• Keine Ergebnisse gefunden

14 Semiprime modules

Im Dokument Modules and Algebras ... (Seite 116-135)

1.Lemma. 2.Properties of trace and torsion submodules. 3.Strongly semiprime mod-ules. 4.Characterizations. 5.SSP and semisimple modmod-ules. 6.Properly semiprime modules. 7.Properties of PSP-modules. 8.Corollary. 9.Relation to strongly prime modules. 10.Polyform SSP modules. 11.Polyform PSP modules. 12.Self-injective PSP modules. 13.Self-injective finitely presented PSP modules. 14.Density property of the self-injective hull. 15.Duo endomorphism rings. 16.Projective PSP modules.

17.Projective SSP modules. 18.Torsion submodules of semiprime rings. 19.Left SSP rings. 20.Left PSP rings. 21.Pseudo regular modules. 22.Left fully idempotent rings.

23.Pseudo regular and P SP modules. 24.Remarks. 25.Exercises.

In this section we extend the notion of semiprimeness from (commutative) rings to modules. We refer to 14.24 for a list of various possibilities to do so. In view of our applications we are mainly interested in non-projective modules.

The two following technical lemmas will be crucial for our investigations.

14.1 Lemma. Let M be an A-module and K, L⊂M. The following are equivalent:

(a) M/L∈σ[K];

(b) for any b ∈M, there exists a finite subset X ⊂K, with AnA(X)b⊂L.

Proof. (a) ⇒ (b) Assume M/L ∈ σ[K] and b ∈ M. Then Ab+L/L ⊂ M/L is a cyclic module in σ[K] and hence a factor module of a cyclic submodule of K(IN). So there existx1, . . . , xk∈K and a morphism

A(x1, . . . , xk)→M/L, a(x1, . . . , xk)7→a(b+L)/L.

This impliesAnA(x1, . . . , xk)b ⊂L.

(b) ⇒ (a) Let b ∈ M and choose x1, . . . , xk ∈K with AnA(x1, . . . , xk)b ⊂ L. We define a map as given above and so (Ab+L)/L∈σ[K]. Hence M/L∈σ[K]. 2

We use the notation for trace and torsion submodules introduced in 11.9.

14.2 Properties of trace and torsion submodules.

Let M be an A-module, K ⊂M andT =EndA(Mc). The following are equivalent:

(a) M/TK(M)∈σ[K];

(b) for any b ∈M, there exists a finite subset X ⊂K, with AnA(X)b⊂ TK(M).

(c) everyK-injective,TK-torsionfree module inσ[M]isM-injective andK-generated;

(d) KT is M-injective and TK(M) +TK(M)M;

(e) Mc=KT ⊕IM(TK(Mc)), where IM denotes the M-injective hull.

Notice that the decomposition of Mc given in (e) is in A-Mod. Though KT obvi-ously is a fully invariant submodule this need not be true for IM(TK(Mc)).

Proof. (a)⇔(b) follows from 14.1.

(a)⇒(c) AssumeQ∈σ[M] isK-injective andTK-torsionfree. For any submodule L ⊂ M, consider a morphism f : L → Q. Since TK(Q) = 0, f factorizes through f0 :L/TK(L)→Q. We have the commutative diagram (with canonical mappings)

0 → L → M

↓ ↓

0 → L/TK(L) → M/TK(M)

f0 Q

Since Q is K-injective and M/TK(M) ∈ σ[K], there exists some M/TK(M) → Q yielding a commutative diagram. HenceQ is M-injective and

Q=T r(M, Q) = T r(M/TK(M), Q) =T r(K, Q).

(c)⇒(a) SinceIM(M/TK(M)) isM-injective andTK-torsionfree, it isK-generated by (c) and soM/TK(M)∈σ[K].

(a) ⇒ (d) As shown above, IM(M/TK(M)) ∈ σ[K]. For a complement U of TK(M) in M, U ⊕ TK(M)M and U is isomorphic to a submodule of M/TK(M).

Hence U ⊂ TK(M) and TK(M) +TK(M)M.

(d)⇒(a) SinceTK(M) +TK(M)M and TK(M)∩ TK(M) = 0, TK(M)'[TK(M) +TK(M)]/TK(M)M/TK(M) . Hence M/TK(M) is isomorphic to a submodule of KTd ∈σ[K].

(d)⇒(e) The assumptions implyMc =KTd ⊕IM(TK(M)). As an injective object, KTd ∈σ[K] is K-generated and hence KTd =KT.

Since TK(M)TK(Mc), KTd =IM(TK(Mc)).

(e)⇒(d) As a direct summand ofMc,KT is M-injective. HenceTK(M)KTd = KT. SinceTK(M)TK(Mc)IM(TK(Mc)), we concludeTK(M) +TK(M)M. 2

Referring to the above relations we define:

14.3 Strongly semiprime modules.

Let M be an A-module and T =EndA(M).c M is called strongly semiprime (SSP) if it satisfies the following equivalent conditions for every submodule K ⊂M:

(a) M/TK(M)∈σ[K];

(b) for any b ∈M, there exists a finite subset X ⊂K, with AnA(X)b⊂ TK(M);

(c) everyK-injectiveTK-torsionfree module inσ[M]isM-injective andK-generated;

(d) KTd ∈σ[K] and TK(M) +TK(M)M;

(e) Mc=KT ⊕IM(TK(Mc)), where IM denotes the M-injective hull.

