Introduction
All rings are assumed to be associative and to have an identity 1 ≠ 0 1 \ne0 1 = 0 ; all modules are unital right modules. An ideal means a two-sided ideal. For a right A A A -module M M M and an ideal B B B of A A A , we denote by M B MB M B the submodule consisting of all finite sums of elements m b mb mb , where m ∈ M m \in M m ∈ M and b ∈ B b \in B b ∈ B . Following Tuganbaev [6 , 7 ] , the module M M M is called a multiplication module (an m-module) if for every submodule N N N of M M M there exists an ideal B B B of A A A such that N = M B N = MB N = M B .
We call M M M an iso-multiplication module (an im-module) if for every submodule N ≤ M N \le M N ≤ M there exists an ideal B B B of A A A such that N ≅ A M B N \cong_A MB N ≅ A M B . The isomorphism is A A A -linear and need not be the restriction of an automorphism of M M M . Every multiplication module is an iso-multiplication module. For noncommutative rings the converse is not true. In particular, a right ideal can be isomorphic to a two-sided ideal without being a two-sided ideal.
The decomposition of representations of the one-arrow quiver, the Morita ideal correspondence, and the standard theorems on Ore localization are not new. The manuscript’s main contributions are the selection of iso-multiplication objects from the standard triangular-module classification and the multiplicity obstruction obtained by applying the iso-multiplication hypothesis after Goldie localization. For comparison with multiplication modules and chain conditions we use [6 , 10 , 11 ] ; no new proof of the pre-existing classifications is claimed.
Scope and provenance of the principal results.
Table 1.
The other statements are applications, obstructions, or supporting examples and should be read accordingly.
The principal examples are the regular right modules of triangular rings. For a field k k k , the right regular module of T 2 ( k ) T_2(k) T 2 ( k ) is iso-multiplication but is not a multiplication module. We give the complete classification of iso-multiplication modules over this ring, including infinitely generated modules. The triangular construction extends to A ( V ) A(V) A ( V ) for arbitrary nonzero k k k -vector spaces V V V , including infinite-dimensional cases. Related examples are obtained from skew polynomial rings, direct products, factor rings, and Morita equivalences.
We prove that an iso-multiplication module over a semiprime right Goldie ring has multiplicity-free classical localization. This is a necessary condition, not a characterization of iso-multiplication. In particular, if the regular module A A A_A A A is iso-multiplication, the classical right ring of quotients of A A A is a finite product of division rings. This condition is necessary but not sufficient. We also obtain bounds on composition length when the ring has finitely many ideals, a cyclicity result for local rings, and results on chain conditions up to isomorphism.
We give some notation. The set of submodules of M M M is denoted by L ( M ) \mathcal{L}(M) L ( M ) . For N ≤ M N \leq M N ≤ M we set
( N : M ) = { a ∈ A : M a ⊆ N } = r A ( M / N ) . (N : M) = \{a \in A : Ma \subseteq N\} = r_A(M/N). ( N : M ) = { a ∈ A : M a ⊆ N } = r A ( M / N ) .
Thus ( N : M ) (N : M) ( N : M ) is a two-sided ideal. Here r A ( Y ) r_A(Y) r A ( Y ) and Ann A ( Y ) \operatorname{Ann}_A(Y) Ann A ( Y ) both denote the two-sided annihilator { a ∈ A : Y a = 0 } \{a \in A : Ya = 0\} { a ∈ A : Y a = 0 } of a right A A A -module Y Y Y ; in particular, ( N : M ) = Ann A ( M / N ) (N : M) = \operatorname{Ann}_A(M/N) ( N : M ) = Ann A ( M / N ) . We write l A ( X ) l_A(X) l A ( X ) for the composition length of X X X (when finite) and u.dim A ( X ) \operatorname{u.dim}_A(X) u.dim A ( X ) for its right uniform or Goldie dimension (the supremum of the numbers of nonzero summands in finite internal direct sums of submodules). We use [ X ] [X] [ X ] only for the isomorphism class of X X X ; [ X : S ] [X : S] [ X : S ] denotes the Jordan–Hölder multiplicity of a simple module S S S in a finite-length module X X X . If S S S is a module and λ \lambda λ a cardinal, S ( λ ) S^{(\lambda)} S ( λ ) denotes a direct sum of λ \lambda λ copies of S S S . For a semiprime right Goldie ring A A A , we write C C C for the set of regular elements, Q c l r ( A ) Q^r_{\mathrm{cl}}(A) Q cl r ( A ) for its classical right ring of quotients, and M Q = M ⊗ A Q M_Q = M \otimes_A Q M Q = M ⊗ A Q when Q = Q c l r ( A ) Q = Q^r_{\mathrm{cl}}(A) Q = Q cl r ( A ) . For a submodule N ≤ M N \leq M N ≤ M we write
N Q : = im ( N ⊗ A Q ⟶ M ⊗ A Q ) ⊆ M Q . N_Q := \operatorname{im}(N \otimes_A Q \longrightarrow M \otimes_A Q) \subseteq M_Q. N Q := im ( N ⊗ A Q ⟶ M ⊗ A Q ) ⊆ M Q .
Exactness of classical Ore localization identifies this image canonically with N ⊗ A Q N \otimes_A Q N ⊗ A Q ; whenever two such localized submodules are called equal, equality is meant inside M Q M_Q M Q , whereas ≅ Q \cong_Q ≅ Q always denotes an abstract Q Q Q -module isomorphism. The terms isomultiplication and isoartinian are used for the corresponding chain conditions up to isomorphism [11 , 12 ] .
Examples and elementary properties
For a right A A A -module M M M and a two-sided ideal B B B of A A A , we use M B MB M B in the sense defined in the Introduction. In particular, for the regular module A A A_A A A , we have A B = B AB=B A B = B . The distinction between right ideals and two-sided ideals is essential in the following examples.
Proposition 2.1. For every right module M M M the following hold:
(i) M M M is a multiplication module if and only if N = M ( N : M ) N=M(N:M) N = M ( N : M ) for every N ∈ L ( M ) N\in\mathcal{L}(M) N ∈ L ( M ) .
(ii) Every multiplication module is an iso-multiplication module; the iso-multiplication property is invariant under isomorphisms.
(iii) The zero module and every simple module are iso-multiplication modules (indeed, multiplication modules).
Proof. If N = M B N=MB N = M B , then B ⊆ ( N : M ) B\subseteq(N:M) B ⊆ ( N : M ) , so N = M B ⊆ M ( N : M ) ⊆ N N=MB\subseteq M(N:M)\subseteq N N = M B ⊆ M ( N : M ) ⊆ N . Conversely, use B = ( N : M ) B=(N:M) B = ( N : M ) . The other assertions follow from the definitions; an isomorphism f : M → M ′ f:M\to M' f : M → M ′ maps M B MB M B onto M ′ B M'B M ′ B while a simple module has only the submodules 0 = M 0 0=M0 0 = M 0 and M = M A M=MA M = M A . □ \square □
Remark 2.2. For an iso-multiplication module, N ≅ M B N\cong MB N ≅ M B does not imply B ⊆ ( N : M ) B\subseteq(N:M) B ⊆ ( N : M ) , and it does not imply N ≅ M ( N : M ) N\cong M(N:M) N ≅ M ( N : M ) . The distinction is demonstrated explicitly in Example 2.4.
Proposition 2.3. The module M A M_A M A is an iso-multiplication module precisely when the map
{ B : B ◃ A } ⟶ { [ N ] : N ∈ L ( M ) } , B ↦ [ M B ] \{B:B\triangleleft A\}\longrightarrow\{[N]:N\in\mathcal{L}(M)\},\qquad B\mapsto[MB] { B : B ◃ A } ⟶ {[ N ] : N ∈ L ( M )} , B ↦ [ M B ] is surjective. In particular, for every N ≤ M N\leq M N ≤ M there is an ideal B B B with Ann A ( N ) = Ann A ( M B ) \operatorname{Ann}_A(N)=\operatorname{Ann}_A(MB) Ann A ( N ) = Ann A ( M B ) ; equality of annihilators alone is not sufficient.
Proof. The first assertion is the definition expressed in terms of isomorphism classes. An A A A -linear isomorphism preserves annihilators. To see why annihilator equality alone does not imply the defining isomorphism, let A = k A=k A = k and M = k 2 M=k^2 M = k 2 . The submodule N = k ⊕ 0 N=k\oplus0 N = k ⊕ 0 and the ideal multiple M A = M MA=M M A = M have the same zero annihilator, but N ̸ ≅ A M N\not\cong_A M N ≅ A M . □ \square □
Examples over triangular and principal right ideal rings
Example 2.4. Let k k k be any field, A = T 2 ( k ) = { ( a b 0 c ) : a , b , c ∈ k } A=T_2(k)=\left\{\begin{pmatrix}a&b\\0&c\end{pmatrix}:a,b,c\in k\right\} A = T 2 ( k ) = { ( a 0 b c ) : a , b , c ∈ k } and M = A A M=A_A M = A A . Denote the matrix units by e 11 , e 12 , e 22 e_{11},e_{12},e_{22} e 11 , e 12 , e 22 . The complete list of two-sided ideals is
0 , J = k e 12 , K = k e 12 ⊕ k e 22 , L = k e 11 ⊕ k e 12 , A . 0,\qquad J=ke_{12},\qquad K=ke_{12}\oplus ke_{22},\qquad L=ke_{11}\oplus ke_{12},\qquad A. 0 , J = k e 12 , K = k e 12 ⊕ k e 22 , L = k e 11 ⊕ k e 12 , A .
The right A A A -module K K K is a direct sum of two copies of S 2 = e 22 A = k e 22 S_2=e_{22}A=ke_{22} S 2 = e 22 A = k e 22 ; both e 12 e_{12} e 12 and e 22 e_{22} e 22 are multiplied by ( a b 0 c ) \begin{pmatrix}a&b\\0&c\end{pmatrix} ( a 0 b c ) through the scalar c c c . Consequently, any right ideal U ⊆ K U\subseteq K U ⊆ K is a k k k -linear subspace and satisfies
U ≅ { 0 , dim k U = 0 , J , dim k U = 1 , K , dim k U = 2. U\cong\begin{cases}
0,&\dim_k U=0,\\
J,&\dim_k U=1,\\
K,&\dim_k U=2.
\end{cases} U ≅ ⎩ ⎨ ⎧ 0 , J , K , dim k U = 0 , dim k U = 1 , dim k U = 2.
If a right ideal U U U is not contained in K K K , choose u = a e 11 + b e 12 + c e 22 ∈ U u=ae_{11}+be_{12}+ce_{22}\in U u = a e 11 + b e 12 + c e 22 ∈ U with a ≠ 0 a\ne0 a = 0 . Then u e 11 = a e 11 ue_{11}=ae_{11} u e 11 = a e 11 and e 11 e 12 = e 12 e_{11}e_{12}=e_{12} e 11 e 12 = e 12 show L = e 11 A ⊆ U L=e_{11}A\subseteq U L = e 11 A ⊆ U . Since A / L A/L A / L is simple, U = L U=L U = L or A A A . Every right ideal is therefore isomorphic to one of 0 0 0 , J J J , K K K , L L L , A A A . Since M B = A B = B MB = AB = B M B = A B = B for each ideal B B B , A A A_A A A is iso-multiplication.
It is not a multiplication module: N = e 22 A = k e 22 N = e_{22}A = ke_{22} N = e 22 A = k e 22 is a right ideal but not two-sided, because e 12 e 22 = e 12 ∉ N e_{12}e_{22} = e_{12} \notin N e 12 e 22 = e 12 ∈ / N . Nevertheless,
e 22 A ⟶ J = e 12 A , e 22 a ↦ e 12 a e_{22}A \longrightarrow J = e_{12}A,\qquad e_{22}a \mapsto e_{12}a e 22 A ⟶ J = e 12 A , e 22 a ↦ e 12 a
is a well-defined A A A -isomorphism. Moreover ( N : M ) = 0 (N : M) = 0 ( N : M ) = 0 , as there is no nonzero two-sided ideal inside k e 22 ke_{22} k e 22 . Thus M ( N : M ) = 0 ≠ N M(N : M) = 0 \ne N M ( N : M ) = 0 = N .
Table 1 (PDF p. 4). Isomorphism types of right ideals of T 2 ( k ) T_2(k) T 2 ( k )
Proposition 2.5. If A A A is a (possibly noncommutative) domain such that every right ideal is principal, then A A A_A A A is iso-multiplication. It is a multiplication module if and only if every right ideal is two-sided.
Proof. For every nonzero right ideal a A aA a A , multiplication A A → a A A_A \to aA A A → a A , r ↦ a r r \mapsto ar r ↦ a r , is an isomorphism: surjectivity is clear and injectivity follows from the absence of zero divisors. Thus each right ideal is isomorphic to A = A A A = A_A A = A A or 0 = A 0 0 = A0 0 = A 0 . For the second assertion use A B = B AB = B A B = B for a two-sided ideal B B B .
Example 2.6. Let K = k ( t ) K = k(t) K = k ( t ) with char k = 0 \operatorname{char} k = 0 char k = 0 , let σ ( t ) = t + 1 \sigma(t) = t + 1 σ ( t ) = t + 1 on K K K , and consider the Ore extension A = K [ x ; σ ] A = K[x;\sigma] A = K [ x ; σ ] subject to x a = σ ( a ) x xa = \sigma(a)x x a = σ ( a ) x . Its degree function is additive on nonzero products; in particular A A A is a domain. Here every nonzero right ideal is principal. Indeed, let I I I be such an ideal and choose 0 ≠ g ∈ I 0 \ne g \in I 0 = g ∈ I of minimal degree d d d . Given f ∈ I f \in I f ∈ I with deg f ≥ d \deg f \ge d deg f ≥ d , write m = deg f − d m = \deg f - d m = deg f − d . If a a a and b b b are the leading coefficients of g g g and f f f , respectively, choose c = σ − d ( a − 1 b ) ∈ K c = \sigma^{-d}(a^{-1}b) \in K c = σ − d ( a − 1 b ) ∈ K . The leading term of g ( c x m ) g(cx^m) g ( c x m ) is a x d c x m = a σ d ( c ) x d + m = b x deg f ax^dcx^m = a\sigma^d(c)x^{d+m} = bx^{\deg f} a x d c x m = a σ d ( c ) x d + m = b x d e g f . Subtracting g ( c x m ) g(cx^m) g ( c x m ) from f f f therefore lowers the degree. Repeating the step gives f = g q + r f = gq + r f = g q + r with deg r < d \deg r < d deg r < d (or r = 0 r = 0 r = 0 ). Since r ∈ I r \in I r ∈ I , minimality of d d d forces r = 0 r = 0 r = 0 . Hence I = g A I = gA I = g A , so A A A_A A A is iso-multiplication by Proposition 2.5.
The right ideal ( x − 1 ) A (x - 1)A ( x − 1 ) A is not two-sided. Indeed, if it were a left ideal, then t ( x − 1 ) = ( x − 1 ) f t(x - 1) = (x - 1)f t ( x − 1 ) = ( x − 1 ) f for some f ∈ A f \in A f ∈ A . Since A A A is a domain with additive degree on nonzero products, f f f must have degree zero, say f = c ∈ K f = c \in K f = c ∈ K . Expanding with x c = σ ( c ) x xc = \sigma(c)x x c = σ ( c ) x gives
t x − t = ( x − 1 ) c = σ ( c ) x − c . tx - t = (x - 1)c = \sigma(c)x - c. t x − t = ( x − 1 ) c = σ ( c ) x − c . The constant coefficient on the left is − t -t − t and on the right is − c -c − c , so c = t c = t c = t . The coefficient of x x x on the left is t t t and on the right is σ ( c ) \sigma(c) σ ( c ) , so σ ( c ) = t \sigma(c) = t σ ( c ) = t . Substitution gives t + 1 = σ ( t ) = t t + 1 = \sigma(t) = t t + 1 = σ ( t ) = t , a contradiction. Hence A A A_A A A is not a classical multiplication module. This supplies an example over a non-Artinian, noncommutative domain.
Example 2.7. For every r ≥ 0 r \ge0 r ≥ 0 , let A r = T 2 ( k ) × k r A_r = T_2(k) \times k^r A r = T 2 ( k ) × k r . Then the right regular module ( A r ) A r (A_r)_{A_r} ( A r ) A r is iso-multiplication but not a multiplication module. Indeed, the T 2 ( k ) T_2(k) T 2 ( k ) factor is iso-multiplication by Example 2.4; each regular k k k -factor is simple, and the finite-product theorem proved below (Theorem 4.23 ) applies. The non-multiplication property is inherited from the factor T 2 ( k ) T_2(k) T 2 ( k ) by considering the right ideal e 22 T 2 ( k ) × 0 e_{22}T_2(k) \times0 e 22 T 2 ( k ) × 0 .
