it suffices to treat the case when M is isotypical of type S; we can then apply the corollary of VIII, p. 65.

7. Description of a Semisimple Module

For the remainder of this section, A is a ring, and S\mathscr{S} is the set of classes of simple A-modules. For every λ∈S\lambda \in \mathscr{S}, we choose a simple module SλS_{\lambda} of class λ\lambda (for example, Sλ=λS_{\lambda}=\lambda ); we denote the opposite ring of the field of endomorphisms of SλS_{\lambda} by DλD_{\lambda}. We view SλS_{\lambda} as an (A,Dλ(A,D_{\lambda})-bimodule.

Let M be an A-module. For every λ∈S\lambda \in \mathscr{S}, HomA(Sλ,M)_A(S_{\lambda},M) is a left vector space over the field DλD_{\lambda}. By VIII, p. 59, and II, §1, No. 6, p. 202, Proposition 6, there exists a unique A-linear mapping, called canonical,

αM:⨁λ∈S(Sλ⊗Dλ\alpha_M:\bigoplus_{\lambda\in\mathscr{S}}(S_{\lambda}\otimes_{D_{\lambda}} HomA(Sλ,M))→M_A(S_{\lambda},M))\rightarrow M

satisfying the relation

αM(s⊗f)=f(s)(10) \alpha_M(s\otimes f) =f(s) \tag{10}

for λ∈S,s∈Sλ\lambda \in \mathscr{S},s\in S_{\lambda}, and f∈f\in HomA(Sλ,M)_A(S_{\lambda},M). If we endow ⨁λ∈S(Sλ⊗Dλ\bigoplus_{\lambda\in\mathscr{S}}(S_{\lambda}\otimes_{D_{\lambda}} HomA(Sλ,M))_A(S_{\lambda},M)) and M with their natural structures of EndA_A(M)-modules, then the mapping αM\alpha_M is EndA_A(M)-linear.

Proposition 7 004Y

Let M be an A-module. The canonical mapping αM\alpha_M is injective. For every λ∈S\lambda \in \mathscr{S}, the image under αM\alpha_M of Sλ⊗DλS_{\lambda}\otimes_{D_{\lambda}} HomA(Sλ,M)_A(S_{\lambda},M) is the isotypical component of M of type λ\lambda. The image of αM\alpha_M is the socle of M. The A-module M is semisimple if and only if the mapping αM\alpha_M is bijective.

Let λ∈S\lambda \in \mathscr{S}. Denote the isotypical component of M of type λ\lambda by MλM_{\lambda}. Every A-linear mapping from SλS_{\lambda} to M takes its values in MλM_{\lambda} (VIII, p. 66, Proposition 5). Consequently, by Proposition 3, a) of VIII, p. 62, the mapping αM\alpha_M induces a bijection from Sλ⊗DλS_{\lambda}\otimes_{D_{\lambda}} HomA(Sλ,M)_A(S_{\lambda},M) to MλM_{\lambda}. The proposition follows because the socle of M is the direct sum of the family (Mλ)λ∈S(M_{\lambda})_{\lambda\in\mathscr{S}} and the module M is semisimple if and only if it is equal to its socle.

Definition 5 004Z

Let M be a semisimple A-module. A description of M (with respect to the family (Sλ)λ∈S(S_{\lambda})_{\lambda\in\mathscr{S}}) is a pair ((Vλ)λ∈S,α)((V_{\lambda})_{\lambda\in\mathscr{S}}, \alpha ), where VλV_{\lambda} is a left vector space over the field DλD_{\lambda} for each λ∈S\lambda \in \mathscr{S} and α:⨁λ∈S(Sλ⊗DλVλ)→M\alpha :\bigoplus_{\lambda\in\mathscr{S}}(S_{\lambda}\otimes_{D_{\lambda}}V_{\lambda})\rightarrow M is an isomorphism of A-modules.

By Proposition 7, every semisimple A-module M has a canonical description: it is the pair ((HomA(Sλ,M)_A(S_{\lambda},M))λ∈S,αM_{\lambda\in\mathscr{S}}, \alpha_M), where αM\alpha_M is the A-linear mapping defined by formula (10).