Any SP module is SSP (Mc = KT). Also every semisimple module is SSP. We state some basic important properties.

14.4 Characterizations.

Let M be an A-module, S = EndA(M) and T = EndA(M). Then the followingc are equivalent:

(a) AM is SSP;

(b) for any N AM, M ∈σ[N], and for any K ⊂AM, TK(M) +TK(M)AM; (c) AMc is SSP;

(d) Mc is a semisimple (A, T)-bimodule.

Proof. (a)⇒(b) ForNM,TN(M)∩TN(M) = 0 impliesTN(M) = 0 andN T =M,c in particularM ∈σ[N].

(b) ⇒ (a) Let K ⊂ M be any submodule. Since TK(M) +TK(M)M, M ∈ σ[TK(M) +TK(M)] =σ[M] by assumption. SoTK(M)+Kis a subgenerator inσ[M]

and hence it generates theM-injective moduleKTd. However,HomA(TK(M),KTd) = 0 implies thatKTd is K-generated and M is an SSP module.

(a) ⇒ (c) For any submodule N ⊂ Mc, put K = N T ∩M. Then KT N T and TK(Mc) = TN(Mc). Consider any N-injective and TN-torsionfree module Q ∈ σ[M].

Then Q is K-injective and TK-torsionfree, and hence M-injective and K-generated since M is SSP (cf. 14.3). As easily seen, Qis also N-generated and soMc is SSP by 14.3.

(c)⇒(a) Essential submodules of SSP modules are obviously SSP.

(c)⇔(d) Put M =Mc. Let U ⊂M be an essential (A, S)-submodule.

Then U = TU(M), and TU(M)∩ TU(M) = 0 implies TU(M) = 0. We see from 14.3 that U = M. So M has no proper essential (A, S)-submodule. Hence it is a semisimple (A, S)-module.

Now assumeM is a semisimple (A, S)-module andK ⊂M anA-submodule. Then M =KS ⊕L for some fully invariant L ⊂ M. This implies HomA(L, KS) = 0 and soL⊂ TK(M). Hence M/TK(M)∈σ[M/L] =σ[K] showing that M is SSP. 2

14.5 SSP and semisimple modules.

Let M be an A-module.

(1) Assume M has essential socle and for every N M, M ∈ σ[N]. Then M is semisimple.

(2) M is semisimple if and only if every module in σ[M] is SSP.

Proof. (1) By assumption,M ∈σ[Soc(M)] and modules inσ[Soc(M)] are semisimple.

(2) We see from (1) that every finitely cogenerated module in σ[M] is semisimple and hence every simple module inσ[M] isM-injective, i.e.,M is co-semisimple.

Let N be the sum of all non-isomorphic simple modules in σ[M] and consider L=M ⊕N. Then TN(L)⊂Rad(L) = 0 (cf. [40], 23.1). Since L is SSP, this implies L/TN(L)∈σ[N]. Hence L and M are semisimple modules. 2

Weakening the conditions for strongly semiprime modules we define:

14.6 Properly semiprime modules.

Let M be a left A-module and T = EndA(Mc). We call M properly semiprime (PSP) if it satisfies the following equivalent conditions:

(a) For every element a∈M, M/Ta(M)∈σ[Ra];

(b) for any a, b∈M, there exist r1, . . . , rn∈A such that AnA(r1a, r2a, . . . , rna)b ⊂ Ta(M);

(c) for every finitely generated submodule K ⊂M, M/TK(M)∈σ[K];

(d) for any cyclic K ⊂M, every K-injective TK-torsionfree module in σ[M] is M-injective and K-generated;

(e) for any cyclic K ⊂M, KTd ∈σ[K] and TK(M) +TK(M)M; (f ) for any cyclic K ⊂M, Mc =KT ⊕IM(TK(Mc)).

Conditions (d)-(f ) also hold for finitely generated submodules.

Proof. For the equivalence of (a), (c), (d), (e) and (f) see 14.2. (c)⇔(b) is clear.

(b) ⇒ (c) Let a1, . . . , ak be a generating subset of K. By 11.10, TK(M) =

Tk

i=1Tai(M). Hence

M/TK(M)⊂

k

M

i=1

M/Tai(M).

But M/Tai(M)∈σ[Rai]⊂σ[K]. Thus M/TK(M)∈σ[K]. 2

14.7 Properties of PSP-modules.

Let M be an A-module and T =EndA(Mc).

(1) Assume M is a PSP-module and U ⊂Mc an (A, T)-submodule of finite uniform dimension. Then U is a semisimple (A, T)-bimodule.

(2) Assume Mc has finite uniform dimension as (A, T)-bimodule. Then M is an SSP-module if and only ifM is a PSP-module.

Proof. (1) First assume U is a uniform (A, T)-bimodule. Let V ⊂U be an (A, T )-submodule, and N ⊂ M ∩V a finitely generated A-submodule. Then N T is an essential (A, T)-submodule of U. TN(U)∩ TN(U) = 0 implies TN(U) = 0.