Example 2.8. In Example 2.4, the right ideal K = k e 12 ⊕ k e 22 K = ke_{12} \oplus ke_{22} K = k e 12 ⊕ k e 22 is a submodule of the iso-multiplication module A A A_A A A , but K A K_A K A is not iso-multiplication. Indeed K ≅ S 2 ⊕ S 2 K \cong S_2 \oplus S_2 K ≅ S 2 ⊕ S 2 . The only ideal multiples of K K K are 0 0 0 and K K K : the action of A A A on K K K factors through the bottom-right diagonal entry, and each two-sided ideal either has zero bottom-right entry or contains e 22 e_{22} e 22 . The proper nonzero submodule J ≅ S 2 J \cong S_2 J ≅ S 2 is isomorphic to neither 0 0 0 nor K K K .
Example 2.9. For any division ring D D D , the simple right module D D D_D D D is a multiplication module, but D D ⊕ D D D_D \oplus D_D D D ⊕ D D is not iso-multiplication: the only ideal multiples are 0 0 0 and D 2 D^2 D 2 , whereas a copy of D D D is a submodule. In particular the iso-multiplication property is not closed under finite direct sums.
Remark 2.10. The iso-multiplication condition is not automatic, even in the commutative setting: for the Prüfer group Z ( p ∞ ) \mathbb{Z}(p^\infty) Z ( p ∞ ) , every nonzero ideal multiple equals the whole group, whereas it contains a submodule of order p p p . We do not pursue commutative classifications here.
Triangular rings of arbitrary dimension
The preceding construction extends to triangular rings of arbitrary dimension. We describe their right ideals explicitly.
Theorem 2.11. Let k k k be a field, 0 ≠ V 0 \ne V 0 = V an arbitrary k k k -vector space, and put
A ( V ) = { ( a v 0 c ) : a , c ∈ k , v ∈ V } , A(V) = \left\{ \begin{pmatrix} a & v \\ 0 & c \end{pmatrix} : a,c \in k,\ v \in V \right\}, A ( V ) = { ( a 0 v c ) : a , c ∈ k , v ∈ V } , ( a v 0 c ) ( a ′ v ′ 0 c ′ ) = ( a a ′ a v ′ + v c ′ 0 c c ′ ) . \begin{pmatrix} a & v \\ 0 & c \end{pmatrix}
\begin{pmatrix} a' & v' \\ 0 & c' \end{pmatrix}
=
\begin{pmatrix} aa' & av' + vc' \\ 0 & cc' \end{pmatrix}. ( a 0 v c ) ( a ′ 0 v ′ c ′ ) = ( a a ′ 0 a v ′ + v c ′ c c ′ ) . Here V V V is regarded as the natural ( k , k ) (k,k) ( k , k ) -bimodule, with the usual scalar multiplication on both sides; this convention is essential for the classification of two-sided ideals. Since K = J ⊕ k e 22 K = J \oplus ke_{22} K = J ⊕ k e 22 , one has dim k K = dim k V + 1 \dim_k K = \dim_k V + 1 dim k K = dim k V + 1 as a cardinal identity. If V V V is infinite-dimensional, cardinal addition absorbs this extra 1 1 1 ; if V V V is finite-dimensional, dim k K \dim_k K dim k K exceeds dim k V \dim_k V dim k V by one. Then the regular right module A ( V ) A ( V ) A(V)_{A(V)} A ( V ) A ( V ) is iso-multiplication but not a classical multiplication module. More precisely, writing
J = { ( 0 v 0 0 ) : v ∈ V } , K = J ⊕ k e 22 , L = e 11 A ( V ) , J = \left\{ \begin{pmatrix} 0 & v \\ 0 & 0 \end{pmatrix} : v \in V \right\}, \qquad K = J \oplus ke_{22}, \qquad L = e_{11}A(V), J = { ( 0 0 v 0 ) : v ∈ V } , K = J ⊕ k e 22 , L = e 11 A ( V ) , every right ideal is either a k k k -subspace of K K K , or L L L , or A ( V ) A(V) A ( V ) ; this is a classification of right ideals of the regular module, not of all A ( V ) A(V) A ( V ) -modules; each subspace of K K K is isomorphic to a two-sided ideal of A ( V ) A(V) A ( V ) .
Proof. The action of A ( V ) A(V) A ( V ) on K K K factors through the homomorphism A ( V ) → k A(V) \to k A ( V ) → k taking a matrix to its bottom-right entry. Hence every k k k -subspace U ⊆ K U \subseteq K U ⊆ K is a right ideal, and U ≅ S 2 ( dim k U ) U \cong S_2^{(\dim_k U)} U ≅ S 2 ( d i m k U ) , where S 2 = k e 22 S_2 = ke_{22} S 2 = k e 22 and the parenthesized exponent permits infinite cardinals. Since V ≠ 0 V \ne0 V = 0 , for each cardinal λ ≤ dim k K \lambda\le\dim_k K λ ≤ dim k K there exists a two-sided ideal isomorphic to S 2 ( λ ) S_2^{(\lambda)} S 2 ( λ ) ; if λ ≤ dim k V \lambda\le\dim_k V λ ≤ dim k V , choose a λ \lambda λ -dimensional subspace W ⊆ V W \subseteq V W ⊆ V and use W ⊆ J W \subseteq J W ⊆ J which is a two-sided ideal; otherwise necessarily dim k V \dim_k V dim k V is finite and λ = dim k V + 1 \lambda= \dim_k V + 1 λ = dim k V + 1 , in which case use the two-sided ideal K K K . (For infinite V V V , dim k K = dim k V \dim_k K = \dim_k V dim k K = dim k V .)
If a right ideal U U U is not contained in K K K , it contains a matrix u u u with nonzero top-left entry a a a . Then u e 11 = a e 11 ue_{11}=ae_{11} u e 11 = a e 11 implies e 11 ∈ U e_{11}\in U e 11 ∈ U and hence L = e 11 A ( V ) ⊆ U L=e_{11}A(V)\subseteq U L = e 11 A ( V ) ⊆ U . Since A ( V ) / L ≅ S 2 A(V)/L \cong S_2 A ( V ) / L ≅ S 2 is simple, U = L U=L U = L or A ( V ) A(V) A ( V ) . Both L L L and A ( V ) A(V) A ( V ) are themselves two-sided ideals. Thus every right ideal is isomorphic to a two-sided ideal, which proves that A ( V ) A ( V ) A(V)_{A(V)} A ( V ) A ( V ) is iso-multiplication.
Finally e 22 A ( V ) = k e 22 e_{22}A(V)=ke_{22} e 22 A ( V ) = k e 22 is a right ideal but not a two-sided ideal: for 0 ≠ v ∈ V 0\ne v\in V 0 = v ∈ V the upper-right matrix j v j_v j v satisfies j v e 22 = j v ∉ k e 22 j_ve_{22}=j_v\notin ke_{22} j v e 22 = j v ∈ / k e 22 . The regular module is therefore not multiplication.
Corollary 2.12. If V V V is infinite dimensional, then A ( V ) A ( V ) A(V)_{A(V)} A ( V ) A ( V ) is iso-multiplication and isoartinian, but it is not isonoetherian. In particular, iso-multiplication plus isoartinianity does not imply isonoetherianity, even for a noncommutative right regular module.
Proof. By Theorem 2.11 every right ideal is a subspace of K K K , or L L L , or A ( V ) A(V) A ( V ) . In a descending chain, terms outside K K K are confined to the two possibilities A ( V ) A(V) A ( V ) and L L L , so after at most two such terms (up to repetition) the chain either remains equal to L L L or enters K K K permanently. Once inside K K K , it is a descending chain of vector spaces. Its sequence of dimension cardinals is nonincreasing and must eventually be constant, since there is no infinite strictly descending sequence of cardinals. The corresponding right ideals are then mutually isomorphic as right modules. Thus the regular module is isoartinian.
For the ascending chain, choose linearly independent vectors v 1 , v 2 , … ∈ V v_1,v_2,\ldots\in V v 1 , v 2 , … ∈ V and define, for n ≥ 1 n\ge1 n ≥ 1 ,
U n = k v 1 ⊕ ⋯ ⊕ k v n ⊆ J . U_n=kv_1\oplus\cdots\oplus kv_n\subseteq J. U n = k v 1 ⊕ ⋯ ⊕ k v n ⊆ J .
Then U 1 ⊊ U 2 ⊊ ⋯ U_1\subsetneq U_2\subsetneq\cdots U 1 ⊊ U 2 ⊊ ⋯ is a strictly ascending chain of right ideals (each U n U_n U n is a subspace of the two-sided ideal J J J ). The action on J J J is through the lower-right scalar, so U n ≅ S 2 ⊕ n U_n\cong S_2^{\oplus n} U n ≅ S 2 ⊕ n for every n n n . In particular, U n ≇ U m U_n\not\cong U_m U n ≅ U m when n ≠ m n\ne m n = m , as their composition lengths are different. This chain never stabilizes up to isomorphism, and hence A ( V ) A ( V ) A(V)_{A(V)} A ( V ) A ( V ) is not isonoetherian.
Remark 2.13. For finite-dimensional V V V , the ring A ( V ) A(V) A ( V ) is right Artinian and consequently both isoartinian and isonoetherian. For infinite-dimensional V V V it has a square-zero Jacobson radical J J J , so it lies outside the semiprime right Goldie setting of Section 5. No comparison with a semiprime classification is implied.
Modules over the triangular algebra T 2 ( k ) T_2(k) T 2 ( k )
The right modules over T 2 ( k ) T_2(k) T 2 ( k ) are representations of the quiver with one arrow. We recall their standard decomposition and determine which of these modules are iso-multiplication, without a finiteness assumption.
Let A = T 2 ( k ) A=T_2(k) A = T 2 ( k ) as in Example 2.4. Put
P = e 11 A , S 1 = e 11 A / e 12 A , S 2 = e 22 A . P=e_{11}A,\qquad S_1=e_{11}A/e_{12}A,\qquad S_2=e_{22}A. P = e 11 A , S 1 = e 11 A / e 12 A , S 2 = e 22 A .
Here P P P is a two-dimensional indecomposable projective, with rad P ≅ S 2 \operatorname{rad}P\cong S_2 rad P ≅ S 2 and top S 1 S_1 S 1 . The two simple right modules are S 1 S_1 S 1 and S 2 S_2 S 2 .
Lemma 3.1. Every right A A A -module, with no finiteness condition, decomposes uniquely up to isomorphism as
M ≅ P ( a ) ⊕ S 1 ( b ) ⊕ S 2 ( c ) , a , b , c cardinals . M\cong P^{(a)}\oplus S_1^{(b)}\oplus S_2^{(c)},\qquad a,b,c\text{ cardinals}. M ≅ P ( a ) ⊕ S 1 ( b ) ⊕ S 2 ( c ) , a , b , c cardinals .
Proof. Put V 1 = M e 11 V_1 = M e_{11} V 1 = M e 11 and V 2 = M e 22 V_2 = M e_{22} V 2 = M e 22 . Right multiplication by e 12 e_{12} e 12 gives a k k k -linear map f : V 1 → V 2 f : V_1 \to V_2 f : V 1 → V 2 , v ↦ v e 12 v \mapsto v e_{12} v ↦ v e 12 . Conversely an arbitrary right A A A -module is determined by this map, because a matrix ( u w 0 z ) \begin{pmatrix} u & w \\ 0 & z \end{pmatrix} ( u 0 w z ) acts on v 1 + v 2 v_1 + v_2 v 1 + v 2 as u v 1 + w v 1 e 12 + z v 2 u v_1 + w v_1 e_{12} + z v_2 u v 1 + w v 1 e 12 + z v 2 . Choose complements V 1 = ker f ⊕ U V_1 = \ker f \oplus U V 1 = ker f ⊕ U and V 2 = f ( U ) ⊕ W V_2 = f(U) \oplus W V 2 = f ( U ) ⊕ W ; they exist even for infinite-dimensional vector spaces. The restriction f ∣ U : U → f ( U ) f|_U : U \to f(U) f ∣ U : U → f ( U ) is an isomorphism. Choose bases of U U U , ker f \ker f ker f and W W W . The cardinal a = dim k U a = \dim_k U a = dim k U gives P ( a ) P^{(a)} P ( a ) , while b = dim k ker f b = \dim_k \ker f b = dim k ker f and c = dim k W c = \dim_k W c = dim k W give S 1 ( b ) S_1^{(b)} S 1 ( b ) and S 2 ( c ) S_2^{(c)} S 2 ( c ) . These cardinals are uniquely determined by the linear map f f f .
Lemma 3.2. For M = P ( a ) ⊕ S 1 ( b ) ⊕ S 2 ( c ) M = P^{(a)} \oplus S_1^{(b)} \oplus S_2^{(c)} M = P ( a ) ⊕ S 1 ( b ) ⊕ S 2 ( c ) , the ideal multiples of M M M , up to A A A -module isomorphism, are
Proof. For the representation f : V 1 → V 2 f : V_1 \to V_2 f : V 1 → V 2 of Lemma 3.1, identify V 1 = M e 11 V_1 = M e_{11} V 1 = M e 11 and V 2 = M e 22 V_2 = M e_{22} V 2 = M e 22 as additive subspaces of M M M . Then M J = im f MJ = \operatorname{im} f M J = im f , M K = V 2 MK = V_2 M K = V 2 , and M L = V 1 ⊕ im f ML = V_1 \oplus\operatorname{im} f M L = V 1 ⊕ im f as additive subspaces of M M M ; the latter is the subrepresentation consisting of all of V 1 V_1 V 1 and only the image of f f f in V 2 V_2 V 2 . These are isomorphic respectively to S 2 ( a ) S_2^{(a)} S 2 ( a ) , S 2 ( a + c ) S_2^{(a+c)} S 2 ( a + c ) , and P ( a ) ⊕ S 1 ( b ) P^{(a)} \oplus S_1^{(b)} P ( a ) ⊕ S 1 ( b ) . Here brackets in the table denote isomorphism classes, not equality of submodules inside M M M ; for example M J MJ M J is the image of f f f inside M e 22 M e_{22} M e 22 . The cases 0 0 0 , A A A are immediate.
Theorem 3.3. Let A = T 2 ( k ) A = T_2(k) A = T 2 ( k ) and let M M M be an arbitrary right A A A -module, possibly infinitely generated. Then M M M is iso-multiplication if and only if
M ≅ 0 , S 1 , S 2 , P , S 1 ⊕ S 2 , or P ⊕ S 2 . M \cong0,\quad S_1,\quad S_2,\quad P,\quad S_1 \oplus S_2,\quad\text{or}\quad P \oplus S_2. M ≅ 0 , S 1 , S 2 , P , S 1 ⊕ S 2 , or P ⊕ S 2 . Among these six isomorphism types the first five are multiplication modules, and P ⊕ S 2 ≅ A A A P \oplus S_2 \cong{}_A A_A P ⊕ S 2 ≅ A A A is iso-multiplication but not a multiplication module. Thus A A A {}_A A_A A A A is the only isomorphism type over T 2 ( k ) T_2(k) T 2 ( k ) separating the two conditions. In particular, every iso-multiplication module over T 2 ( k ) T_2(k) T 2 ( k ) is finite dimensional over k k k .
Proof. By Lemma 3.1, write M = P ( a ) ⊕ S 1 ( b ) ⊕ S 2 ( c ) M = P^{(a)} \oplus S_1^{(b)} \oplus S_2^{(c)} M = P ( a ) ⊕ S 1 ( b ) ⊕ S 2 ( c ) , where a , b , c a,b,c a , b , c are arbitrary cardinals. Assume first that M M M is iso-multiplication.
If b > 0 b > 0 b > 0 , the summand S 1 S_1 S 1 is a submodule. By Lemma 3.2, the only ideal multiples with any S 1 S_1 S 1 composition factor are M L = P ( a ) ⊕ S 1 ( b ) ML = P^{(a)} \oplus S_1^{(b)} M L = P ( a ) ⊕ S 1 ( b ) and M M M itself. Thus, in order that an ideal multiple be isomorphic to S 1 S_1 S 1 , necessarily a = 0 a = 0 a = 0 and b = 1 b = 1 b = 1 . Therefore M = S 1 ⊕ S 2 ( c ) M = S_1 \oplus S_2^{(c)} M = S 1 ⊕ S 2 ( c ) . If c > 0 c > 0 c > 0 , a copy of S 2 S_2 S 2 must also be isomorphic to an ideal multiple; now M K = S 2 ( c ) MK = S_2^{(c)} M K = S 2 ( c ) is the only nonzero ideal multiple of pure S 2 S_2 S 2 -type, so c = 1 c = 1 c = 1 . Hence this case gives S 1 S_1 S 1 or S 1 ⊕ S 2 S_1 \oplus S_2 S 1 ⊕ S 2 .