Proposition 8 0050

Let M be a semisimple A-module and ((Vλ)λ∈S,α)((V_{\lambda})_{\lambda\in\mathscr{S}}, \alpha ) a description of M.

a) For every λ∈S\lambda \in \mathscr{S}, the mapping α\alpha induces an isomorphism from the A-module Sλ⊗DλVλS_{\lambda}\otimes_{D_{\lambda}}V_{\lambda} to the isotypical component of M of type λ\lambda.

b) For every λ∈S\lambda \in \mathscr{S}, the mapping βλ:Vλ→\beta_{\lambda}: V_{\lambda}\rightarrow HomA(Sλ,M)_A(S_{\lambda},M) defined by βλ(v)(s)=α(s⊗v)\beta_{\lambda}(v)(s) =\alpha (s\otimes v) is DλD_{\lambda}-linear and bijective.

c) Let N be a submodule of M. There exists a unique family (Wλ)λ∈S(W_{\lambda})_{\lambda\in\mathscr{S}} with the following properties: WλW_{\lambda} is a DλD_{\lambda}-linear subspace of VλV_{\lambda} for every λ∈S\lambda \in \mathscr{S}, and N is the image under α\alpha of the module ⨁λ∈S(Sλ⊗DλWλ)\bigoplus_{\lambda\in\mathscr{S}}(S_{\lambda}\otimes_{D_{\lambda}}W_{\lambda}) identified with its canonical image in the module ⨁λ∈S(Sλ⊗DλVλ)\bigoplus_{\lambda\in\mathscr{S}}(S_{\lambda}\otimes_{D_{\lambda}}V_{\lambda}). For every λ∈S,Wλ\lambda \in \mathscr{S},W_{\lambda} is the set of elements v∈Vλv\in V_{\lambda} such that α(s⊗v)\alpha (s\otimes v) belongs to N for every s∈Sλs\in S_{\lambda}.

The A-module Sλ⊗DλVλS_{\lambda}\otimes_{D_{\lambda}}V_{\lambda} is isotypical of type λ\lambda for every λ∈S\lambda \in \mathscr{S}. Assertion a) then follows from the facts that α\alpha is an isomorphism and that M is the direct sum of the family (Mλ)λ∈S(M_{\lambda})_{\lambda\in\mathscr{S}} (VIII, p. 65, Proposition 4, a)).

Let λ∈S\lambda \in \mathscr{S}. Denote the restriction of α\alpha to Sλ⊗DλVλS_{\lambda}\otimes_{D_{\lambda}}V_{\lambda} by αλ:Sλ⊗DλVλ→\alpha_{\lambda}: S_{\lambda}\otimes_{D_{\lambda}}V_{\lambda}\rightarrow M. With the notation of No. 3 applied to the (A,Dλ(A,D_{\lambda})-bimodule SλS_{\lambda}, we have

βλ=γ(αλ)=H(αλ)∘βVλ \beta_{\lambda}=\gamma (\alpha_{\lambda}) =\mathscr{H}(\alpha_{\lambda})\circ \beta_{V_{\lambda}}

where the last equality follows from VIII, p. 60. Hence, βλ\beta_{\lambda} is the composition of the DλD_{\lambda}-linear homomorphism H(αλ)\mathscr{H}(\alpha_{\lambda}) and βVλ\beta_{V_{\lambda}}. Assertion b) then follows from Proposition 3, b) of VIII, p. 62.

Let N be a submodule of M. We have N=⨁λ∈S(N∩Mλ)N =\bigoplus_{\lambda\in\mathscr{S}}(N\cap M_{\lambda}) (VIII, p. 66, Corollary), so c) follows from Theorem 2 of VIII, p. 62.

Corollary 0051

Let M be a semisimple A-module. For every submodule N of M and every element λ\lambda of S\mathscr{S}, identify HomA(Sλ,N)_A(S_{\lambda},N) with the DλD_{\lambda}-linear subspace of HomA(Sλ,M)_A(S_{\lambda},M) consisting of the mappings with image contained in N.

a) The mapping N↦N\mapsto (HomA(Sλ,N)_A(S_{\lambda},N))λ∈S_{\lambda\in\mathscr{S}} is a bijection from the set of A-submodules of M to the set of families (Wλ)λ∈S(W_{\lambda})_{\lambda\in\mathscr{S}} such that for every λ∈S\lambda \in \mathscr{S}, WλW_{\lambda} is a DλD_{\lambda}-linear subspace of HomA(Sλ,M)_A(S_{\lambda},M).

b) The inverse bijection sends a family (Wλ)λ∈S(W_{\lambda})_{\lambda\in\mathscr{S}} to the A-submodule ∑λ∈S∑w∈Wλw(Sλ)\sum_{\lambda\in\mathscr{S}}\sum_{w\in W_{\lambda}}w(S_{\lambda}) of M.