From this we deduce N T V U as A-modules. Since N T is M-injective, we conclude N T =V =U and U is a simple (A, T)-bimodule.

Now assumeU has finite uniform dimension as (A, T)-bimodule. Hence there exist uniform submodulesVi ⊂U,i= 1, . . . , n, such thatLni=1ViU as (A, T)-submodule.

Let Ni be finitely generated submodule of the left A-module Vi ∩ M and N =

Ln

i=1Ni. As shown above, all Vi = NiT are simple (A, T)-bimodules. Hence N T =

Pn

i=1NiT =Lni=1ViU as A-submodule. Therefore U =N T =Lni=1Vi is a finitely generated semisimple (A, T)-bimodule.

(2) If M is SSP then obviouslyM is PSP.

If M is PSP and Mc has finite uniform dimension as (A, T)-bimodule, Mc is a semisimple (A, T)-bimodule by (1) and hence M is SSP by 14.4. 2 14.8 Corollary. For a finitely generated left A-module M and T = EndA(Mc), the

following are equivalent:

(a) M is an SSP-module;

(b) Mc is (finitely generated and) semisimple as an (A, T)-bimodule;

(c) M is PSP and Mc has finite uniform dimension as an (A, T)-bimodule.

Proof. Since M is a finitely generated A-module, Mc = M T is a finitely generated (A, T)-bimodule. Hence the assertions follow from 14.4 and 14.7. 2

The next result shows the relation to strongly prime modules.

14.9 Relation to strongly prime modules.

For anA-module M with T =EndA(Mc), the following are equivalent:

(a) M is strongly prime;

(b) M is PSP (or SSP) and Mc is a uniform (A, T)-bimodule.

(c) Mc is a simple (A, T)-bimodule.

In particular, for a uniform A-module M, the conditions strongly prime, SSP, and PSP are equivalent.

Proof. (a)⇒(b) For every submoduleK ⊂M,KT =Mc and the assertion is clear.

(b) ⇒ (a) Assume M is PSP and K ⊂ M is a finitely generated submodule. By the uniformity condition, TK(Mc)∩ TK(Mc) = 0 implies TK(Mc) = 0 and Mc = KT. SoM is strongly prime.

(a)⇔(c) This follows with the same arguments as applied in the proof of 14.4. 2 In general, SSP modules need not be polyform. Modules satisfying both conditions have particularly nice structure properties.

14.10 Polyform SSP modules.

For a polyform A-module M and T =EndA(Mc), the following are equivalent:

(a) M is an SSP-module;

(b) for every N M, M ∈σ[N];

(c) for every submodule K ⊂M, KTc ∈σ[K];

(d) for every submodule K ⊂M, Mc =KT ⊕ TK(Mc).

Proof. The statements follow from 11.11, 14.3 and 14.4. 2 We know that a module M is SSP if and only if Mc is SSP. In general, for a PSP module M,Mc need not be PSP. For polyform modules the PSP property extends at least to the idempotent closure:

14.11 Polyform PSP modules.

Let M be a polyform A-module, T = EndA(Mc), B the Boolean ring of central idempotents of T, and Mf the idempotent closure of M. Then the following are equiv-alent:

(a) M is a PSP module;

(b) for every m∈M, AmT =M ε(m);c

(c) for every finitely generated submodule K ⊂M, KT =M ε(K);c (d) Mf is a PSP module.

Under the given conditions, Tm(Mc) =Mc(id−ε(m)).

Proof. (a) ⇒ (b) By 14.6(f) and 11.11(1), Mc = AmT ⊕ Tm(Mc). Now (b) follows from the definition of the idempotentε(m).

(b)⇒(a) Since M ε(m) isc M-injective,

Pierce stalks of polyform modules will be considered in 18.16.

By 14.4, any injective SSP module is semisimple as a bimodule. For self-injective polyform PSP modules we get a weaker structure theorem:

14.12 Self-injective PSP modules.

Let M be a self-injective polyform A-module and T = EndA(Mc). Denote Λ = A⊗ZZTo and C =EndΛ(M). Then the following conditions are equivalent:

(a) AM is a PSP A-module;

(b) every cyclic Λ-submodule of M is a direct summand;

(c) every finitely generated Λ-submodule of M is a direct summand;

(d) as a Λ-module, M is a selfgenerator;

(e) for any m ∈M and f ∈EndC(M), there exists h ∈Λ with f(m) =hm.

Proof. Notice that Λ-submodules of M are just fully invariant submodules and C can be identified with the centre of T. HenceC is a commutative regular ring.

(a)⇔(b)⇔(c) is clear by 14.11, (c)⇒(d) is obvious.

(d)⇒(e) This follows from the proof of the Density Theorem (e.g., [40, 15.7]).

(e) ⇒ (b) Choose any m ∈ M and ε(m) ∈ C as defined in 11.12. Since mC ' ε(m)C is a direct summand, for any n ∈M, there exists f ∈EndC(M) with f(m) = nε(m). By (e),f(m) =hm for some h∈Λ and hence M ε(m)⊂Λm.