If b = 0 b = 0 b = 0 and a = 0 a = 0 a = 0 then M = S 2 ( c ) M = S_2^{(c)} M = S 2 ( c ) . Since all nonzero ideal multiples equal M M M , the presence of a simple submodule forces c = 1 c = 1 c = 1 unless M = 0 M = 0 M = 0 .
Finally suppose b = 0 b = 0 b = 0 and a > 0 a > 0 a > 0 . The module M M M contains S 2 ⊆ rad P S_2 \subseteq\operatorname{rad} P S 2 ⊆ rad P . The only nonzero pure S 2 S_2 S 2 ideal multiples in Lemma 3.2 are S 2 ( a ) S_2^{(a)} S 2 ( a ) and S 2 ( a + c ) S_2^{(a+c)} S 2 ( a + c ) ; hence a = 1 a = 1 a = 1 (a simple module cannot be isomorphic to an infinite direct sum of simples). If c ≥ 2 c \ge2 c ≥ 2 , the semisimple submodule S 2 ⊕ S 2 ⊆ S 2 ( 1 + c ) = M K S_2 \oplus S_2 \subseteq S_2^{(1+c)} = MK S 2 ⊕ S 2 ⊆ S 2 ( 1 + c ) = M K has no isomorphic ideal multiple, since the only nonzero pure S 2 S_2 S 2 multiples have multiplicities 1 1 1 and 1 + c ≥ 3 1+c \ge3 1 + c ≥ 3 , including the possibility that c c c is infinite. Thus c ≤ 1 c \le1 c ≤ 1 . We obtain P P P and P ⊕ S 2 P \oplus S_2 P ⊕ S 2 .
The following table records the necessity argument for arbitrary cardinals a , b , c a,b,c a , b , c . The entries in its third column refer to the list of ideal multiples in Lemma 3.2 .
Table 2.
Here the exclusion of c ≥ 2 c\geq2 c ≥ 2 includes all infinite cardinals: 1 + c ≥ 3 1+c\geq3 1 + c ≥ 3 as cardinals, so neither S 2 ( 1 ) S_2^{(1)} S 2 ( 1 ) nor S 2 ( 1 + c ) S_2^{(1+c)} S 2 ( 1 + c ) is isomorphic to S 2 ( 2 ) S_2^{(2)} S 2 ( 2 ) . Hence the six surviving types all have finite k k k -dimension, proving the final assertion of the theorem without any finite-generation assumption.
Conversely, 0 0 0 , S 1 S_1 S 1 , S 2 S_2 S 2 are simple or zero and hence m. The only submodules of P P P are 0 0 0 , rad P = P J ≅ S 2 \operatorname{rad}P=PJ\cong S_2 rad P = P J ≅ S 2 , and P P P , so P P P is a multiplication module. Because S 1 , S 2 S_1,S_2 S 1 , S 2 are nonisomorphic simple modules, the submodules of S 1 ⊕ S 2 S_1\oplus S_2 S 1 ⊕ S 2 are 0 0 0 , S 1 S_1 S 1 , S 2 S_2 S 2 , M M M , and they are respectively M 0 M_0 M 0 , M L M_L M L , M K M_K M K , M A M_A M A ; hence this module is a multiplication module. Lastly P ⊕ S 2 ≅ e 11 A ⊕ e 22 A = A A P\oplus S_2\cong e_{11}A\oplus e_{22}A=A_A P ⊕ S 2 ≅ e 11 A ⊕ e 22 A = A A . For completeness, its right ideals fall into the following isomorphism types: 0 0 0 , S 2 S_2 S 2 , S 2 ⊕ S 2 S_2\oplus S_2 S 2 ⊕ S 2 , P P P , and A A A . Indeed a right ideal contained in K = k e 12 ⊕ k e 22 K=ke_{12}\oplus ke_{22} K = k e 12 ⊕ k e 22 is a k k k -subspace of K K K and is semisimple of S 2 S_2 S 2 -type, whereas a right ideal not contained in K K K contains e 11 e_{11} e 11 and hence is L = e 11 A L=e_{11}A L = e 11 A or A A A . These five types are represented by the two-sided ideals 0 0 0 , J J J , K K K , L L L , A A A , respectively. Therefore A A A_A A A is iso-multiplication; it is not multiplication because e 22 A e_{22}A e 22 A is not two-sided, as shown in Example 2.4.
Table 2 (PDF p. 8). Classification summary over T 2 ( k ) T_2(k) T 2 ( k )
Remark 3.4. The normal form is the standard decomposition of representations of the one-arrow quiver of type A 2 A_2 A 2 . Theorem 3.3 determines which of these representations are iso-multiplication. Among them, only the regular module fails to be a multiplication module.
Corollary 3.5. Let A = T 2 ( k ) A=T_2(k) A = T 2 ( k ) and let M M M be an iso-multiplication right A A A -module. Then M M M is a classical multiplication module if and only if M ≇ A A M\not\cong A_A M ≅ A A .
Proof. By Theorem 3.3 , exactly six isomorphism classes have the iso-multiplication property. The first five classes are classical multiplication modules, while P ⊕ S 2 ≅ A A P\oplus S_2\cong A_A P ⊕ S 2 ≅ A A is not a classical multiplication module, by Example 2.4.
Remark 3.6. The decomposition lemma and classification include infinite-dimensional representations. The classification shows a posteriori that no infinite-dimensional module over T 2 ( k ) T_2(k) T 2 ( k ) satisfies the iso-multiplication property. This does not extend automatically to triangular algebras with higher-dimensional off-diagonal bimodules.
Composition length and local rings
The number of two-sided ideals of a ring gives restrictions on the composition lengths of its iso-multiplication modules.
Proposition 4.1. Let M A M_A M A be an iso-multiplication module of finite composition length r r r . Then
{ 0 , 1 , … , r } ⊆ { l A ( M B ) : B ⊲ A } . \{0,1,\ldots,r\}\subseteq\{l_A(MB):B\lhd A\}. { 0 , 1 , … , r } ⊆ { l A ( M B ) : B ⊲ A } . In particular there are at least r + 1 r+1 r + 1 distinct ideal multiples M B MB M B . This conclusion does not imply that the corresponding ideals can be chosen in a chain under inclusion.
Proof. Choose a composition series 0 = M 0 ⊊ M 1 ⊊ ⋯ ⊊ M r = M 0=M_0\subsetneq M_1\subsetneq\cdots\subsetneq M_r=M 0 = M 0 ⊊ M 1 ⊊ ⋯ ⊊ M r = M . For every j j j the defining condition supplies a two-sided ideal B j B_j B j with M j ≅ M B j M_j\cong MB_j M j ≅ M B j . Hence l A ( M B j ) = j l_A(MB_j)=j l A ( M B j ) = j . Since distinct lengths preclude isomorphism, the multiples M B j MB_j M B j are distinct. Nothing in this argument compares B i B_i B i with B j B_j B j by inclusion. □ \square □
Theorem 4.2. Let A A A be any unital ring and M A M_A M A an iso-multiplication module. Assume every finitely generated submodule of M M M has finite composition length. If the collection { [ M B ] : B ⊲ A } \{[MB]:B\lhd A\} {[ M B ] : B ⊲ A } consists of exactly u < ∞ u<\infty u < ∞ isomorphism types, then M M M has finite length and
l A ( M ) ≤ u − 1. l_A(M)\leq u-1. l A ( M ) ≤ u − 1. In particular, M M M is both Noetherian and Artinian, and thus both isonoetherian and isoartinian. Neither A A A nor M M M is assumed Noetherian or Artinian in advance.
Proof. Every finitely generated submodule F F F has a composition series of length r = l A ( F ) r=l_A(F) r = l A ( F ) . All its r + 1 r+1 r + 1 terms are submodules of M M M of pairwise different composition lengths. By the iso-multiplication property these yield r + 1 r+1 r + 1 distinct isomorphism types among the ideal multiples M B MB M B ; therefore r ≤ u − 1 r\leq u-1 r ≤ u − 1 . Choose a finitely generated F ≤ M F\leq M F ≤ M of maximal possible length, which exists since the possible lengths lie in the finite set { 0 , … , u − 1 } \{0,\ldots,u-1\} { 0 , … , u − 1 } . If F ≠ M F\neq M F = M , select m ∈ M ∖ F m\in M\setminus F m ∈ M ∖ F . The finitely generated module F + m A F+mA F + m A has finite length, and the nonzero quotient ( F + m A ) / F (F+mA)/F ( F + m A ) / F forces l A ( F + m A ) > l A ( F ) l_A(F+mA)>l_A(F) l A ( F + m A ) > l A ( F ) , a contradiction. Hence M = F M=F M = F and l A ( M ) ≤ u − 1 l_A(M)\leq u-1 l A ( M ) ≤ u − 1 . Finite-length modules satisfy both classical chain conditions. The bound is sharp, as witnessed by A = k [ x ] / ( x u − 1 ) A=k[x]/(x^{u-1}) A = k [ x ] / ( x u − 1 ) and M = A A M=A_A M = A A , which has u u u distinct ideal multiples and length u − 1 u-1 u − 1 (for $u>2). □ \square □
Remark 4.3. The mere finiteness of { [ M B ] } \{[MB]\} {[ M B ]} does not imply finite length: for a principal right ideal domain which is not a field, its regular module has only the two types [ 0 ] [0] [ 0 ] and [ A ] [A] [ A ] among ideal multiples but may have infinite composition length. The theorem uses finite length of finitely generated submodules, not just a finite type spectrum.
Theorem 4.4. Let A A A be a right Artinian ring having exactly t < ∞ t<\infty t < ∞ two-sided ideals, counted including 0 0 0 and A A A . If M A M_A M A is an iso-multiplication module, then M M M has finite composition length, and
l A ( M ) ≤ t − 1. l_A(M)\leq t-1. l A ( M ) ≤ t − 1. In particular M M M is finitely generated. The bound is best possible for each integer $t\geq 2.
Proof. A right Artinian ring is right Noetherian by the Hopkins–Levitzki theorem. Thus each finitely generated right A A A -module is both Artinian and Noetherian, hence has finite composition length. Since there are only t t t two-sided ideals, there are at most t t t isomorphism classes [MB]. Theorem 4.2 therefore yields l A ( M ) ≤ t − 1 l_A(M) \le t - 1 l A ( M ) ≤ t − 1 , as claimed.
For sharpness, fix t ≥ 2 t \ge2 t ≥ 2 and let A = k [ x ] / ( x t − 1 ) A = k[x]/(x^{t-1}) A = k [ x ] / ( x t − 1 ) over a field k k k . Its t t t distinct ideals are ( 1 ) , ( x ) , … , ( x t − 1 ) = 0 (1),(x),\ldots,(x^{t-1}) = 0 ( 1 ) , ( x ) , … , ( x t − 1 ) = 0 . The regular module A A A_A A A is a classical multiplication module and has composition length t − 1 t-1 t − 1 . Thus equality is attained for every t ≥ 2 t \ge2 t ≥ 2 .
Corollary 4.5. If A A A is right Artinian with finitely many two-sided ideals, every iso-multiplication module M A M_A M A is a finite-length module. In particular, it is Noetherian and Artinian and hence satisfies both chain conditions up to isomorphism in the sense of Facchini-Nazemian.
Proof. Theorem 4.4 yields finite composition length. The remaining assertions are standard consequences of a composition series; no cancellation or mutual-embedding hypothesis is needed.
Theorem 4.6. Let A A A be a right Artinian ring with precisely t < ∞ t < \infty t < ∞ two-sided ideals, and let M A M_A M A be an iso-multiplication module satisfying l A ( M ) = t − 1 l_A(M) = t - 1 l A ( M ) = t − 1 . Then the lengths of the ideal multiples M B M_B M B are exactly 0 , 1 , … , t − 1 0,1,\ldots,t-1 0 , 1 , … , t − 1 , and exactly one two-sided ideal B B B gives each length. Consequently, if N , L ≤ M N,L \le M N , L ≤ M then
l A ( N ) = l A ( L ) ⟹ N ≅ A L . l_A(N) = l_A(L) \Longrightarrow N \cong_A L. l A ( N ) = l A ( L ) ⟹ N ≅ A L . In particular, composition length induces a bijection from the set of isomorphism classes of submodules of M M M onto { 0 , … , t − 1 } \{0,\ldots,t-1\} { 0 , … , t − 1 } . The conclusion does not assert that L ( M ) \mathcal{L}(M) L ( M ) is linearly ordered.
Proof. Put r = l A ( M ) = t − 1 r = l_A(M) = t - 1 r = l A ( M ) = t − 1 , and fix a composition series 0 = M 0 ⊊ M 1 ⊊ ⋯ ⊊ M r = M 0 = M_0 \subsetneq M_1 \subsetneq\cdots\subsetneq M_r = M 0 = M 0 ⊊ M 1 ⊊ ⋯ ⊊ M r = M . For each j j j , the iso-multiplication property gives a two-sided ideal B j B_j B j satisfying M j ≅ A M B j M_j \cong_A M B_j M j ≅ A M B j , hence l A ( M B j ) = j l_A(M B_j) = j l A ( M B j ) = j . Since these lengths are different, B 0 , … , B r B_0,\ldots,B_r B 0 , … , B r are r + 1 = t r+1=t r + 1 = t distinct two-sided ideals. They exhaust all the ideals of A A A , and their corresponding ideal multiples have pairwise distinct lengths. Every submodule N ≤ M N \le M N ≤ M has some length j ∈ { 0 , … , r } j \in\{0,\ldots,r\} j ∈ { 0 , … , r } and is isomorphic to an ideal multiple M B M B M B . Necessarily B = B j B = B_j B = B j , so N ≅ A M B j N \cong_A M B_j N ≅ A M B j . The same reasoning applied to L L L proves the assertion for equal lengths. Every length j j j occurs along the chosen composition series, proving the bijection. No step implies comparability of different submodules.
Remark 4.7. For A = k [ x ] / ( x r ) A = k[x]/(x^r) A = k [ x ] / ( x r ) and M = A A M = A_A M = A A , the r + 1 r+1 r + 1 ideals are ( 1 ) , ( x ) , … , ( x r ) = 0 (1),(x),\ldots,(x^r) = 0 ( 1 ) , ( x ) , … , ( x r ) = 0 and l A ( M ) = r l_A(M) = r l A ( M ) = r , so equality occurs for every r ≥ 1 r \ge1 r ≥ 1 . This familiar uniserial example verifies sharpness, but the conclusion of Theorem 4.6 is uniqueness of isomorphism type by length, not uniseriality in general. More abstractly, the same proof works when M M M has finite length r r r and there are precisely r + 1 r+1 r + 1 distinct isomorphism types among the ideal multiples M B M B M B : the r + 1 r+1 r + 1 distinct lengths supplied by a composition series then exhaust those types.
Proposition 4.8. Let M A M_A M A have finite length r r r . Assume that any two submodules of M M M having the same composition length are isomorphic and that for every j ∈ { 0 , … , r } j \in\{0,\ldots,r\} j ∈ { 0 , … , r } some two-sided ideal B j B_j B j satisfies l A ( M B j ) = j l_A(M B_j) = j l A ( M B j ) = j . Then M M M is an iso-multiplication module. Conversely, an iso-multiplication module of finite length always satisfies the second assumption, whether or not the first assumption holds.
Proof. For N ≤ M N \le M N ≤ M set j = l A ( N ) j = l_A(N) j = l A ( N ) . By assumption, l A ( M B j ) = j l_A(M B_j) = j l A ( M B j ) = j and hence N ≅ A M B j N \cong_A M B_j N ≅ A M B j . For the converse, take a composition series of M M M and use the iso-multiplication property on its terms. This establishes the length realization, but does not automatically yield uniqueness of submodule types of equal length.
Example 4.9. Let k k k be a field and A = k × k A = k \times k A = k × k . For M = k ⊕ k M = k \oplus k M = k ⊕ k with the natural coordinatewise action the two simple submodule types are different, but M M M is a classical multiplication module (all four submodules are ideal multiples). Thus the first hypothesis of Proposition 4.8 is sufficient, not necessary: iso-multiplication does not force equal-length submodules to be isomorphic in general.
Example 4.10. The algebra T 2 ( k ) T_2(k) T 2 ( k ) has five two-sided ideals, so Theorem 4.4 gives ℓ A ( M ) ≤ 4 \ell_A(M) \le4 ℓ A ( M ) ≤ 4 for every iso-multiplication module M M M . The stronger complete classification in Theorem 3.3 shows that the largest possible length is actually three, attained by A A ≅ P ⊕ S 2 A_A \cong P \oplus S_2 A A ≅ P ⊕ S 2 . The counting theorem is thus a universal constraint rather than a substitute for a detailed classification.