This is a reformulation of Proposition 8, c) applied to the canonical description of M.

Proposition 9 0052

Let M and M′M' be semisimple A-modules, and consider descriptions ((Vλ)λ∈S,α)((V_{\lambda})_{\lambda\in\mathscr{S}}, \alpha ) and ((Vλ′)λ∈S,α′)((V'_{\lambda})_{\lambda\in\mathscr{S}}, \alpha ') of M and M′M', respectively. For every family f=(fλ)λ∈S\boldsymbol{f}= (f_{\lambda})_{\lambda\in\mathscr{S}} in ∏λ∈S\prod_{\lambda\in\mathscr{S}} HomDλ(Vλ,Vλ′)_{D_{\lambda}}(V_{\lambda},V_{\lambda}'), there exists a unique A-linear mapping φ(f)∈\varphi (\boldsymbol{f})\in HomA(M,M′)_A(M,M') for which the following diagram commutes:

⨁(Sλ⊗DVλ)α\bigoplus(S_{\lambda}\otimes_DV_{\lambda})^{\alpha} // M

λ∈Sλ \lambda \in \mathscr{S}\lambda
⨁(1S⊗fλ) \bigoplus(1_S\otimes f_{\lambda})
λφ(f) _{\lambda}\varphi (\boldsymbol{f})
λ∈S\lambda \in \mathscr{S}

⨁(Sλ⊗DVλ′)α′\bigoplus(S_{\lambda}\otimes_DV'_{\lambda})^{\alpha'} // M′M'.

λ∈Sλ\lambda \in \mathscr{S}\lambda

The mapping φ:∏λ∈S\varphi :\prod_{\lambda\in\mathscr{S}} HomDλ(Vλ,Vλ′)→_{D_{\lambda}}(V_{\lambda},V'_{\lambda})\rightarrow HomA(M,M′)_A(M,M') defined this way is a group isomorphism. When we have M=M′,Vλ=Vλ′M = M',V_{\lambda}= V'_{\lambda} for every λ∈S\lambda \in \mathscr{S}, and α=α′\alpha =\alpha ', the mapping φ\varphi is a ring isomorphism from ∏λ∈S\prod_{\lambda\in\mathscr{S}} EndDλ(Vλ)_{D_{\lambda}}(V_{\lambda}) to EndA(M)_A(M).

In view of the description of the isotypical components of M and M′M' given in Proposition 8, a), this follows from Theorem 3 of VIII, p. 64 and Proposition 5, b) of VIII, p. 66.

Corollary 0053

Let M be a semisimple A-module and M′M' an A-module. The mapping u↦u\mapsto (Hom(1Sλ,u_{S_{\lambda}}, u))λ∈S_{\lambda\in\mathscr{S}} from HomA(M,M′)_A(M,M') to

∏λ∈S\prod_{\lambda\in\mathscr{S}} HomDλ_{D_{\lambda}}(HomA(Sλ,M)_A(S_{\lambda},M), HomA(Sλ,M′)_A(S_{\lambda},M'))

is a group isomorphism. When M′M' is equal to M, it is an isomorphism from the ring EndA(M)_A(M) to the ring ∏λ∈S\prod_{\lambda\in\mathscr{S}} EndDλ_{D_{\lambda}}(HomA(Sλ,M)_A(S_{\lambda},M)).

This is a reformulation of Proposition 9 applied to the canonical descriptions of M and of the socle of M′M'.

8. Multiplicities and Lengths in Semisimple Modules

Proposition 10 0054

Let M be a semisimple A-module. Let (Mi)i∈I(M_i)_{i\in I} be a family of simple submodules with direct sum M. The following properties are equivalent:

(i) M has finite length.

(ii) M is Artinian.

(iii) M is Noetherian.

(iv) M is finitely generated.

(v) I is finite.

If M has these properties, then the length of M is equal to the cardinal of I.