On the other hand, m = mε(m) and Λm ⊂ (ΛM)ε(m) ⊂ M ε(m). So Λm = M ε(m) is a direct summand and the assertion is proved. 2

ModulesM, whose finitely generated submodules are direct summands, are closely related to M being regular in σ[M]. In fact, if M is finitely presented inσ[M], these two notions coincide (7.3). As a special case we derive from above:

14.13 Self-injective finitely presented PSP modules.

Let M be a self-injective polyform A-module and T = EndA(Mc). Denote Λ = A⊗ZZ To and C =EndΛ(M). Assume M is finitely generated as a Λ-module. Then the following conditions are equivalent:

(a) M is a PSP A-module and finitely presented in σ[ΛM];

(b) ΛM is regular and projective in σ[ΛM];

(c) ΛM is a (projective) generator in σ[ΛM];

(d) for any m1, . . . , mn ∈M and f ∈EndC(M), there exists h∈Λ with f(mi) = hmi, for i= 1, . . . , n (density property).

Proof. Since ΛM is finitely generated, MC is a generator in C-Modby 11.12. So MC is a faithfully flat C-module.

(a) ⇔ (b) By 14.12, every finitely generated Λ-submodule of M is a direct sum-mand. Now the assertion follows from 7.3.

(b)⇒(c)⇒(d) are obvious (Density Theorem 5.4).

(d) ⇒ (a) Since MC is a generator in C-Mod, M is (finitely generated and) pro-jective as an EndC(M)-module (5.5).

By the density property, the categories σ[ΛM] and σ[EndC(M)M] coincide (5.4).

SoM is projective (hence finitely presented) in σ[ΛM].

By 14.12, M is a PSP A-module. 2

For modules M with EndA(Mc) commutative we are now able to characterize the density property of Mc as bimodule.

14.14 Density property of the self-injective hull.

Let M be a polyformA-module, assume T =EndA(Mc) to be commutative and put Λ =A⊗ZZT. Then the following are equivalent:

(a) For any a1, . . . , an∈Mc and f ∈EndT(Mc), there exists h∈Λ with hai =f ai for i= 1, . . . , n.

(b) For any m1, . . . , mn, m∈M there exist r1, . . . , rk∈A such that for s1, . . . , sn∈A, the relations

n

X

l=1

slrjml= 0 for j = 1, . . . , k, imply s1m∈ TU m1(M) for U =AnA(m2, . . . , mn).

(c) M is a PSP module and for any m1, . . . , mn ∈ M, there exist r1, . . . , rk ∈ A such that fors1, . . . , sn ∈A the relations

n

X

l=1

slrjml= 0 for j = 1, . . . , k, imply s1m1 ∈ TU m1(M) for U =AnA(m2, . . . , mn).

If Mc is finitely presented in σ[ΛMc], the above are equivalent to:

(d) Mc is a PSP A-module.

Proof. PutU =AnA(m2, . . . , mn).

(a)⇒(b) PutN =Pni=1miT. By 11.12,Mcis a non-singular andN is an injective T-module. Since in a non-singular module the intersection of injective submodules is again injective,K =m1T∩N isT-injective. Hencem1T =K⊕Lfor some submodule L⊂m1T. By 11.12, AnT(m1) = (1−ε(m1))T. This means that the map

m1T →ε(m1)T, m1t7→ε(m1)t,

is an isomorphism. So there exist idempotents u, v ∈T with the properties (∗) uv = 0, u+v =ε(m1), K =m1uT 'uT and L=m1vT 'vT.

Since m1uT =K ⊂N and U N = 0, we have U m1u= 0 and U m1 =U[m1ε(m1)] = U[m1(u+v)] =U m1v.

By 11.13, U m1vAm1v, and by 11.12, ε(U m1v) =ε(Am1v) =ε(m1v).

The isomorphismm1vT 'vT implies ε(m1v) =v and hence we have (∗∗) ε(U m1) =v .

Since m∈M and TU m1(M) = M ∩ TU m1(Mc), by 11.14,

(TU m1(M) :m)A = (TU m1(Mc) :m)A =AnA(mv).

As an injective submodule,m1vT ⊕N is a direct summand inMc. Hence there exists a T-endomorphism ψ of Mc satisfying ψN = 0 and ψm1v =mv (recall m1vT ' vT).

Since m1u∈N, we have ψm1u= 0 and ψm1 =ψ[m(u+v)] =mv.

By assumption, there exists h∈Λ satisfying

hm1 =ψm1 =mv, and hmi =ψmi for i= 2, . . . , n . Now the assertion follows from 11.14.

(b) ⇒ (c) Putting m1 = · · · = mn = 0 we see that M is a PSP module. The second part of the conditions in (c) follows from (b) form =m1.

(c) ⇒ (a) Since Mc = M T it suffices to show that for any m1, . . . , mn ∈ M and f ∈EndT(Mc), there existsh ∈Λ such thatf mi =hmi, for all i= 1, . . . , n.

We prove this by induction on the cardinality|I|of minimal subsetsI ⊂ {1, . . . , n}

satisfying Pi∈ImiT =Pni=1miT.

Consider the case |I|= 1, i.e., I ={1}. By 14.11, M ε(mc 1) = Λm1. Since f m1 =f[m1ε(m1)] = (f m1)ε(m1)∈M ε(mc 1) = Λm1 ,

f m1 =hm1 for someh∈Λ. By assumption, Pni=1miT =m1T, implyingf mi =hmi, for all i= 1, . . . , n.