Composition-factor vectors
Composition length does not distinguish the simple composition factors of submodules. The following proposition gives a necessary condition in terms of Jordan–Hölder multiplicities.
Proposition 4.12. Suppose M A M_A M A has finite composition length, and let S 1 , … , S q S_1,\ldots,S_q S 1 , … , S q represent the isomorphism classes of simple composition factors of M M M . For X ≤ M X \le M X ≤ M put
c ( X ) = ( [ X : S 1 ] , … , [ X : S q ] ) ∈ N q , c(X) = ([X:S_1],\ldots,[X:S_q]) \in\mathbb{N}^q, c ( X ) = ([ X : S 1 ] , … , [ X : S q ]) ∈ N q , where [ X : S i ] [X:S_i] [ X : S i ] is the Jordan–Hölder multiplicity of S i S_i S i . If M M M is an iso-multiplication module, then
{ c ( N ) : N ≤ M } = { c ( M B ) : B ◃ A } . \{c(N):N \le M\} = \{c(MB):B \triangleleft A\}. { c ( N ) : N ≤ M } = { c ( M B ) : B ◃ A } . Thus a single composition-factor vector of a submodule which is not realized by any ideal multiple disproves iso-multiplication.
Proof. For each N ≤ M N \le M N ≤ M , the iso-multiplication property yields N ≅ M B N \cong MB N ≅ M B for some two-sided ideal B B B . Isomorphic finite-length modules have the same Jordan–Hölder multiplicities, so c ( N ) = c ( M B ) c(N) = c(MB) c ( N ) = c ( M B ) . Conversely, M B ≤ M MB \le M M B ≤ M for every two-sided ideal B B B , proving the reverse inclusion. The equality follows.
Example 4.13. Let A = T 2 ( k ) A = T_2(k) A = T 2 ( k ) , with radical J = k e 12 J = ke_{12} J = k e 12 , and two-sided ideals L = k e 11 ⊕ k e 12 L = ke_{11} \oplus ke_{12} L = k e 11 ⊕ k e 12 and K = k e 12 ⊕ k e 22 K = ke_{12} \oplus ke_{22} K = k e 12 ⊕ k e 22 . Write S 1 = e 11 A / e 12 A S_1 = e_{11}A/e_{12}A S 1 = e 11 A / e 12 A and S 2 = e 22 A S_2 = e_{22}A S 2 = e 22 A for the two nonisomorphic simple right modules, and set
M = S 1 ⊕ S 1 ⊕ S 2 . M = S_1 \oplus S_1 \oplus S_2. M = S 1 ⊕ S 1 ⊕ S 2 . The five two-sided ideals of A A A are 0 0 0 , J J J , L L L , K K K , A A A . Their actions on M M M are
M 0 = M J = 0 , M L = S 1 ⊕ 2 , M K = S 2 , M A = M . M0 = MJ = 0,\qquad ML = S_1^{\oplus2},\qquad MK = S_2,\qquad MA = M. M 0 = M J = 0 , M L = S 1 ⊕ 2 , M K = S 2 , M A = M . Thus the lengths of ideal multiples realize every value in { 0 , 1 , 2 , 3 } \{0,1,2,3\} { 0 , 1 , 2 , 3 } , which is the entire possible scalar length spectrum of the submodules of M M M . Nevertheless S 1 ≤ M S_1 \le M S 1 ≤ M has factor vector ( 1 , 0 ) (1,0) ( 1 , 0 ) , while the ideal multiples have only the vectors
( 0 , 0 ) , ( 2 , 0 ) , ( 0 , 1 ) , ( 2 , 1 ) . (0,0),\qquad(2,0),\qquad(0,1),\qquad(2,1). ( 0 , 0 ) , ( 2 , 0 ) , ( 0 , 1 ) , ( 2 , 1 ) . Therefore S 1 S_1 S 1 is not isomorphic to any M B MB M B , and M M M is not iso-multiplication. In particular, merely realizing all intermediate composition lengths by ideal multiples is not sufficient for the defining property.
Remark 4.14. When a simple submodule S ≤ M S \leq M S ≤ M occurs, the iso-multiplication property requires an ideal B B B for which M B ≅ A S MB \cong_A S M B ≅ A S , not merely l A ( M B ) = 1 l_A(MB)=1 l A ( M B ) = 1 . In a finite-length module the vector c ( S ) c(S) c ( S ) has precisely one nonzero coordinate, equal to 1 1 1 . Therefore failure of any ideal multiple to realize that coordinate vector excludes the iso-multiplication property. This is a practical necessary test, not a sufficient criterion: distinct nonsplit extensions may have the same Jordan–Hölder factor vector.
Remark 4.15. The scalar-length and factor-vector tests are obstructions, not general recognition theorems. Even matching complete composition-factor vectors is weaker than matching module isomorphism classes in nonsemisimple categories. Proposition 4.8 does give a sufficient criterion, but only because it adds the independent hypothesis that submodules with the same length are isomorphic.
General consequences
Theorem 4.16. If A A A is a simple ring and M A ≠ 0 M_A \neq0 M A = 0 , then M M M is iso-multiplication if and only if every nonzero submodule N N N of M M M is isomorphic to M M M .
Proof. The only two-sided ideals are 0 , A 0,A 0 , A , so the only ideal multiples are 0 , M 0,M 0 , M . □ \square □
Corollary 4.17. For a simple Artinian ring A A A , the iso-multiplication modules are precisely the zero module and the simple right modules.
Proof. A right module over A A A is semisimple; if it has more than one simple summand, a simple submodule is not isomorphic to the whole module, contradicting Theorem 4.16. □ \square □
Theorem 4.18. Let A A A be any associative ring with identity and let M A M_A M A be a semisimple right module, not necessarily finitely generated. Then
M is im ⟺ M is m . M\text{ is im}\ \Longleftrightarrow\ M\text{ is m}. M is im ⟺ M is m . In particular, every simple isomorphism type occurs with multiplicity at most one in an iso-multiplication semisimple module.
Proof. Write M = ⨁ λ ∈ Λ S λ ( κ λ ) M=\bigoplus_{\lambda\in\Lambda}S_\lambda^{(\kappa_\lambda)} M = ⨁ λ ∈ Λ S λ ( κ λ ) , with the S λ S_\lambda S λ pairwise nonisomorphic simple modules and κ λ \kappa_\lambda κ λ cardinals. For every two-sided ideal B B B , simplicity implies S λ B = 0 S_\lambda B=0 S λ B = 0 or S λ B = S λ S_\lambda B=S_\lambda S λ B = S λ ; consequently M B = ⨁ S λ B = S λ S λ ( κ λ ) MB=\bigoplus_{S_\lambda B=S_\lambda}S_\lambda^{(\kappa_\lambda)} M B = ⨁ S λ B = S λ S λ ( κ λ ) . Assume M M M is iso-multiplication and fix μ \mu μ with κ μ > 0 \kappa_\mu>0 κ μ > 0 . A single copy S μ S_\mu S μ is a submodule and so is isomorphic to some M B MB M B . Comparing simple summand multiplicities in the displayed decomposition forces κ μ = 1 \kappa_\mu=1 κ μ = 1 . Thus M M M is multiplicity-free. For any submodule N ≤ M N\leq M N ≤ M , semisimplicity implies that N N N is a direct sum of simple modules. Since M M M contains at most one copy of each simple isomorphism type, every simple submodule of M M M is contained in a unique simple direct summand S λ S_\lambda S λ . Indeed, the coordinate projections of a simple submodule into the summands S λ S_\lambda S λ are homomorphisms, and a nonzero homomorphism between simple modules is an isomorphism. Since the S λ S_\lambda S λ are pairwise nonisomorphic, at most one coordinate projection can be nonzero. Consequently, every submodule N N N is the direct sum of a uniquely determined subset of the simple direct summands of M M M . If N ≅ M B N\cong MB N ≅ M B , comparison of the multiplicities of the simple isomorphism types shows that this subset is exactly the set of summands occurring in M B MB M B . Hence N = M B N=MB N = M B , and M M M is an m m m -module. The converse is immediate. □ \square □
Corollary 4.19. If A A A is semisimple Artinian, then M A M_A M A is iso-multiplication if and only if it is a multiplicity-free module, precisely when each simple A A A -module occurs with multiplicity at most one in M M M . In particular, this applies without assuming M M M finitely generated.
Corollary 4.20. If a semisimple module M M M contains two isomorphic nonzero simple direct summands, then M M M is not iso-multiplication. This applies, in particular, to S ⊕ S S \oplus S S ⊕ S for every simple S S S over every ring.
Proposition 4.21. For every right A A A -module M M M , set
I ( M ) = { [ M B ] : B ◃ A } , S ( M ) = { [ N ] : N ∈ L ( M ) } , \mathcal{I}(M)=\{[MB] : B \triangleleft A\}, \qquad\mathcal{S}(M)=\{[N] : N \in\mathcal{L}(M)\}, I ( M ) = {[ M B ] : B ◃ A } , S ( M ) = {[ N ] : N ∈ L ( M )} , where brackets denote right-module isomorphism types. Then M M M is iso-multiplication if and only if I ( M ) = S ( M ) \mathcal{I}(M)=\mathcal{S}(M) I ( M ) = S ( M ) . Consequently, if A A A has only t t t two-sided ideals and M M M is iso-multiplication, at most t t t isomorphism types occur among the submodules of M M M .
Proof. Every M B MB M B is a submodule, so I ( M ) ⊆ S ( M ) \mathcal{I}(M) \subseteq\mathcal{S}(M) I ( M ) ⊆ S ( M ) without assumptions. The reverse inclusion is exactly the definition of iso-multiplication. The cardinality estimate is immediate.
Remark 4.22. The preceding equality gives an obstruction to the iso-multiplication property. It does not characterize rings admitting iso-multiplication modules.
Theorem 4.23. Let A = A 1 × ⋯ × A n A=A_1 \times\cdots\times A_n A = A 1 × ⋯ × A n and M = ⨁ i = 1 n M i M=\bigoplus_{i=1}^{n}M_i M = ⨁ i = 1 n M i , where M i = M e i M_i=Me_i M i = M e i is regarded as a right A i A_i A i -module. Then
M A is im ⟺ ( M i ) A i is im for every i . M_A\text{ is im} \Longleftrightarrow(M_i)_{A_i}\text{ is im for every }i. M A is im ⟺ ( M i ) A i is im for every i . Proof. Every right submodule N N N decomposes as ⨁ N e i \bigoplus Ne_i ⨁ N e i , and each two-sided ideal as B 1 × ⋯ × B n B_1 \times\cdots\times B_n B 1 × ⋯ × B n . Consequently M B = ⨁ M i B i MB=\bigoplus M_iB_i M B = ⨁ M i B i . Componentwise isomorphisms prove sufficiency. Conversely, given N i ≤ M i N_i\leq M_i N i ≤ M i , view it as a submodule of M M M supported only in component i i i . An isomorphism N i ≅ M B N_i\cong MB N i ≅ M B must preserve each central idempotent and hence restrict to N i ≅ M i B i N_i\cong M_iB_i N i ≅ M i B i , proving necessity.
Remark 4.24. The Jordan–Hölder vector obstruction was proved in Proposition 4.12 ; it is not an independent recognition criterion. For example, P P P and S 1 ⊕ S 2 S_1\oplus S_2 S 1 ⊕ S 2 over T 2 ( k ) T_2(k) T 2 ( k ) have equal composition-factor vectors but are not isomorphic.
Local Loewy rigidity
We now consider local rings, meaning rings with a unique maximal right ideal (equivalently, rings whose quotient by the Jacobson radical is a division ring). In particular, every proper two-sided ideal lies in the Jacobson radical. The next theorem holds without a commutativity assumption; its converse fails for noncommutative rings.
Theorem 4.25. Let A A A be any (not necessarily commutative) local ring with Jacobson radical J J J , and let M A M_A M A be a right module satisfying M J h = 0 MJ^h=0 M J h = 0 for some integer h ≥ 1 h\geq1 h ≥ 1 . If M M M is an iso-multiplication module, then M M M is cyclic. Neither A A A nor M M M is assumed to be noetherian, artinian, or finitely generated. In particular, every iso-multiplication module over a right artinian local ring is cyclic and has finite composition length.
Proof. Assume M ≠ 0 M\ne0 M = 0 and choose h h h minimal with M J h = 0 MJ^h=0 M J h = 0 . Then M J h − 1 ≠ 0 MJ^{h-1}\ne0 M J h − 1 = 0 , so there is an x ∈ M x\in M x ∈ M with x J h − 1 ≠ 0 xJ^{h-1}\ne0 x J h − 1 = 0 . As x A ≤ M xA\leq M x A ≤ M and M M M is iso-multiplication, x A ≅ M B xA\cong MB x A ≅ M B for some two-sided ideal B ◃ A B\triangleleft A B ◃ A . If B ≠ A B\ne A B = A then B ⊆ J B\subseteq J B ⊆ J because A A A is local: every proper two-sided ideal is contained in its unique maximal right ideal. Consequently
( M B ) J h − 1 = M ( B J h − 1 ) ⊆ M J h = 0. (1) (MB)J^{h-1}=M(BJ^{h-1})\subseteq MJ^h=0. \tag*{(1)} ( M B ) J h − 1 = M ( B J h − 1 ) ⊆ M J h = 0. (1)
On the other hand, ( x A ) J h − 1 (xA)J^{h-1} ( x A ) J h − 1 contains x J h − 1 ≠ 0 xJ^{h-1} \ne0 x J h − 1 = 0 , contradicting the right A A A -module isomorphism x A ≅ M B xA \cong MB x A ≅ M B . Thus B = A B=A B = A and M ≅ x A M\cong xA M ≅ x A . An image of a cyclic module under an isomorphism is cyclic; hence M M M is cyclic. If A A A is right artinian, J J J is nilpotent and a cyclic right A A A -module has finite length. □ \square □
Example 4.27. Let k k k be any field, put D = k ( t ) D=k(t) D = k ( t ) , and let
T = ( 0 t 1 0 ) ∈ M 2 ( D ) . T=\begin{pmatrix}0 & t \\ 1 & 0\end{pmatrix}\in M_2(D). T = ( 0 1 t 0 ) ∈ M 2 ( D ) . The polynomial X 2 − t X^2-t X 2 − t is irreducible over D = k ( t ) D=k(t) D = k ( t ) . Indeed, if t = ( f ( t ) / g ( t ) ) 2 t=(f(t)/g(t))^2 t = ( f ( t ) / g ( t ) ) 2 with nonzero coprime polynomials f , g ∈ k [ t ] f,g\in k[t] f , g ∈ k [ t ] , then f ( t ) 2 = t g ( t ) 2 f(t)^2=tg(t)^2 f ( t ) 2 = t g ( t ) 2 . Since t t t is prime in k [ t ] k[t] k [ t ] , it divides f f f , say f = t h f=th f = t h . Substitution gives t 2 h 2 = t g 2 t^2h^2=tg^2 t 2 h 2 = t g 2 , hence g 2 = t h 2 g^2=th^2 g 2 = t h 2 and t ∣ g t\mid g t ∣ g , so t ∣ g t\mid g t ∣ g . This contradicts gcd ( f , g ) = 1 \gcd(f,g)=1 g cd( f , g ) = 1 . Thus t t t is not a square in D D D and the quadratic polynomial X 2 − t X^2-t X 2 − t is irreducible. Thus T T T has no one-dimensional invariant D D D -subspace in D 2 D^2 D 2 . The matrix T T T is transcendental over k k k : if p ( T ) = 0 p(T)=0 p ( T ) = 0 for p ∈ k [ X ] p\in k[X] p ∈ k [ X ] , write p ( X ) = p 0 ( X 2 ) + X p 1 ( X 2 ) p(X)=p_0(X^2)+Xp_1(X^2) p ( X ) = p 0 ( X 2 ) + X p 1 ( X 2 ) . As T 2 = t I 2 T^2=tI_2 T 2 = t I 2 , we have p ( T ) = p 0 ( t ) I 2 + p 1 ( t ) T p(T)=p_0(t)I_2+p_1(t)T p ( T ) = p 0 ( t ) I 2 + p 1 ( t ) T ; the off-diagonal entries imply p 1 ( t ) = 0 p_1(t)=0 p 1 ( t ) = 0 , then the diagonal entries imply p 0 ( t ) = 0 p_0(t)=0 p 0 ( t ) = 0 . As t t t is transcendental over k k k , both polynomials vanish, so p = 0 p=0 p = 0 . Thus the rule t ↦ T t\mapsto T t ↦ T extends to a unital field embedding ρ : D = k ( t ) ↪ M 2 ( D ) \rho:D=k(t)\hookrightarrow M_2(D) ρ : D = k ( t ) ↪ M 2 ( D ) , because, writing p ( T ) = u ( t ) I 2 + v ( t ) T p(T)=u(t)I_2+v(t)T p ( T ) = u ( t ) I 2 + v ( t ) T , its determinant is u ( t ) 2 − t v ( t ) 2 u(t)^2-tv(t)^2 u ( t ) 2 − t v ( t ) 2 , which is nonzero whenever ( u , v ) ≠ ( 0 , 0 ) (u,v)\ne(0,0) ( u , v ) = ( 0 , 0 ) by the irreducibility of X 2 − t X^2-t X 2 − t over D D D . Take V = D 2 V=D^2 V = D 2 as a right D D D -vector space of column vectors, with coordinatewise right multiplication v ⋅ b = ( v 1 b , v 2 b ) v\cdot b=(v_1b,v_2b) v ⋅ b = ( v 1 b , v 2 b ) . Independently define its left action by a ⋅ v = ρ ( a ) v a\cdot v=\rho(a)v a ⋅ v = ρ ( a ) v . The matrices ρ ( a ) \rho(a) ρ ( a ) act D D D -linearly on the right, so a ⋅ ( v b ) = ( a ⋅ v ) b a\cdot(vb)=(a\cdot v)b a ⋅ ( v b ) = ( a ⋅ v ) b as required for a bimodule. The two actions do not commute in the sense required for a commutative trivial extension: for v = ( 1 , 0 ) T v=(1,0)^T v = ( 1 , 0 ) T , one has ρ ( t ) v = ( 0 , 1 ) T ≠ ( t , 0 ) T = v t \rho(t)v=(0,1)^T\ne(t,0)^T=vt ρ ( t ) v = ( 0 , 1 ) T = ( t , 0 ) T = v t . Thus the ring constructed below is noncommutative. This makes V V V a ( D , D ) (D,D) ( D , D ) -bimodule with no nonzero proper subbimodule: a nonzero proper subbimodule would have to be a right D D D -line invariant under ρ ( t ) = T \rho(t)=T ρ ( t ) = T . Such a line would supply an eigenvector over D D D , contradicting the irreducibility of X 2 − t X^2-t X 2 − t .