If the set I is finite, then M has properties (i), (ii), (iii), and (iv). Suppose that the set I is infinite. By Example 2 of VIII, p. 2, the module M is neither Artinian nor Noetherian; because every module of finite length is Artinian and Noetherian (VIII, p. 2, Proposition 1), M also does not have finite length. Finally, every element of M belongs to the sum of a finite number of submodules MiM_i, so M is not finitely generated. This proves the equivalence of properties (i) through (v). If these hold, then we have long(M) =∑i∈I=\sum_{i\in I} long(Mi_i) == Card(I) (II, §1, No. 10, p. 213, Corollary 5).

Proposition 11 0055

Let M be a semisimple A-module that is the direct sum of a family (Mi)i∈I(M_i)_{i\in I} of simple submodules. For every λ∈S\lambda \in \mathscr{S}, we denote by I(λ)I(\lambda ) the set of indices i∈Ii\in I such that MiM_i is of class λ\lambda. The cardinal of I(λ)I(\lambda ) is equal to the dimension of the left DλD_{\lambda}-vector space HomA(Sλ,M)_A(S_{\lambda},M).

The isotypical component of A of type λ\lambda is isomorphic to Sλ(I(λ))S_{\lambda}^{(I(\lambda))} (VIII, p. 65, Proposition 4, b)). The DλD_{\lambda}-vector space HomA(Sλ,M)_A(S_{\lambda},M) may be identified with HomA(Sλ,Mλ)_A(S_{\lambda},M_{\lambda}), so is isomorphic to Dλ(I(λ))D_{\lambda}^{(I(\lambda))} (No. 2). This proves the proposition.

Every simple module is primordial (VIII, p. 45), so every semisimple module is semiprimordial. Let M be a semisimple A-module, and let λ∈S\lambda \in \mathscr{S}. The multiplicity of λ\lambda in M is the primordial multiplicity [M:λ][M :\lambda ] of λ\lambda in M defined in VIII, p. 34. Proposition 11 translates to the equality

(11) [M:λ]=[M :\lambda ] = dimDλ_{D_{\lambda}}(HomA(Sλ,M)_A(S_{\lambda},M)).

More generally, if ((Vλ)λ∈S,α)((V_{\lambda})_{\lambda\in\mathscr{S}}, \alpha ) is a description of M, then [M:λ][M :\lambda ] is equal to dimDλ(Vλ)_{D_{\lambda}}(V_{\lambda}). By Proposition 6 of VIII, p. 68, we also have

(12) [M:λ]=[M :\lambda ] = dimDλ_{D_{\lambda}}(HomA(M,Sλ)_A(M,S_{\lambda}))

when the multiplicity [M:λ][M :\lambda ] is finite. Semisimple A-modules M and M′M' are isomorphic if and only if we have [M:λ]=[M′:λ][M :\lambda ] = [M':\lambda ] for every λ∈S\lambda \in \mathscr{S}.

Let M be a semisimple A-module. There exists a cardinal I\mathbf{I} that has the following property: for every decomposition M=⨁i∈IMiM =\bigoplus_{i\in I}M_i of M into a direct sum of simple modules, the cardinal of I is equal to I\mathbf{I} (VIII, p. 34, Corollary 2). This cardinal is called the length of the semisimple A-module M and denoted by longA(M)_A(M) or long(M). When M has finite length, this definition is compatible with that of II, §1, No. 10, p. 212, by Proposition 10.

The simple A-modules are the semisimple A-modules of length 1, and we have the formula

(13) longA(⨁j∈JMj)=∑j∈J_A(\bigoplus_{j\in J}M_j)=\sum_{j\in J} longA(Mj)_A(M_j)

for every family (Mj)j∈J(M_j)_{j\in J} of semisimple A-modules. By Proposition 11, we have

(14) longA(M)=∑λ∈S_A(M) =\sum_{\lambda\in\mathscr{S}} dimDλ_{D_{\lambda}} HomA(Sλ,M)_A(S_{\lambda},M).

By applying this formula to MλM_{\lambda}, we obtain [M:λ]=[M :\lambda ] = longA(Mλ)_A(M_{\lambda}) for every λ∈S\lambda \in \mathscr{S}.

When A is a field, the simple A-modules are the vector spaces of dimension 1; every A-module is then semisimple (II, §7, No. 1, p. 292, Theorem 1), and its length is simply its dimension as a vector space over A (II, §7, No. 2, p. 293) .

Exercises

See the exercises for § 4.

[^1]: By Corollary 2 of VIII, p. 56, this definition coincides with that given in VII, §2, No. 2, p. 9

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