Now assume |I| = n and consider N = Pni=2miT. As shown above, there exist idempotentsu, v ∈T satisfying (∗) and (∗∗). By 11.14, there exists h1 ∈Λ with

h1m1 =m1v and h1mi = 0, for i= 2, . . . , n . Notice that h1m1u= (h1m1)u= (m1v)u=m1(vu) = 0.

By induction hypothesis, there existsh2 ∈Λ with h2mi =f mi, fori= 2, . . . , n .

Obviously, h2x=f xfor any x∈N. From (∗) we obtain h2m1u=f m1u.

Consider the elementm =f m1v−h2m1v. Clearly m=mv and so m ∈M v. By (∗),c vε(m1) = v. Now it follows from 14.11 that

Λm1v = (Λm1)v =M ε(mc 1)v =M v .c Hence there existsh3 ∈Λ with

h3m1v =m=f m1v−h2m1v . Puttingh=h3h1 +h2, we have

hm1 =h3(h1m1u) +h3(h1m1v) +h2m1u+h2m1v

=h3m1v+f m1u+h2m1v

= (f m1v−h2m1v) +f m1u+h2m1v

=f m1v+f m1u=f m1 ,

hmi =h3(h1mi) +h2mi =f mi, for i= 2, . . . , n .

(a)⇔(d) This is clear by 14.13. 2

We have seen in 14.12 and 14.13 that self-injective polyform PSP modules have nice structural properties. Hence we may ask for which modulesM, the M-injective hullMc is a PSP module.

Now we turn to the question which additional conditons on an SSP or PSP module M imply that M is polyform. It is interesting to observe that this is achieved by commutativity conditions on the endomorphism ring as well as by projectivity of the module. Recall that a ring is said to be (left and right) duo if all its one-sided ideals are two-sided.

14.15 Duo endomorphism rings.

Let M be an A-module, T =EndA(Mc) and assume, for every NM, M ∈σ[N].

(1) Jac(T)∩Z(T) = 0.

(2) Suppose T is a duo ring. Then M is polyform and SSP.

Proof. (2) Assume for f ∈T, N =KefM. Thenc M ∈ σ[N] which is equivalent to N T = Mc. This implies M fc = (N T)f = (N f)T = 0 and hence f = 0. So the Jacobson radical of T is zero, i.e., M is polyform. By 14.10,M is SSP.

(1) A similar argument also implies this assertion. 2 Projectivity makes any PSP module polyform. This applies in particular for the left module structure of the ring itself.

14.16 Projective PSP modules.

Let M be a PSP module which is projective in σ[M]. Then:

(1) For any submodules K, N ⊂M, TN(K)is a complement ofTN(K)in K, hence TN(K) +TN(K)K and [TN(K) +TN(K)]/TN(K)K/TN(K).

(2) M is polyform.

Proof. M is projective in σ[M] if and only if M(Λ) is self-projective, for any set Λ.

From this it is obvious that M/X is projective in σ[M/X], for every fully invariant submoduleX ⊂M.

(1) First we show TN(K) 6= 0, for any finitely generated submodules K, N ⊂M with Hom(K, N) 6= 0. We may assume that there is an epimorphism f : K → N.

Then N ⊂ TK(M) and

TK(M)∩N = 0 =TK(M)∩K.

PutL=TK(M) andM =M/L. There are canonical inclusionsN ⊂M andK ⊂M.

Since M is PSP, M ∈σ[K] and so σ[M] =σ[K].

As outlined above, M is projective in σ[K] and hence is a submodule of K(Λ), for some set Λ. Therefore the compostion of the inclusion N ⊂ M with a suitable map M →K yields a non-zero morphism N →K. This meansTN(K)6= 0.

From 11.10 we know TN(K) ∩ TN(K) = 0. We have to show that TN(K) is maximal with respect to this property.

For x ∈ K \ TN(K), there exists a non-zero g : Ax → Ncand 0 6= (y)g ∈ N, for some y∈Ax. From the above we know

06=TN(Ay)⊂ TN(K)∩Ax.

This shows the maximality of TN(K) which implies the assertions (see [40, 140]).

(2) AssumeM is not polyform. Then there exist a cyclic submoduleK ⊂M and a non-zero morphismf :K →M withKefK. Put N = (K)f and M =M/TN(M).

The map K →f M →M factorizes through ¯f :K/TN(K)→M. Since TN(K)∩Kef TN(K) and TN(K)K/TN(K) by (1), we observeKef¯K/TN(K) and hence N 'Imf¯is an M-singular module.

By assumption, M ∈σ[N] =σ[M]. So, in particular, M is M-singular. However, as noted above, M is projective in σ[M] and hence cannot be M-singular.

Therefore M is polyform. 2

14.17 Projective SSP modules.

LetM be projective inσ[M]andT =EndA(Mc). Then the following are equivalent:

(a) M is an SSP-module;

(b) for every submodule K ⊂M, Mc =KT ⊕ TK(Mc).

(c) M is polyform and for any N M, M ∈σ[N].