Form the trivial extension
A = D × V , ( a , v ) ( b , w ) = ( a b , a ⋅ w + v b ) . A=D\times V,\qquad(a,v)(b,w)=(ab,a\cdot w+vb). A = D × V , ( a , v ) ( b , w ) = ( ab , a ⋅ w + v b ) . The subset J = 0 × V J=0\times V J = 0 × V satisfies J 2 = 0 J^2=0 J 2 = 0 , and A / J ≅ D A/J\cong D A / J ≅ D . An element ( a , v ) (a,v) ( a , v ) with a ≠ 0 a\ne0 a = 0 is invertible: since ( 0 , v ) 2 = 0 (0,v)^2=0 ( 0 , v ) 2 = 0 , the usual inverse of the unit ( a , 0 ) (a,0) ( a , 0 ) extends to ( a , v ) (a,v) ( a , v ) by a one-step nilpotent correction. Conversely, elements of J J J are nonunits. Thus A A A is a local noncommutative ring with Jacobson radical J J J . Its two-sided ideals are exactly 0 , J , A 0,J,A 0 , J , A : every proper two-sided ideal is contained in J J J , and such an ideal is exactly a ( D , D ) (D,D) ( D , D ) -subbimodule of V V V under the identification J = 0 × V J=0\times V J = 0 × V .
As a right A A A -module, J J J is a two-dimensional right D D D -vector space, whereas every one-dimensional right D D D -subspace W ≤ V W\leq V W ≤ V gives a nonzero right ideal 0 × W 0\times W 0 × W of A A A . The ideal multiples of the regular right module A A A_A A A are exactly 0 , J , A 0,J,A 0 , J , A . Their composition lengths are 0 , 2 , 3 0,2,3 0 , 2 , 3 , respectively:
J J J is a direct sum of two copies of the unique simple right A A A -module D A D_A D A , and A / J ≅ D A A/J \cong D_A A / J ≅ D A . The right ideal 0 × W 0 \times W 0 × W has composition length 1. Therefore A A A_A A A is cyclic, but not iso-multiplication. This shows that the converse to Theorem 4.25 fails even for a local ring with square-zero Jacobson radical. The argument uses the irreducibility of X 2 − t X^2-t X 2 − t over k ( t ) k(t) k ( t ) , which remains valid when char k = 2 \operatorname{char} k=2 char k = 2 ; it verifies in turn that ρ \rho ρ is a field embedding, that V V V has no proper nonzero subbimodule, that J 2 = 0 J^2=0 J 2 = 0 , and that the only two-sided ideals are 0 0 0 , J J J , A A A . The one-dimensional right subspace W W W then supplies the required right ideal of length one.
Corollary 4.28. For every right artinian local ring A A A , every iso-multiplication right module is a finitely generated cyclic module. If A A A is commutative, the converse holds; for noncommutative A A A it can fail even for the regular module A A A_A A A .
Proof. The forward assertion follows from Theorem 4.25 . For commutative A A A , the converse follows because every cyclic module is an ordinary multiplication module. The explicit local ring in Example 4.27 has square-zero radical and is right artinian (it has a finite composition series of length three as a right module over itself), providing the claimed counterexample. Notice that “right artinian” here is a property of A A A_A A A and does not require finite dimension over the central field k k k . □
The preceding examples separate multiplication modules from iso-multiplication modules. The next section gives restrictions of a different kind using Goldie localization.
Localization over semiprime right Goldie rings
We use the classical quotient ring of a semiprime right Goldie ring. The necessary facts on Ore localization and Goldie’s theorem can be found in [4 , 5 ] . No commutativity assumption on products of right ideals is made.
Let A A A be semiprime right Goldie and C C C the set of regular elements of A A A . Then C C C is a right denominator set and Q = A C − 1 Q=AC^{-1} Q = A C − 1 is semisimple Artinian. Write
Q ≅ ∏ i = 1 s M n i ( D i ) , Q \cong\prod_{i=1}^{s} M_{n_i}(D_i), Q ≅ i = 1 ∏ s M n i ( D i ) ,
where the D i D_i D i are division rings and n i ≥ 1 n_i \ge1 n i ≥ 1 . For a right A A A -module M M M , write M Q = M ⊗ A Q M_Q=M\otimes_A Q M Q = M ⊗ A Q . The module M Q M_Q M Q is semisimple, with possibly infinite multiplicities. For each N ≤ M N \le M N ≤ M , write ι N : N ⊗ A Q ⟶ M ⊗ A Q \iota_N:N\otimes_A Q \longrightarrow M\otimes_A Q ι N : N ⊗ A Q ⟶ M ⊗ A Q for the map induced by N ↪ M N \hookrightarrow M N ↪ M , and define
Loc M ( N ) : = im ( ι N ) ≤ M Q . \operatorname{Loc}_M(N):=\operatorname{im}(\iota_N)\le M_Q. Loc M ( N ) := im ( ι N ) ≤ M Q .
Because the classical right Ore localization is exact, ι N \iota_N ι N is injective. Thus N ⊗ A Q N\otimes_A Q N ⊗ A Q is canonically identified with the actual submodule Loc M ( N ) \operatorname{Loc}_M(N) Loc M ( N ) of M Q M_Q M Q . We reserve N Q N_Q N Q for this image when no confusion is possible; equality of these images is stronger than an abstract isomorphism of localized modules. The right-fraction convention Q = A C − 1 Q=AC^{-1} Q = A C − 1 requires Q Q Q to be flat as a left A A A -module in the tensor functor − ⊗ A Q -\otimes_A Q − ⊗ A Q ; this is the standard exactness property of the classical right Ore localization [4 , 5 ] .
Lemma 5.2. If B B B is a two-sided ideal of A A A , then B Q BQ B Q is a two-sided ideal of Q Q Q . Moreover, for every right A A A -module M M M there is a natural Q Q Q -module isomorphism
( M B ) ⊗ A Q ≅ M Q ( B Q ) . (MB) \otimes_A Q \cong M_Q(BQ). ( M B ) ⊗ A Q ≅ M Q ( B Q ) . Consequently, M Q ( B Q ) M_Q(BQ) M Q ( B Q ) is the direct sum of entire isotypic components of M Q M_Q M Q .
Proof. We use the right Ore convention Q = A C − 1 Q=AC^{-1} Q = A C − 1 : elements are represented by right fractions a c − 1 ac^{-1} a c − 1 with a ∈ A a\in A a ∈ A and c ∈ C c\in C c ∈ C . Put B Q = { ∑ j b j q j : b j ∈ B , q j ∈ Q } BQ=\{\sum_j b_jq_j:b_j\in B,q_j\in Q\} B Q = { ∑ j b j q j : b j ∈ B , q j ∈ Q } , a right Q Q Q -ideal. Because B B B is a two-sided ideal, a ( B Q ) ⊆ B Q a(BQ)\subseteq BQ a ( B Q ) ⊆ B Q for every a ∈ A a\in A a ∈ A . For a regular element c ∈ C c\in C c ∈ C , left multiplication
L c : B Q ⟶ B Q , y ↦ c y L_c:BQ\longrightarrow BQ,\qquad y\mapsto cy L c : B Q ⟶ B Q , y ↦ cy
is right Q Q Q -linear and injective: c c c is a unit in Q Q Q . As a right submodule of the finite-length semisimple regular module Q Q Q_Q Q Q , the module B Q BQ B Q has finite composition length. Since left multiplication commutes with right multiplication in Q Q Q , L c L_c L c is a right Q Q Q -module endomorphism; its kernel is zero because c c c is invertible in Q Q Q . Therefore this injective endomorphism of a finite-length module is surjective. In particular, for each y ∈ B Q y\in BQ y ∈ B Q the unique solution to c z = y cz=y cz = y in Q Q Q belongs to B Q BQ B Q , and hence c − 1 ( B Q ) ⊆ B Q c^{-1}(BQ)\subseteq BQ c − 1 ( B Q ) ⊆ B Q . Thus ( a c − 1 ) ( B Q ) ⊆ B Q (ac^{-1})(BQ)\subseteq BQ ( a c − 1 ) ( B Q ) ⊆ B Q for all a ∈ A a\in A a ∈ A and c ∈ C c\in C c ∈ C : the right ideal B Q BQ B Q is also a left Q Q Q -ideal. Since Q Q Q is semisimple Artinian, a central idempotent e e e satisfies B Q = e Q BQ=eQ B Q = e Q .
For a classical right Ore localization at regular elements, Q Q Q is flat as a left A A A -module (this is the standard Ore-localization exactness theorem); thus the − ⊗ A Q -\otimes_A Q − ⊗ A Q is exact. More explicitly, apply it to 0 → M B → M 0\to MB\to M 0 → M B → M ; exactness gives an embedding ( M B ) ⊗ A Q ↪ M Q (MB)\otimes_A Q\hookrightarrow M_Q ( M B ) ⊗ A Q ↪ M Q . The image is the right Q Q Q -span of elements ( m b ) ⊗ 1 (mb)\otimes1 ( mb ) ⊗ 1 with m ∈ M m\in M m ∈ M and b ∈ B b\in B b ∈ B . As ( m b ) ⊗ 1 = ( m ⊗ 1 ) b (mb)\otimes1=(m\otimes1)b ( mb ) ⊗ 1 = ( m ⊗ 1 ) b , this image is exactly M Q ( B Q ) M_Q(BQ) M Q ( B Q ) . This identifies the two localized modules without presupposing that M B ↪ M MB\hookrightarrow M M B ↪ M remains injective under an arbitrary nonflat tensor product. Finally M Q e M_Qe M Q e is the full direct sum of the simple constituents supported on the simple factors of Q Q Q on which e e e is 1 1 1 ; it cannot select only some copies of one simple type.
Lemma 5.3. For every simple Q Q Q -submodule S S S of M Q M_Q M Q there exists a cyclic right A A A -submodule N N N of M M M such that
N ⊗ A Q ≅ Q S . N\otimes_A Q\cong_Q S. N ⊗ A Q ≅ Q S . This holds even if M M M is not finitely generated or has nonzero C C C -torsion.
Proof. Choose 0 ≠ x ∈ S 0\ne x\in S 0 = x ∈ S . By the right Ore common-denominator property, every finite collection of right fractions in Q = A C − 1 Q=AC^{-1} Q = A C − 1 can be written with one denominator on the right. Consequently, for fractions q j = a j ′ c − 1 q_j=a'_jc^{-1} q j = a j ′ c − 1 occurring in x = ∑ j = 1 t ( m j ⊗ 1 ) q j x=\sum_{j=1}^{t}(m_j\otimes1)q_j x = ∑ j = 1 t ( m j ⊗ 1 ) q j , there is a common right denominator c ∈ C c\in C c ∈ C and numerators a j ′ ∈ A a'_j\in A a j ′ ∈ A with q j = a j ′ c − 1 q_j=a'_jc^{-1} q j = a j ′ c − 1 . Since tensor multiplication is right Q Q Q -linear, the finite sum becomes
x = ( ∑ j = 1 t m j a j ′ ⊗ 1 ) c − 1 = ( m ⊗ 1 ) c − 1 , m = ∑ j = 1 t m j a j ′ ∈ M . (2) x=\left(\sum_{j=1}^{t}m_ja'_j\otimes1\right)c^{-1}=(m\otimes1)c^{-1},\qquad m=\sum_{j=1}^{t}m_ja'_j\in M. \tag*{(2)} x = ( j = 1 ∑ t m j a j ′ ⊗ 1 ) c − 1 = ( m ⊗ 1 ) c − 1 , m = j = 1 ∑ t m j a j ′ ∈ M . (2)
Thus m ⊗ 1 = x c m \otimes1 = xc m ⊗ 1 = x c is nonzero and belongs to S S S . For N = m A N = mA N = m A , exact localization identifies N ⊗ A Q N \otimes_A Q N ⊗ A Q with its canonical image N Q = ( m ⊗ 1 ) Q N_Q = (m \otimes1)Q N Q = ( m ⊗ 1 ) Q inside M Q M_Q M Q . Since m ⊗ 1 ∈ S m \otimes1 \in S m ⊗ 1 ∈ S is nonzero, and S S S is simple, this image equals S S S as a submodule of M Q M_Q M Q . In particular the conclusion holds whether or not M M M has nonzero C C C -torsion.
Theorem 5.4. Let A A A be an arbitrary semiprime right Goldie ring and M A M_A M A an iso-multiplication module. Then
M ⊗ A Q c l r ( A ) M \otimes_A Q^{r}_{\mathrm{cl}}(A) M ⊗ A Q cl r ( A ) is multiplicity-free semisimple. Equivalently, if Q ≅ ∏ i = 1 s M n i ( D i ) Q \cong\prod_{i=1}^{s} M_{n_i}(D_i) Q ≅ ∏ i = 1 s M n i ( D i ) and V i V_i V i denotes the unique simple right module for its i i i th matrix factor, then
M ⊗ A Q ≅ ⨁ i ∈ E V i M \otimes_A Q \cong\bigoplus_{i \in E} V_i M ⊗ A Q ≅ i ∈ E ⨁ V i for some E ⊆ { 1 , … , s } E \subseteq\{1,\ldots,s\} E ⊆ { 1 , … , s } . No finiteness hypothesis on M M M is required. The conclusion also permits M Q = 0 M_Q = 0 M Q = 0 for a nonzero module M M M with C C C -torsion; the zero localization is multiplicity-free by convention.
Proof. Fix an index i i i for which the multiplicity of V i V_i V i in M Q M_Q M Q is nonzero, and choose a simple submodule S ≅ V i S \cong V_i S ≅ V i of M Q M_Q M Q . By Lemma 5.3, some cyclic submodule N ≤ M N \leq M N ≤ M has N Q ≅ S N_Q \cong S N Q ≅ S . Since M M M is iso-multiplication, N ≅ A M B N \cong_A MB N ≅ A M B for a two-sided ideal B ⊲ A B \lhd A B ⊲ A . Localizing this isomorphism and applying Lemma 5.2, we obtain
V i ≅ Q N Q ≅ Q ( M B ) Q ≅ Q M Q ( B Q ) . V_i \cong_Q N_Q \cong_Q (MB)_Q \cong_Q M_Q(BQ). V i ≅ Q N Q ≅ Q ( M B ) Q ≅ Q M Q ( B Q ) .