Proof. (a)⇒(b) By 14.16, M is polyform and now apply 14.10.

(b) ⇒(a) The decomposition implies in particular that every fully invariant sub-module is a direct summand in Mc as (A, T)-submodule. Hence Mc is a semisimple (A, T)-bimodule and M is SSP by 14.4.

(a)⇔(c) By 14.16, M is polyform and the assertion follows from 14.10. 2 Definition. A ring A is calledleft PSP (SSP), if AA is a PSP (SSP) module.

Notice that the definition also applies to rings without units, considering such rings as modules over rings with units in a canonical way. Rings with units are (left) projective and hence are left non-singular if they are left PSP (or SSP) (by 14.16). We will see that they are also semiprime. Before we want to describe the torsion modules related to semiprime rings.

14.18 Torsion submodules of semiprime rings.

Let A be a semiprime ring and N ⊂A a left ideal. Then:

(1) TN(A) = AnA(N).

(2) A/TN(A)∈σ[N]if and only if there exists a finite subsetX ⊂N withAnA(X) = AnA(N).

Proof. (1) The relation TN(A) ⊂ AnA(N) always holds. Clearly N AnA(N) is a nilpotent left ideal and hence is zero.

Consider f ∈HomA(AnA(N), IA(N)) and put K = (N)f−1. Then (Kf)2 ⊂N(Kf) = (N K)f ⊂(N AnA(N))f = 0.

Since A is semiprime, Kf = 0 and so Im f ∩N = 0, implying f = 0. Hence TN(A)⊃AnA(N).

(2) Assume A/TN(A)∈σ[N]. By 14.1, there exists a finite subset X ⊂N with AnA(X)1⊂ TN(A) = AnA(N)⊂AnA(X).

Now assume AnA(X) =AnA(N) for some finite X ⊂N. Then AnA(X) =TN(A) by (1), and for any b ∈A,AnA(X)b⊂ TN(A). Now apply 14.1. 2

Applying our module theoretic results to AA, we obtain characterizations of 14.19 Left SSP rings.

For the ring A put Q:=Qmax(A). The following are equivalent:

(a) A is left SSP;

(b) for every essential left ideal N ⊂A, A∈σ[N];

(c) every essential left ideal N ⊂A contains a finite subset X with AnA(X) = 0;

(d) for every left ideal I ⊂A, Q=IQ⊕ TI(Q);

(e) A is semiprime and every left ideal I ⊂ A contains a finite subset X ⊂ I with AnA(X) = AnA(I);

(f ) Q is a semisimple (A, Q)-module.

If A satisfies these conditions, then Q is left self-injective, von Neumann regular, and a finite product of simple rings.

Proof. (a)⇒(b) is shown in 14.4.

(b) ⇒(a) Assume for every essential left ideal N ⊂ A, A∈ σ[N]. Any such N is a faithful A-module.

First we show that A is semiprime. For this consider an ideal I ⊂A with I2 = 0.

Let J be the right annihilator of I, and L ⊂ A any non-zero left ideal. Obviously, I ⊂J. Assume L∩J = 0. Then IL6= 0. However, IL⊂L∩J = 0, a contradiction.

This implies that J is an essential left ideal in A. By our assumption, J is a faithful left module and IJ = 0 means I = 0.

In view of 14.4 and 11.11, it remains to show thatA is left non-singular

If the left singular ideal S(A)⊂ A is non-zero, S(A)⊕AnA(S(A)) is an essential left ideal inA. Hence there area1, . . . , ak withAnA(a1, . . . , ak) = 0. From this we see that there is a monomorphism S(A)→ S(A)(b1, . . . , br), withb1, . . . , br ∈ S(A).

The kernel of this map isS(A)∩AnA(b1)∩. . .∩AnA(br). Since all the AnA(bi) are essential left ideals in A, this intersection could not be zero, a contradiction. Hence A is left non-singular.

(b)⇔(c) This is obvious by 14.1.

(a) ⇒(d) By 14.16, A is semiprime and left non-singular. Therefore Q = Ab and EndA(Q) = Q. Now the assertion follows from 14.10.

(d) ⇒ (a) We show that A is left non-singular. Assume for a ∈ A, AnA(a) is an essential left ideal in A. Since Q = RaQ⊕ Ta(Q), we have RaQ = eQ, for some idempotente ∈Q. Write e=Pni=1riaqi, withri ∈A, qi ∈Q.

Obviously, L = Tni=1(AnA(a) : ri)A is an essential left ideal in A and Le = 0.

ThereforeL∩Qe= 0 and also L∩(Qe∩A) = 0. However,L⊂A is an essential left ideal and Qe∩A 6= 0, a contradiction. Hence A is left non-singular and so Q = Ab and EndA(Q) =Q. Now apply 14.10.

(a)⇔(e)⇔(f) follow from 14.17, 14.3 and 14.4. 2 14.20 Left PSP rings.

For a ring A, the following conditions are equivalent:

(a) A is a left PSP-ring;

(b) Ais semiprime (and left non-singular) and for every finitely generated left ideal N ⊂A, AnA(N) =AnA(X), for some finite subset X ⊂N.