The canonical images N Q N_Q N Q and ( M B ) Q (MB)_Q ( M B ) Q are submodules of M Q M_Q M Q ; the displayed arrows assert abstract Q Q Q -module isomorphisms, not equality of those images. In particular, the argument does not identify an arbitrary isomorphism N ≅ A M B N \cong_A MB N ≅ A M B with an automorphism of M M M or of M Q M_Q M Q . By Lemma 5.2 there is a central idempotent e ∈ Q e \in Q e ∈ Q with B Q = e Q BQ = eQ B Q = e Q , so M Q ( B Q ) = M Q e M_Q(BQ) = M_Qe M Q ( B Q ) = M Q e is the sum of entire isotypic components of M Q M_Q M Q . Since this module is isomorphic to one simple module of type V i V_i V i , its V i V_i V i -multiplicity in M Q M_Q M Q is exactly one, and every other component selected by e e e vanishes. This holds for every occurring simple type.
Corollary 5.5. If A A A is semiprime right Goldie and A A A_A A A is iso-multiplication, then
Q c l r ( A ) ≅ D 1 × ⋯ × D s Q^{r}_{\mathrm{cl}}(A) \cong D_1 \times\cdots\times D_s Q cl r ( A ) ≅ D 1 × ⋯ × D s for division rings D 1 , … , D s D_1,\ldots,D_s D 1 , … , D s . In particular, the regular module of an order in a semisimple Artinian ring with a genuine matrix factor M n ( D ) M_n(D) M n ( D ) , n ≥ 2 n \geq2 n ≥ 2 , cannot be iso-multiplication.
Proof. For M = A A M = A_A M = A A we have M Q = Q Q M_Q = Q_Q M Q = Q Q . As a right module over M n ( D ) M_n(D) M n ( D ) , the regular module is the direct sum of n n n copies of its unique simple module. Theorem 5.4 forces every n i = 1 n_i = 1 n i = 1 .
Corollary 5.6. If A A A is semiprime right Goldie and the regular module A A A_A A A is iso-multiplication, then every idempotent of A A A is central.
Proof. By Corollary 5.5, the canonical embedding A ↪ Q c l r ( A ) A \hookrightarrow Q^{r}_{\mathrm{cl}}(A) A ↪ Q cl r ( A ) has codomain a finite direct product of division rings. Every idempotent in that product is a central coordinate idempotent, since a division ring has only the idempotents $0$ and 1 1 1 . Thus, for e 2 = e ∈ A e^2 = e \in A e 2 = e ∈ A and a ∈ A a \in A a ∈ A , the images of e a ea e a and a e ae a e in Q Q Q agree; injectivity gives e a = a e ea = ae e a = a e in $A.
Corollary 5.8. If A A A is prime right Goldie and M A M_A M A is iso-multiplication, then M Q M_Q M Q is either zero or a simple right module over the simple Artinian ring Q Q Q . In particular, if A A A {}_A A_A A A A is iso-multiplication, then A A A is a domain and Q Q Q is a division ring.
Proof. Goldie’s theorem makes Q Q Q simple Artinian; Theorem 5.4 gives the first statement. For the regular module Q Q Q_Q Q Q , multiplicity-freeness forces Q Q Q to be a division ring. Since A ↪ Q A \hookrightarrow Q A ↪ Q , nonzero elements of A A A have no zero product.
Theorem 5.9. Let A A A be semiprime right Goldie, let Q = Q c l r ( A ) Q=Q_{\mathrm{cl}}^{r}(A) Q = Q cl r ( A ) , and suppose that the canonical map M ⟶ M Q = M ⊗ A Q M\longrightarrow M_Q=M\otimes_A Q M ⟶ M Q = M ⊗ A Q is injective. If M Q M_Q M Q has finite composition length over Q Q Q , then
u.dim A ( M ) = l Q ( M Q ) . \operatorname{u.dim}_{A}(M)=l_Q(M_Q). u.dim A ( M ) = l Q ( M Q ) . In particular, if M M M is also an iso-multiplication module and Q ≅ ∏ i = 1 s M n i ( D i ) Q\cong\prod_{i=1}^{s}M_{n_i}(D_i) Q ≅ ∏ i = 1 s M n i ( D i ) , then
u.dim A ( M ) = l Q ( M Q ) = ∣ { i : the i th simple type occurs in M Q } ∣ ≤ s . \operatorname{u.dim}_{A}(M)=l_Q(M_Q)=\left|\left\{i:\text{ the }i\text{th simple type occurs in }M_Q\right\}\right|\leq s. u.dim A ( M ) = l Q ( M Q ) = ∣ { i : the i th simple type occurs in M Q } ∣ ≤ s . Neither finite generation of M M M nor noetherianity of A A A is assumed.
The uniform-dimension equality in the first statement is the standard Goldie-rank localization comparison; see [4 , 5 ] . The contribution specific to iso-multiplication is the bound by the number of Wedderburn factors, obtained by combining this equality with Theorem 5.4 .
Proof. Let d = l Q ( M Q ) d=l_Q(M_Q) d = l Q ( M Q ) . If N 1 ⊕ ⋯ ⊕ N t ≤ M N_1\oplus\cdots\oplus N_t\leq M N 1 ⊕ ⋯ ⊕ N t ≤ M is a direct sum of nonzero submodules, every ( N j ) Q (N_j)_Q ( N j ) Q is nonzero because M ⟶ M Q M\longrightarrow M_Q M ⟶ M Q is injective. Exactness of Ore localization gives a direct sum of t t t nonzero Q Q Q -submodules of M Q M_Q M Q , hence t ≤ d t\leq d t ≤ d . This proves u.dim A ( M ) ≤ d \operatorname{u.dim}_{A}(M)\leq d u.dim A ( M ) ≤ d .
Conversely, choose a decomposition M Q = S 1 ⊕ ⋯ ⊕ S d M_Q=S_1\oplus\cdots\oplus S_d M Q = S 1 ⊕ ⋯ ⊕ S d into simple right Q Q Q -modules. For each j j j select 0 ≠ x j ∈ S j 0\ne x_j\in S_j 0 = x j ∈ S j . The common-denominator argument in Lemma 5.3 writes x j = ( m j ⊗ 1 ) c j − 1 x_j=(m_j\otimes1)c_j^{-1} x j = ( m j ⊗ 1 ) c j − 1 with m j ∈ M m_j\in M m j ∈ M , c j ∈ C c_j\in C c j ∈ C . Since m j ⊗ 1 = x j c j ∈ S j m_j\otimes1=x_jc_j\in S_j m j ⊗ 1 = x j c j ∈ S j is nonzero, the submodule m j A m_jA m j A of M M M is nonzero and its image in M Q M_Q M Q lies in S j S_j S j . The sum m 1 A ⊕ ⋯ ⊕ m d A m_1A\oplus\cdots\oplus m_dA m 1 A ⊕ ⋯ ⊕ m d A is therefore direct inside M M M (by injectivity of M ⟶ M Q M\longrightarrow M_Q M ⟶ M Q ). Thus u.dim A ( M ) ≥ d \operatorname{u.dim}_{A}(M)\geq d u.dim A ( M ) ≥ d . Equality follows. The last displayed formula uses Theorem 5.4 .
Corollary 5.10. Let A A A be a semiprime right Goldie ring, let C C C be its set of regular elements, and write
Q = Q c l r ( A ) ≅ ∏ i = 1 s M n i ( D i ) . Q=Q_{\mathrm{cl}}^{r}(A)\cong\prod_{i=1}^{s}M_{n_i}(D_i). Q = Q cl r ( A ) ≅ i = 1 ∏ s M n i ( D i ) . Suppose that the right A A A -module M M M is faithful, C C C -torsion-free, and iso-multiplication. If V i V_i V i denotes the unique simple right module associated with the i i i th simple Artinian factor, then
M ⊗ A Q ≅ Q ⨁ i = 1 s V i , u.dim A ( M ) = s . M\otimes_A Q\cong_Q\bigoplus_{i=1}^{s}V_i,\qquad\operatorname{u.dim}_{A}(M)=s. M ⊗ A Q ≅ Q i = 1 ⨁ s V i , u.dim A ( M ) = s . In particular, no simple quotient-ring component is missing and no simple type occurs twice. No finite-generation hypothesis on M M M is needed.
Proof. Write M Q = M ⊗ A Q M_Q=M\otimes_A Q M Q = M ⊗ A Q . Since M M M is C C C -torsion-free, the canonical map M → M Q M\to M_Q M → M Q is injective. By Theorem 5.4 , each simple right Q Q Q -module type occurs in M Q M_Q M Q with multiplicity at most one. We prove that every type occurs.
Fix 1 ≤ i ≤ s 1\le i\le s 1 ≤ i ≤ s and let e i e_i e i be the nonzero central idempotent of Q Q Q corresponding to its i i i th factor. In the classical right Ore quotient ring, express e i = a c − 1 e_i=ac^{-1} e i = a c − 1 with a ∈ A a\in A a ∈ A and c ∈ C c\in C c ∈ C . Thus a = e i c ≠ 0 a=e_ic\ne0 a = e i c = 0 in Q Q Q , hence also in A A A since A ↪ Q A\hookrightarrow Q A ↪ Q is injective; moreover a ∈ e i Q ∩ A a\in e_iQ\cap A a ∈ e i Q ∩ A . This explicitly exhibits a nonzero element of A A A supported on the i i i th central component of Q Q Q . If M Q e i = 0 M_Qe_i=0 M Q e i = 0 , then ( m ⊗ 1 ) a = ( m ⊗ 1 ) e i c = 0 (m\otimes1)a=(m\otimes1)e_ic=0 ( m ⊗ 1 ) a = ( m ⊗ 1 ) e i c = 0 for every m ∈ M m\in M m ∈ M . Injectivity of M → M Q M\to M_Q M → M Q implies m a = 0 ma=0 ma = 0 for all m m m , contradicting faithfulness and a ≠ 0 a\ne0 a = 0 . Therefore M Q e i ≠ 0 M_Qe_i\ne0 M Q e i = 0 , so its i i i th simple type occurs. Multiplicity-freeness now yields the asserted direct sum. Finally Theorem 5.9 gives u.dim A ( M ) = l Q ( M Q ) = s \operatorname{u.dim}_A(M)=l_Q(M_Q)=s u.dim A ( M ) = l Q ( M Q ) = s .
Example 5.12. For A = Z × Z A=\mathbb{Z}\times\mathbb{Z} A = Z × Z and M = A A M=A_A M = A A , the iso-multiplication property is classical multiplication and Q = Q × Q Q=\mathbb{Q}\times\mathbb{Q} Q = Q × Q . We have u.dim A ( M ) = 2 = l Q ( M Q ) \operatorname{u.dim}_A(M)=2=l_Q(M_Q) u.dim A ( M ) = 2 = l Q ( M Q ) , so the general upper bound s s s can be attained. For the skew polynomial right principal ideal domain of Example 2.6 , the corresponding values are both 1.
Corollary 5.13. Let A A A be prime right Goldie, and let M ≠ 0 M\ne0 M = 0 be a C C C -torsion-free iso-multiplication right A A A -module. Then M M M is uniform. In particular, every nonzero pair of submodules of M M M has nonzero intersection.
Proof. Here Q Q Q is simple Artinian. Theorem 5.4 says M Q M_Q M Q is either zero or simple; torsion-freeness and M ≠ 0 M\ne0 M = 0 rule out zero. If N 1 , N 2 ≤ M N_1,N_2\le M N 1 , N 2 ≤ M are nonzero with zero intersection, then N 1 ⊕ N 2 N_1\oplus N_2 N 1 ⊕ N 2 embeds in M M M . Localizing gives two nonzero Q Q Q -submodules in direct sum inside the simple module M Q M_Q M Q , a contradiction.
Corollary 5.14. Let M A M_A M A be iso-multiplication and put A ‾ = A / Ann A ( M ) \overline{A}=A/\operatorname{Ann}_A(M) A = A / Ann A ( M ) . If A ‾ \overline{A} A is semiprime right Goldie, rational rigidity applies to M M M as a right A ‾ \overline{A} A -module. If M M M is also torsion-free over the regular elements of A ‾ \overline{A} A , the exact uniform-dimension formula applies.
Proof. Every submodule of M M M is the same over A A A and A ‾ \overline{A} A . If N ≅ M B N\cong MB N ≅ M B as A A A -modules for a two-sided ideal B ⊲ A B\lhd A B ⊲ A , this is an A ‾ \overline{A} A -isomorphism and M B = M B ‾ MB=M\overline{B} M B = M B for B ‾ = ( B + Ann A ( M ) ) / Ann A ( M ) ⊲ A ‾ \overline{B}=(B+\operatorname{Ann}_A(M))/\operatorname{Ann}_A(M)\lhd\overline{A} B = ( B + Ann A ( M )) / Ann A ( M ) ⊲ A . Hence M M M is iso-multiplication over A ‾ \overline{A} A , and the assertions follow from Theorems 5.4 and 5.9 .
Example 5.16. The skew polynomial ring A = k ( t ) [ x ; σ ] A=k(t)[x;\sigma] A = k ( t ) [ x ; σ ] of Example 2.6 is a noncommutative principal right ideal domain. Its right classical quotient ring is a division ring, and A A A_A A A is iso-multiplication without being a classical multiplication module. Thus Corollary 5.5 does not force commutativity, right duo behavior, or the ordinary multiplication property.
Example 5.17. The right Artinian algebra A = T 2 ( k ) A=T_2(k) A = T 2 ( k ) is right Goldie, and its regular module is iso-multiplication by Example 2.4. Yet A A A is not semiprime: its radical k e 12 ke_{12} k e 12 has square zero.
Thus the semiprime hypothesis is essential for applying the above quotient-ring mechanism; the example does not provide a semiprime counterexample to the conclusion.
Remark 5.18. A zero rational localization is vacuously multiplicity-free and therefore cannot characterize the iso-multiplication property; for example ( Z / p Z ) 2 (\mathbb{Z}/p\mathbb{Z})^2 ( Z / p Z ) 2 is not iso-multiplication over Z \mathbb{Z} Z although its tensor product with Q \mathbb{Q} Q vanishes. The next example shows that even a nonzero simple rational localization is insufficient.
Example 5.19. Let A = k [ x ] A=k[x] A = k [ x ] , S = A / ( x ) S=A/(x) S = A / ( x ) , and M = A ⊕ S M=A\oplus S M = A ⊕ S . With Q = k ( x ) Q=k(x) Q = k ( x ) we have
so the localization is nonzero and simple. Nevertheless M M M is not an iso-multiplication module. Indeed, for each ideal B = ( f ) B=(f) B = ( f ) the isomorphism type of M B M_B M B is 0 0 0 if f = 0 f=0 f = 0 , is [ A ] [A] [ A ] if 0 ≠ f ∈ ( x ) 0\ne f\in(x) 0 = f ∈ ( x ) , and is [ A ⊕ S ] [A\oplus S] [ A ⊕ S ] if f ∉ ( x ) f\notin(x) f ∈ / ( x ) . But S ≤ M S\le M S ≤ M is not isomorphic to any of these. Hence rational multiplicity-freeness is strictly a necessary condition; even a simple nonzero localization is not enough. This contrasts with the preceding example where all of M M M vanished under localization.
Remark 5.20. Corollary 5.5 strengthens Lemma 7.5 in a different direction: it applies to every semiprime right Goldie ring, without any principal-ideal, right-noetherian, or isoartinian assumption. It follows from the stated Goldie rigidity theorem and is not a separate assertion of Goldie’s theorem.
Rational submodules and essentiality
We also describe the submodules of M Q M_Q M Q and the essential submodules detected by localization. We retain the notation C C C , Q Q Q from Section 5 and the convention N Q = Loc M ( N ) = im ( N ⊗ A Q → M Q ) N_Q=\operatorname{Loc}_M(N)=\operatorname{im}(N\otimes_A Q\to M_Q) N Q = Loc M ( N ) = im ( N ⊗ A Q → M Q ) . In particular, formulas involving N Q N_Q N Q or L Q L_Q L Q below are equalities of submodules of the fixed ambient Q Q Q -module M Q M_Q M Q .