Proof. (a) ⇒ (b) By 14.16, A is left non-singular. To show that A is semiprime, consider a cyclic left ideal N ∈ A with N2 = 0. By assumption, A/TN(A) ∈ σ[N].

Since N K = 0 for any K ∈σ[N], in particular N(A/TN(A)) = 0 and N A ⊂ TN(A).

This impliesN A ⊂ TN(A)∩ TN(A) = 0 and N = 0. Now refer to 14.18.

(b)⇒(a) also follows from 14.18. 2

Now we consider a condition on modules which is closely related to regularity properties.

14.21 Pseudo regular modules.

Let M be a left A-module, S = EndA(M), T = EndA(Mc) and D = EndT(Mc).

Then the following conditions are equivalent:

(a) For every m∈M, there exists h∈D, such that hM ⊂AmS and hm =m;

(b) for any (A, S)-submodule N ⊂ M and m1, . . . , mk ∈ N, there exists h ∈ D, such that hM ⊂N and hmi =mi, for all i= 1, . . . , k.

A module satisfying these conditions is called pseudo regular.

Proof. (a) ⇒ (b) Let N ⊂ M be an (A, S)-submodule of M and m1, . . . , mk ∈ N.

Fork = 1 there is nothing to show.

Assume the assertion holds for anyk−1 elements inN. Choose f ∈D such that f mk=mk and f M ⊂AmkS⊂N.

Clearly mi −f mi ∈ N, for all i = 1, . . . , k −1, and by hypothesis, there exists g ∈D with gM ⊂N and

g(mi −f mi) = mi−f mi, for i= 1, . . . , k−1.

Puth= 1−(1−g)(1−f). Then hm =gm−gf m+f m∈N, for allm ∈M, which means hM ⊂N, and for i= 1, . . . , k,

hmi =mi−(1−g)(1−f)mi =mi−(1−g)(mi−f mi) =mi.

(b)⇒(a) is obvious. 2

The next result shows what pseudo regularity means for rings.

14.22 Left fully idempotent rings.

A ring A is left fully idempotent if and only if the left A-module A is pseudo regular.

Proof. RecallEndA(A) =A. AssumeAis left fully idempotent. LetN be an ideal of A andm ∈N. Since (Am)2 =Am, there exist h∈AmAsuch thathm =m. Clearly hA⊂N and h∈A⊂BiendA(A). Henceb AA is pseudo regular.

Now assumeAAto be pseudo regular andm∈A. Then there existsh∈BiendA(A)b such that hm=m andhA ⊂AmA. Hence r=h1∈AmA andrm= (h1)m =hm = m. So m =rm∈(Am)2, (Am)2 =Amand the ring A is left fully idempotent. 2 It follows from the above observation that a commutative ring A is von Neumann regular if and only ifAA is pseudo regular.

Of particular interest for our investigations is the following relationship:

14.23 Pseudo regular and PSP modules.

Let M be a left A-module, S = EndA(M) and T = EndA(Mc). The following are equivalent:

(a) For every m∈M, there is a central idempotent e∈T, with AmS =M e;

(b) for every m∈M, M =AmS ⊕ Tm(M) and Mc=AmT ⊕ Tm(Mc);

(c) for every finitely generated submodule N ⊂ M, there is a central idempotent e∈T, with N S =M e;

(d) for every finitely generated submodule N ⊂M, M =N S⊕ TN(M) and Mc=N T ⊕ TN(Mc);

(e) M is a pseudo regular P SP-module over A.

Proof. Denote D=EndT(Mc).

(a)⇒(b) Form ∈M choose a central idempotente∈T, with AmS =M e. Since Mc =M T we have

Mc = M T e⊕M T(1−e) =M eT ⊕M T(1−e)

= AmST ⊕M T(1−e) = AmT ⊕M T(1−e).

Hence AmT is M-injective and M T e = AmT = Tm(Mc) = IM(Tm(Mc)) (see 11.10).

Since Mc is M-injective,

HomA(M T(1−e), IM(Tm(Mc))) = {t∈T |M T(1−e)t⊂IM(Tm(Mc))}

= {t∈T |M T(1−e)t⊂M T e}= 0.

SoM T(1−e)⊂ Tm(Mc). ButM T e=AmT,Mc=M T e⊕M T(1−e) and M T e∩ Tm(Mc) = Tm(Mc)∩ Tm(Mc) = 0.

HenceTm(Mc) =M T(1−e) andMc =AmT⊕Tm(Mc). ClearlyAmS =M e=M∩M T e and

Tm(M) =M ∩ Tm(Mc) =M ∩M T(1−e) =M(1−e).

SoM =AmT ⊕ Tm(M).

(b) ⇒ (a) Let e be the projection of Mc onto AmT along Tm(Mc). Since both submodules are fully invariant, e is a central idempotent of T. From AmS ⊂ AmT and Tm(M)⊂ Tm(Mc), we concludeM e=AmS.

(c)⇔(d) can be shown with the above proof.

(c)⇒(a) is obvious.

(a)⇒(e) SinceM/Tm(M) = AmS ∈σ[Am], M isP SP.