Lemma 5.21. Let A A A be semiprime right Goldie and M A M_A M A any module. For each Q Q Q -submodule W ≤ M Q W\le M_Q W ≤ M Q there is a submodule N ≤ M N\le M N ≤ M satisfying N Q = W N_Q=W N Q = W . If M → M Q M\to M_Q M → M Q is injective, one may take
N = { m ∈ M : m ⊗ 1 ∈ W } . N=\{m\in M:m\otimes1\in W\}. N = { m ∈ M : m ⊗ 1 ∈ W } . Proof. Let N N N be the preimage of W W W under the canonical map M → M Q M\to M_Q M → M Q . Clearly N Q ⊆ W N_Q\subseteq W N Q ⊆ W . Conversely, every w ∈ W w\in W w ∈ W can be written as ( m ⊗ 1 ) c − 1 (m\otimes1)c^{-1} ( m ⊗ 1 ) c − 1 with m ∈ M m\in M m ∈ M and c ∈ C c\in C c ∈ C by clearing finitely many right denominators. Then m ⊗ 1 = w c ∈ W m\otimes1=wc\in W m ⊗ 1 = w c ∈ W , so m ∈ N m\in N m ∈ N and w = ( m ⊗ 1 ) c − 1 ∈ N Q w=(m\otimes1)c^{-1}\in N_Q w = ( m ⊗ 1 ) c − 1 ∈ N Q . Thus N Q = W N_Q=W N Q = W . The same inverse-image formula defines N N N even if M → M Q M\to M_Q M → M Q is not injective; under injectivity it may additionally be identified with the intersection of M M M and W W W inside M Q M_Q M Q .
Theorem 5.22. Let A A A be semiprime right Goldie and let M A M_A M A be iso-multiplication. Write
M Q ≅ ⨁ i ∈ E V i M_Q\cong\bigoplus_{i\in E}V_i M Q ≅ i ∈ E ⨁ V i with pairwise nonisomorphic simple right Q Q Q -modules V i V_i V i , as guaranteed by Theorem 5.4. Then the map
L ( M ) ⟶ L Q ( M Q ) , N ↦ N Q \mathcal{L}(M)\longrightarrow\mathcal{L}_Q(M_Q),\qquad N\mapsto N_Q L ( M ) ⟶ L Q ( M Q ) , N ↦ N Q
is surjective, and L Q ( M Q ) \mathcal{L}_{Q}(M_{Q}) L Q ( M Q ) is a finite Boolean lattice with 2 ∣ E ∣ 2^{|E|} 2 ∣ E ∣ elements. Moreover every localized submodule is a localized ideal multiple: for each N ≤ M N \leq M N ≤ M there is B ⊲ A B \lhd A B ⊲ A such that
N Q = ( M B ) Q = M Q ( B Q ) . N_{Q}=(MB)_{Q}=M_{Q}(BQ). N Q = ( M B ) Q = M Q ( B Q ) .
Thus each rational submodule is of the form M Q e M_{Q}e M Q e for a central idempotent e ∈ Q e \in Q e ∈ Q ; e e e is unique up to components annihilating M Q M_{Q} M Q . In particular, the isomorphism-class information of localized submodules is controlled by the support set E E E .
Proof. Surjectivity follows from Lemma 5.21. Distinct V i V_i V i are nonisomorphic simple modules, so each Q Q Q -submodule of their direct sum is uniquely the sum of a subset of the V i V_i V i . This gives the Boolean lattice. By the iso-multiplication property, N ≅ M B N \cong MB N ≅ M B for some two-sided ideal B B B ; tensoring this abstract isomorphism yields N ⊗ A Q ≅ ( M B ) ⊗ A Q N \otimes_A Q \cong(MB) \otimes_A Q N ⊗ A Q ≅ ( M B ) ⊗ A Q . Exactness embeds both localized modules into M Q M_Q M Q as N Q N_Q N Q and ( M B ) Q (MB)_Q ( M B ) Q , respectively. Lemma 5.2 identifies ( M B ) Q = M Q ( B Q ) (MB)_Q=M_Q(BQ) ( M B ) Q = M Q ( B Q ) , a sum of entire isotypic components. Since M Q M_Q M Q contains each simple Q Q Q -type at most once, two isomorphic submodules of M Q M_Q M Q must be the same sum of the same simple components. Only at this point does the abstract isomorphism imply N Q = ( M B ) Q N_Q=(MB)_Q N Q = ( M B ) Q as submodules of M Q M_Q M Q .
Theorem 5.23. Let A A A be semiprime right Goldie and suppose M ⟶ M Q M \longrightarrow M_Q M ⟶ M Q is injective. For every N ≤ M N \leq M N ≤ M ,
N is essential in M ⟺ N Q = M Q . N\text{ is essential in }M \Longleftrightarrow N_Q=M_Q. N is essential in M ⟺ N Q = M Q . If, in addition, M M M is iso-multiplication, this is equivalent to saying that N Q N_Q N Q contains every simple Q Q Q -type occurring in M Q M_Q M Q .
Proof. First suppose N Q = M Q N_Q=M_Q N Q = M Q . Let 0 ≠ L ≤ M 0 \neq L \leq M 0 = L ≤ M . The injectivity of M ⟶ M Q M \longrightarrow M_Q M ⟶ M Q implies L Q ≠ 0 L_Q \neq0 L Q = 0 . Because Ore localization is exact, it preserves the intersection of the embedded submodules N , L N,L N , L :
( N ∩ L ) Q = N Q ∩ L Q = M Q ∩ L Q = L Q ≠ 0. (N \cap L)_Q=N_Q \cap L_Q=M_Q \cap L_Q=L_Q \neq0. ( N ∩ L ) Q = N Q ∩ L Q = M Q ∩ L Q = L Q = 0. Thus N ∩ L ≠ 0 N \cap L \neq0 N ∩ L = 0 , proving that N N N is essential in M M M .
For the reverse implication, suppose that N N N is essential in M M M but N Q ≠ M Q N_Q \neq M_Q N Q = M Q . Since Q Q Q is semisimple Artinian, N Q N_Q N Q is a direct summand of M Q M_Q M Q . Choose a nonzero simple submodule S S S of a complementary summand; then S ∩ N Q = 0 S \cap N_Q=0 S ∩ N Q = 0 . Lemma 5.3 supplies a cyclic submodule L = m A ≤ M L=mA \leq M L = m A ≤ M whose localized image is L Q = S L_Q=S L Q = S . The equality L Q = S ≠ 0 L_Q=S \neq0 L Q = S = 0 shows that L ≠ 0 L \neq0 L = 0 . Because the localization functor is exact, it preserves kernels and finite intersections: the intersection is the kernel of the difference map N ⊕ L ⟶ M N \oplus L \longrightarrow M N ⊕ L ⟶ M , ( n , l ) ↦ n − l (n,l) \mapsto n-l ( n , l ) ↦ n − l . Thus, inside M Q M_Q M Q ,
( N ∩ L ) Q = N Q ∩ L Q = N Q ∩ S = 0. (N \cap L)_Q=N_Q \cap L_Q=N_Q \cap S=0. ( N ∩ L ) Q = N Q ∩ L Q = N Q ∩ S = 0. As M ⟶ M Q M \longrightarrow M_Q M ⟶ M Q is injective, its restriction to N ∩ L N \cap L N ∩ L is also injective. Thus a nonzero element of N ∩ L N \cap L N ∩ L would have a nonzero image in ( N ∩ L ) Q (N \cap L)_Q ( N ∩ L ) Q , contradicting the displayed equality. Hence N ∩ L = 0 N \cap L=0 N ∩ L = 0 , in conflict with the essentiality of N N N . This proves N Q = M Q N_Q=M_Q N Q = M Q . If M M M is also iso-multiplication, Theorem 5.4 decomposes M Q M_Q M Q into pairwise nonisomorphic simple constituents; thus N Q = M Q N_Q=M_Q N Q = M Q is equivalent to N Q N_Q N Q containing all their types.
Corollary 5.24. Under the hypotheses of Theorem 5.22, the ring End Q ( M Q ) \operatorname{End}_{Q}(M_{Q}) End Q ( M Q ) is a finite direct product of division rings (with the zero ring in case M Q = 0 M_{Q}=0 M Q = 0 ).
Proof. The V i V_i V i are pairwise nonisomorphic simple modules. Therefore all off-diagonal Hom groups vanish and
End Q ( M Q ) ≅ ∏ i ∈ E End Q ( V i ) . \operatorname{End}_{Q}(M_{Q}) \cong\prod_{i\in E} \operatorname{End}_{Q}(V_i). End Q ( M Q ) ≅ i ∈ E ∏ End Q ( V i ) .
where each factor is a division ring by Schur’s lemma. Since E E E is a subset of the finite set of simple factors of Q Q Q , the product is finite.
Example 5.25. Let A = Z × Z A = \mathbb{Z} \times\mathbb{Z} A = Z × Z and M = A A M = A_A M = A A . Write Q r a t = Q Q_{\mathrm{rat}} = \mathbb{Q} Q rat = Q for the field of rational numbers and Q c l = Q c l r ( A ) ≅ Q r a t × Q r a t Q_{\mathrm{cl}} = Q_{\mathrm{cl}}^{r}(A) \cong Q_{\mathrm{rat}} \times Q_{\mathrm{rat}} Q cl = Q cl r ( A ) ≅ Q rat × Q rat . Then M ⊗ A Q c l ≅ Q r a t ⊕ Q r a t M \otimes_A Q_{\mathrm{cl}} \cong Q_{\mathrm{rat}} \oplus Q_{\mathrm{rat}} M ⊗ A Q cl ≅ Q rat ⊕ Q rat has two nonisomorphic simple Q c l Q_{\mathrm{cl}} Q cl -constituents. Its localized submodule lattice is the four-element Boolean lattice. The submodule N = ( 2 Z ) × ( 3 Z ) N = (2\mathbb{Z}) \times(3\mathbb{Z}) N = ( 2 Z ) × ( 3 Z ) is essential, and N ⊗ A Q c l = M ⊗ A Q c l N \otimes_A Q_{\mathrm{cl}} = M \otimes_A Q_{\mathrm{cl}} N ⊗ A Q cl = M ⊗ A Q cl inside the fixed localization; by contrast, Z × 0 \mathbb{Z} \times0 Z × 0 is not essential, and its localization is only one Boolean atom.
Remark 5.26. The Boolean conclusion is a statement about the localized lattice, not a claim that L ( M ) \mathcal{L}(M) L ( M ) itself is Boolean. For example, the regular module over Z \mathbb{Z} Z is iso-multiplication but its actual submodule lattice contains many proper nonzero ideals. The injectivity hypothesis of Theorem Theorem 5.23 cannot be omitted: for M = Z / p Z M = \mathbb{Z}/p\mathbb{Z} M = Z / p Z , every submodule localizes to zero, although 0 0 0 is not essential in M M M .
Morita equivalence and factor rings
We first consider the behavior of iso-multiplication modules under Morita equivalence. Under the standard matrix equivalence, M A M_A M A corresponds to a row module, not to the regular module of the matrix ring. In particular, Morita equivalence does not identify the regular right A A A -module A A A_A A A with the regular right M n ( A ) M_n(A) M n ( A ) -module M n ( A ) M n ( A ) M_n(A)_{M_n(A)} M n ( A ) M n ( A ) when n > 1 n > 1 n > 1 .
Theorem 6.1. Let A A A be an arbitrary unital ring, n ≥ 1 n \ge1 n ≥ 1 , S = M n ( A ) S = M_n(A) S = M n ( A ) , and M A M_A M A any right module. Endow M 1 × n M^{1\times n} M 1 × n with its natural right S S S -module structure by matrix multiplication. Then
M A is im ⟺ M S 1 × n is im M_A\text{ is im}\ \Longleftrightarrow\ M^{1\times n}_S\text{ is im} M A is im ⟺ M S 1 × n is im The same equivalence holds for classical multiplication modules.
Proof. Let e = e 11 ∈ S e = e_{11} \in S e = e 11 ∈ S . The standard Morita equivalence between right A A A -modules and right S S S -modules sends M M M to F ( M ) = M 1 × n F(M) = M^{1\times n} F ( M ) = M 1 × n , with inverse X ↦ X e X \mapsto Xe X ↦ X e ; its exactness and its lattice bijection carry every A A A -submodule N ≤ M N \le M N ≤ M to the corresponding S S S -submodule F ( N ) ≤ F ( M ) F(N) \le F(M) F ( N ) ≤ F ( M ) .
Every two-sided ideal of S S S has the form M n ( B ) M_n(B) M n ( B ) for a unique two-sided ideal B B B of A A A . Direct calculation of the entries of row-matrix products gives the equality of submodules
F ( M ) M n ( B ) = ( M B ) 1 × n = F ( M B ) . F(M)M_n(B) = (MB)^{1\times n} = F(MB). F ( M ) M n ( B ) = ( M B ) 1 × n = F ( M B ) .
Suppose M M M is iso-multiplication and take X ≤ F ( M ) X \le F(M) X ≤ F ( M ) . Write X = F ( N ) X = F(N) X = F ( N ) under the equivalence. For some B ⊲ A B \lhd A B ⊲ A , N ≅ M B N \cong MB N ≅ M B . Applying F F F yields X ≅ F ( M B ) = F ( M ) M n ( B ) X \cong F(MB) = F(M)M_n(B) X ≅ F ( M B ) = F ( M ) M n ( B ) ; hence F ( M ) F(M) F ( M ) is iso-multiplication. Conversely, if F ( M ) F(M) F ( M ) is iso-multiplication, apply the inverse equivalence to the isomorphism F ( N ) ≅ F ( M ) M n ( B ) F(N) \cong F(M)M_n(B) F ( N ) ≅ F ( M ) M n ( B ) to obtain N ≅ M B N \cong MB N ≅ M B . Replacing isomorphisms by equalities and using the submodule equality displayed above gives the identical argument for multiplication modules.
Corollary 6.2. For any equivalence F : Mod - A → Mod - S F : \operatorname{Mod}\text{-}A \to\operatorname{Mod}\text{-}S F : Mod - A → Mod - S of categories of right modules, M A M_A M A is iso-multiplication if and only if F ( M ) S F(M)_S F ( M ) S is iso-multiplication. The same is true of the classical multiplication condition.
Proof. Choose an ( A , S ) (A,S) ( A , S ) -bimodule A P S {}_A P_S A P S that is a Morita progenerator with F ( M ) ≅ M ⊗ A P F(M) \cong M \otimes_A P F ( M ) ≅ M ⊗ A P for right modules; thus tensoring on the right by P P P maps Mod - A \operatorname{Mod}\text{-}A Mod - A to Mod - S \operatorname{Mod}\text{-}S Mod - S . For a two-sided ideal B ⊲ A B \lhd A B ⊲ A , let J ⊲ S J \lhd S J ⊲ S be the corresponding ideal under the Morita ideal-lattice correspondence, characterized by B P = P J BP = PJ B P = P J . Since tensoring with P P P is an exact Morita equivalence, the inclusion M B ↪ M MB \hookrightarrow M M B ↪ M induces an injection ( M B ) ⊗ A P ↪ M ⊗ A P (MB) \otimes_A P \hookrightarrow M \otimes_A P ( M B ) ⊗ A P ↪ M ⊗ A P . Identifying its image with a submodule of M ⊗ A P M \otimes_A P M ⊗ A P , we obtain
( M B ) ⊗ A P = ( M ⊗ A P ) J . (MB) \otimes_A P = (M \otimes_A P)J. ( M B ) ⊗ A P = ( M ⊗ A P ) J .
Indeed, in M ⊗ A P M \otimes_A P M ⊗ A P one has m b ⊗ p = m ⊗ b p mb \otimes p = m \otimes bp mb ⊗ p = m ⊗ b p , so the image of ( M B ) ⊗ A P (MB) \otimes_A P ( M B ) ⊗ A P is the submodule generated by the elements m ⊗ b p m \otimes bp m ⊗ b p with b ∈ B b \in B b ∈ B . Since B P = P J BP = PJ B P = P J , this is exactly the submodule generated by m ⊗ p j m \otimes pj m ⊗ p j with p ∈ P p \in P p ∈ P and j ∈ J j \in J j ∈ J , namely ( M ⊗ A P ) J (M \otimes_A P)J ( M ⊗ A P ) J . The equivalence also preserves and reflects submodules and their isomorphisms. Thus the defining statement “each submodule is isomorphic to an ideal multiple” is carried to precisely the corresponding statement over S S S . Equalities yield the m-assertion. □
Example 6.3. Let A = T 2 ( k ) A = T_2(k) A = T 2 ( k ) and n ≥ 1 n \ge1 n ≥ 1 , and set S = M n ( A ) S = M_n(A) S = M n ( A ) . By Example 2.4 and Theorem 6.1 , the right S S S -module A 1 × n A^{1 \times n} A 1 × n is iso-multiplication but is not a classical multiplication module. This is compatible with the fact that the right regular module S S S_S S S is not iso-multiplication for n ≥ 2 n \ge2 n ≥ 2 : A 1 × n A^{1 \times n} A 1 × n and S S S_S S S are different right S S S -modules. The latter failure follows from a rank obstruction or, since A A A is not semiprime, directly from the Morita inverse S e ≅ A n Se \cong A^n S e ≅ A n and the classification of modules over A A A : A n A^n A n is not iso-multiplication when n ≥ 2 n \ge2 n ≥ 2 .