Let π be the projection of Mc onto AmT along Tm(Mc). Obviously, π ∈D. Since

Consider the following properties of an A-module M:

(i) A/AnA(M) is cogenerated by every essential submodule of M; (ii) for every N M,M ∈σ[N];

(iii) for every N M,M ⊂N(Λ), for some set Λ;

(iv) for every N M,A/AnA(M)⊂Nr, for some r∈IN; (v) M is cogenerated by every essential submodule of M; (vi) M is polyform;

(vii) for every submodule K ⊂M, M/TK(M)∈σ[K];

(viii) for every cyclic submoduleK ⊂M, M/TK(M)∈σ[K].

The conditions (i)-(vi) stated for every submodule were considered in 13.9. (vii) is used to define strongly semiprime modules, and (viii) defines properly semiprime

modules. For any module M, (iii) ⇒ (v) ⇒ (i), (iii) ⇒ (ii) ⇒ (i), (iv) ⇒ (i) and (vii)⇒(ii).

ForM projective inσ[M], (ii)⇔(iii) and (vii) is equivalent toM satisfying both (ii) and (vi) (cf. 14.16, 14.17).

For A commutative and M = A, any of the properties (i), (v), (vi) and (viii) characterize A as a semiprime ring (hence Qmax(A) is regular). Properties (ii), (iii), (iv) and (vii) imply that Qmax(A) is a finite product of fields.

For M =AA, the conditions (ii), (iii), (iv) and (vii) are equivalent and describe SSP rings (see 14.19). Left ideals N ⊂ A, which contain a finite subset X with AnA(X) = 0, are also called insulated. Hence A is left SSP if and only if every essential left ideal is insulated. So A is left strongly semiprime ring in the sense of Handelman [149, Theorem 1]. Such rings are also investigated in Kutami-Oshiro [182]

and generalize left strongly prime rings (see 13.9).

14.25 Exercises.

Recall that A is an associative algebra with unit.

(1) The definitions of SSP and PSP modulesM refer to the surrounding category.

Nevertheless submodules of such modules are of the same type.

LetM be anA-module and L⊂M a submodule. Prove: If M is SSP (resp. PSP) (in σ[M]), then L is also SSP (resp. PSP) (in σ[L]).

(2) An A-module M is called weakly semisimple if it is polyform, every finitely generated submodule has finite uniform dimension, and for every non-zero submodule N ⊂M, there exists f ∈HomA(M, N) with f|N 6= 0 (weakly compressible).

Let M be such a module. Prove ([285, 1.1]):

(i) For any m∈M and N M, there exists a monomorphismAm→N ∩Am.

(ii) M is strongly semiprime.

(3) Show that for a ring A, the following are equivalent ([285, 5.1, 5.2]):

(a) A has a faithful weakly semisimple left ideal (of finite uniform dimension);

(b) A is semiprime, left non-singular, with a faithful left ideal which contains a (finite) direct sum of uniform left ideals.

(4) A ring is said to beleft fully idempotentif every left ideal is idempotent. Show that for a ringA, the following are equivalent ([77, 4.2]):

(a) A is biregular;

(b) A is a left fully idempotent left PSP ring;

(c) A is a left PSP ring whose prime factor rings are left fully idempotent;

(d) A is a left PSP ring with all prime ideals maximal.

(5) Let A be a reduced ring. Prove:

(i) For any subset U ⊂A, the left annihilator AnA(U) of U coincides with the right annihilator of U in A ([2, p. 286]).

(ii) A is left (and right) PSP.

(6) Let Abe a reduced ring. Prove that the following are equivalent (see [77, 4.2], [81, Theorem 8]):

(a) A is biregular;

(b) A is left fully idempotent;

(c) all prime ideals in A are maximal.

(7) Let A be a reduced ring. Prove that the following are equivalent ([226]):

(a) Qmax(A) is reduced;

(b) closed left ideals in A are two-sided ideals;

(c) for non-zero elements x, y ∈A, Ax∩Ay= 0 implies xy= 0 =yx.

(8) Let A be a semiprime left non-singular ring. Prove that the following are equivalent ([182, Proposition 2.3]):

(a) Qmax(A) is a direct sum of simple rings;

(b) the set of central idempotents ofQmax(A)is finite;

(c) A contains no infinite direct sum of ideals.

(9) Show that for a ring A, the following are equivalent ([182, Theorem 2.5]):

(a) A is left SSP;

(b) Qmax(A) is a direct sum of simple rings and for every idempotent e∈Qmax(A), AeQmax(A) =Qmax(A)eQmax(A).

(10) Show that for a ring A, the following are equivalent ([182, Theorem 3.4]):

(a) A/S2(A) is left SSP;

(b) every non-singular self-injective leftA-module is injective;

(c) any finite direct sum of non-singular self-injective leftA-modules is self-injective.

(11) Let A be a semiprime ring. Prove ([149, Lemma 7, Corollary 23]):

(i) IfAhas no infinite direct sums of ideals, thenAsatisfies acc and dcc on annihilators of ideals.

(ii) If A has dcc on left annihilators then A is left SSP.

References. Beidar-Wisbauer [75, 76, 77], Birkenmeier-Kim-Park [81], Faith [12], Handelman [149], Kutami-Oshiro [182], Wisbauer [273], Zelmanowitz [285].

Im Dokument Modules and Algebras ... (Seite 116-135)