Corollary 6.4. Let S = M n ( T 2 ( k ) ) S = M_n(T_2(k)) S = M n ( T 2 ( k )) and let F F F be the standard Morita equivalence Mod - T 2 ( k ) → Mod - S \operatorname{Mod}\text{-}T_2(k) \to\operatorname{Mod}\text{-}S Mod - T 2 ( k ) → Mod - S . The iso-multiplication right S S S -modules, with no finiteness restriction, are exactly the Morita images
F ( 0 ) , F ( S 1 ) , F ( S 2 ) , F ( P ) , F ( S 1 ⊕ S 2 ) , F ( P ⊕ S 2 ) . F(0), \qquad F(S_1), \qquad F(S_2), \qquad F(P), \qquad F(S_1 \oplus S_2), \qquad F(P \oplus S_2). F ( 0 ) , F ( S 1 ) , F ( S 2 ) , F ( P ) , F ( S 1 ⊕ S 2 ) , F ( P ⊕ S 2 ) . Among these six types, only F ( P ⊕ S 2 ) F(P \oplus S_2) F ( P ⊕ S 2 ) fails to be a classical multiplication module.
Proof. An equivalence is essentially surjective and bijects isomorphism classes. Combine Theorem 6.1 with Theorem 3.3 , applying the last assertion of Theorem 6.1 as well. □
Factor rings and applications
We give a factor-ring criterion and some consequences of the preceding results.
Passing through arbitrary factor rings. The following property is preserved when a module is viewed over a ring and over a factor ring acting faithfully on it.
Theorem 6.6. Let A A A be any unital ring, H ⊲ A H \lhd A H ⊲ A a two-sided ideal, and R = A / H R = A/H R = A / H . Let M M M be a unital right R R R -module, also considered as a right A A A -module via the quotient homomorphism A → R A \to R A → R . Then
M A is im ⟺ M R is im , M A is an m -module ⟺ M R is a multiplication module . M_A\text{ is im} \iff M_R\text{ is im}, \qquad M_A\text{ is an }m\text{-module} \iff M_R\text{ is a multiplication module}. M A is im ⟺ M R is im , M A is an m -module ⟺ M R is a multiplication module . No commutativity, chain condition, or finiteness assumption is required.
Proof. Since H H H annihilates M M M , the A A A -submodules and R R R -submodules of M M M coincide. Moreover, a map between such modules is A A A -linear if and only if it is R R R -linear. For any two-sided ideal B ◃ A B \triangleleft A B ◃ A , put B R = ( B + H ) / H ◃ R B_R=(B+H)/H \triangleleft R B R = ( B + H ) / H ◃ R . With the two actions distinguished,
M ⋅ A B = M ⋅ R B R M \cdot_A B=M \cdot_R B_R M ⋅ A B = M ⋅ R B R
as equal additive subgroups of M M M . Conversely, every two-sided ideal of R R R arises as the image of a two-sided ideal of A A A containing H H H . Thus the same submodules arise as ideal multiples for both actions, and the same isomorphisms are available. The assertions for equality and for isomorphism of submodules follow separately.
Example 6.7. Let A = T n ( k ) A=T_n(k) A = T n ( k ) , n ≥ 3 n\ge3 n ≥ 3 , and define the surjective homomorphism π : A → T 2 ( k ) \pi:A\to T_2(k) π : A → T 2 ( k ) by taking the upper-left 2 × 2 2\times2 2 × 2 block. Indeed the upper-left 2 × 2 2\times2 2 × 2 block of a product of upper triangular matrices depends only on their upper-left blocks. Take M = T 2 ( k ) T 2 ( k ) M=T_2(k)_{T_2(k)} M = T 2 ( k ) T 2 ( k ) , with the A A A -action x ⋅ a = x π ( a ) x\cdot a=x\pi(a) x ⋅ a = x π ( a ) . By Theorem 6.6 and Example 2.4, M A M_A M A is iso-multiplication but not a multiplication module. A witness is the right submodule k e 22 ≤ T 2 ( k ) ke_{22}\le T_2(k) k e 22 ≤ T 2 ( k ) ; it is isomorphic to k e 12 ke_{12} k e 12 , but is not itself an ideal multiple. This construction proves existence for all T n ( k ) T_n(k) T n ( k ) ; it does not classify the iso-multiplication modules over T n ( k ) T_n(k) T n ( k ) .
Example 6.8. Let 0 ≠ V 0\ne V 0 = V be any k k k -vector space, including dim k V = 2 \dim_k V=2 dim k V = 2 , and let A ( V ) A(V) A ( V ) be the triangular algebra of Theorem 2.11 . Its right regular module is iso-multiplication but not a multiplication module. When dim k V = 2 \dim_k V=2 dim k V = 2 this is the two-arrow (Kronecker) triangular algebra, giving a direct example in a representation-infinite setting. The result is an example for the regular module, not a classification of arbitrary Kronecker representations.
A sharpened consequence of Goldie rigidity.
Corollary 6.9. Suppose A A A is semiprime right Goldie and A A A_A A A is iso-multiplication. Then
Q c l r ( A ) ≅ D 1 × ⋯ × D s , A ↪ D 1 × ⋯ × D s , u.dim ( A A ) = s , Q_{\mathrm{cl}}^r(A)\cong D_1\times\cdots\times D_s,\qquad A\hookrightarrow D_1\times\cdots\times D_s,\qquad\operatorname{u.dim}(A_A)=s, Q cl r ( A ) ≅ D 1 × ⋯ × D s , A ↪ D 1 × ⋯ × D s , u.dim ( A A ) = s , for division rings D i D_i D i . In particular, A A A has no nonzero nilpotent elements. If A A A is prime, it is a domain and has right uniform dimension one.
Proof. Corollary 5.5 gives the displayed decomposition of the classical right quotient ring. The classical localization map of a semiprime right Goldie ring is injective, so A A A embeds in this direct product. Every nilpotent element in a division ring is zero, and the same holds coordinatewise in a finite direct product, proving reducedness of A A A . The right uniform dimension of A A A is the length of the right regular Q Q Q -module, by the standard Goldie rank identity. Each division-ring factor contributes exactly one simple summand, so this dimension equals s s s . If A A A is prime, its classical quotient ring is simple Artinian, hence there is only one factor; its embedding in a division ring makes A A A a domain.
Remark 6.10. The conclusion that Q Q Q is a finite product of division rings is an obstruction to the iso-multiplication property, not a converse. Even when Q Q Q is a field, an A A A -module can have simple nonzero rationalization and still fail iso-multiplication: the module k [ x ] k[x] k [ x ] over k [ x ] / ( x ) k[x]/(x) k [ x ] / ( x ) described in Example 5.19 has precisely this feature. Its torsion summand is invisible to rationalization but visible among integral submodule types.
A direct length-spectrum falsification test.
Example 6.12. Let A = k [ ϵ ] / ( ϵ 2 ) A=k[\epsilon]/(\epsilon^2) A = k [ ϵ ] / ( ϵ 2 ) and M = k ⊕ k M=k\oplus k M = k ⊕ k , with ϵ M = 0 \epsilon M=0 ϵ M = 0 . Then ℓ A ( M ) = 2 \ell_A(M)=2 ℓ A ( M ) = 2 , while its ideal multiples are M 0 = 0 M0=0 M 0 = 0 , M ( ϵ ) = 0 M(\epsilon)=0 M ( ϵ ) = 0 , and M A = M MA=M M A = M : the ideal ( ϵ ) (\epsilon) ( ϵ ) annihilates M M M , whereas the unit ideal A A A acts as the identity. Thus Λ ( M ) = { 0 , 2 } \Lambda(M)=\{0,2\} Λ ( M ) = { 0 , 2 } and length one is missing. Hence M M M is not iso-multiplication. This calculation also illustrates that one need not classify all submodules before applying the test.
Corollary 6.13. Suppose A A A has exactly t t t two-sided ideals, M A M_A M A is iso-multiplication and ℓ A ( M ) = t − 1 < ∞ \ell_A(M)=t-1<\infty ℓ A ( M ) = t − 1 < ∞ . Then the t t t numbers ℓ A ( M B ) \ell_A(MB) ℓ A ( M B ) , as B B B runs through the distinct two-sided ideals, are precisely 0 , 1 , … , t − 1 0,1,\ldots,t-1 0 , 1 , … , t − 1 , each occurring exactly once.
Proof. Proposition 4.1 produces all t t t different lengths 0 , … , t − 1 0,\ldots,t-1 0 , … , t − 1 among the ideal multiples, whereas there are only t t t choices for B B B . The pigeonhole principle proves the asserted one-to-one assignment.
The term isonoetherian refers to stabilization of isomorphism types along ascending chains, whereas isoartinian refers to such stabilization along descending chains [11 ] . The families above show that iso-multiplication, unlike the classical ascending and descending chain conditions, is an isomorphism-spectrum representation condition. In particular the infinite-dimensional A ( V ) A(V) A ( V ) example is iso-multiplication and isoartinian but not isonoetherian (Corollary 2.12 ); finite-length iso-multiplication modules satisfy both classical chain conditions, independently of the iso-multiplication definition.
A brief commutative comparison.
This observation is included only to delimit the scope of the present investigation, not as an independent main result. Existing studies of classical multiplication modules, including the locally cyclic projective and one-dimensional domain cases [1 , 2 ] , are relevant to further comparisons. We do not claim either a general equivalence or a separating example for arbitrary commutative rings. Our main results concern genuinely noncommutative examples, classifications, and obstructions.
Chain conditions and further remarks
Recall that a module is isonoetherian (respectively, isoartinian) if the terms of each ascending (respectively, descending) chain are eventually isomorphic [11 ] . The iso-multiplication condition alone does not compare ideal multiples by inclusion.
Theorem 7.1. Assume M A M_A M A is iso-multiplication, and that (i) only finitely many isomorphism types [ M B ] [M_B] [ M B ] occur as B B B ranges over the two-sided ideals of A A A , and (ii) whenever two submodules of M M M embed in one another, they are isomorphic. Then M M M is both isonoetherian and isoartinian. In particular, hypothesis (i) holds if A A A has only finitely many two-sided ideals.
Proof. The inclusion chain of submodules induces a monotone chain in the finite preorder of embeddability on isomorphism types. Condition (ii) makes the preorder antisymmetric, hence a finite partial order. Every ascending and descending chain in a finite partial order is eventually constant, so the submodules are eventually isomorphic.
Remark 7.2. Finiteness of the ideal lattice alone is not substituted for the Schröder–Bernstein condition (ii): different isomorphism types of infinite modules may embed in one another. The more general embedding criterion below keeps these logically independent requirements visible.
Remark 7.3. For the triangular algebra A ( V ) A(V) A ( V ) with V V V infinite-dimensional, Corollary 2.12 gives an iso-multiplication module which is isoartinian but not isonoetherian. Thus the two iso-chain conditions must be investigated separately even within the present class.
Corollary 7.4. If D D D is a domain in which every right ideal is principal, then D D D_D D D is isonoetherian and isoartinian in addition to being iso-multiplication. The latter property is classical multiplication precisely when all right ideals are two-sided.
Proof. The iso-multiplication and classical-multiplication assertions are Proposition 2.5 . Every nonzero right ideal is isomorphic to D D D_D D D ; hence along any ascending or descending chain, the isomorphism type can change between [ 0 ] [0] [ 0 ] and [ D ] [D] [ D ] at most once.
Lemma 7.5. Let D D D be a principal right ideal domain and n ≥ 2 n \ge2 n ≥ 2 . The regular right module of S = M n ( D ) S=M_n(D) S = M n ( D ) is not iso-multiplication.
Proof. Every two-sided ideal of S S S is M n ( I ) M_n(I) M n ( I ) for a two-sided ideal I I I of D D D . If I ≠ 0 I \ne0 I = 0 , then I = d D I=dD I = d D for some d ≠ 0 d \ne0 d = 0 , so left multiplication by the diagonal matrix d 1 n d1_n d 1 n gives an injective right S S S -linear map S → S S \to S S → S with image M n ( d D ) = M n ( I ) M_n(dD)=M_n(I) M n ( d D ) = M n ( I ) . Hence every nonzero two-sided ideal of S S S is isomorphic to S S S_S S S . But the nonzero right ideal e 11 S e_{11}S e 11 S is not isomorphic to S S S_S S S : under the standard Morita equivalence, they correspond to D D D_D D D and D D n D_D^n D D n , respectively. A principal right ideal domain is a right Ore domain and has well-defined rank over its classical division ring of fractions, so these latter modules have different ranks when n ≥ 2 n \ge2 n ≥ 2 . Thus e 11 S e_{11}S e 11 S cannot be isomorphic to any ideal multiple of S S S .
The following corollary is not an independent classification of semiprime isoartinian rings. It combines the previously established ring classification of Facchini–Nazemian [11 Theorem 4.7(1)] with the matrix-size obstruction above. The classification into products of matrix rings is the previously published result of Facchini–Nazemian, not an original result of this paper; the additional conclusion eliminates factors of size greater than one when the right regular module is iso-multiplication.
Corollary 7.6. Let A A A be a semiprime right noetherian right isoartinian ring. Then A A A_A A A is im if and only if A ≅ D 1 × ⋯ × D s A \cong D_1 \times\cdots\times D_s A ≅ D 1 × ⋯ × D s for principal right ideal domains D 1 , … , D s D_1,\ldots,D_s D 1 , … , D s . In this case A A A_A A A is both isoartinian and isonoetherian.
Proof. Assume that A A A_A A A is iso-multiplication. Under the stated semiprime, right-noetherian and right-isoartinian hypotheses, [11 Theorem 4.7(1)] gives a finite decomposition
A ≅ ∏ i = 1 s M n i ( D i ) , A \cong\prod_{i=1}^{s} M_{n_i}(D_i), A ≅ i = 1 ∏ s M n i ( D i ) , where each D i D_i D i is a principal right ideal domain. By Theorem 4.23, the regular module of each matrix factor is iso-multiplication. Lemma 7.5 forces n i = 1 n_i=1 n i = 1 for every i i i , proving the required decomposition.
Conversely, let A = ∏ i = 1 s D i A=\prod_{i=1}^{s}D_i A = ∏ i = 1 s D i , where every D i D_i D i is a principal right ideal domain. Each right ideal is N = ∏ i = 1 s N i N=\prod_{i=1}^{s}N_i N = ∏ i = 1 s N i , with each N i N_i N i either zero or isomorphic to ( D i ) D i (D_i)_{D_i} ( D i ) D i . Let I = ∏ i = 1 s I i I=\prod_{i=1}^{s}I_i I = ∏ i = 1 s I i , where I i = D i I_i=D_i I i = D i when N i ≠ 0 N_i\ne0 N i = 0 and I i = 0 I_i=0 I i = 0 otherwise. Then I ◃ A I\triangleleft A I ◃ A and N ≅ A A I = A I N\cong_A AI=AI N ≅ A A I = A I , proving iso-multiplication. For an ascending or descending chain of right ideals, the supports { i : N i ≠ 0 } \{i:N_i\ne0\} { i : N i = 0 } eventually stabilize, since they form a monotone sequence of subsets of the finite set { 1 , … , s } \{1,\ldots,s\} { 1 , … , s } . All terms with the same support are isomorphic as right A A A -modules. Thus A A A_A A A is both isonoetherian and isoartinian.
Remark 7.7. The right-noetherian assumption in Corollary 7.6 is part of the cited classification, not an assumption that may silently be removed. Nazemian-Yazdonov [13 ] constructed a prime right and left isoartinian ring which is neither right nor left noetherian (preprint dated 16 October 2026). That construction does not itself establish whether their ring is iso-multiplication; it does decisively invalidate any attempt to assume semiprime isoartinian rings must be noetherian.
Corollary 7.8. If A A A is a simple Artinian ring, then A A A_A A A is iso-multiplication if and only if A A A is a division ring.
Proof. This is immediate from Corollary 5.5 and the fact that a simple Artinian division-ring product has one factor.
An embedding criterion
For an arbitrary right module M M M let T ( M ) = { [ M B ] : B ◃ A } \mathcal{T}(M)=\{[MB]:B\triangleleft A\} T ( M ) = {[ M B ] : B ◃ A } be the collection of isomorphism classes of ideal multiples, where [ X ] [X] [ X ] denotes the class of X X X . Define [ X ] ≤ [ Y ] [X]\leq[Y] [ X ] ≤ [ Y ] when X X X embeds as a submodule of Y Y Y . This is in general a preorder, rather than a partial order: mutual embeddings need not imply isomorphism for arbitrary modules.
Theorem 7.9. Suppose M M M is iso-multiplication and the following hold:
(i) whenever X = M B X=MB X = M B and Y = M C Y=MC Y = M C embed in one another, then X ≅ Y X\cong Y X ≅ Y ;
(ii) the resulting partially ordered set of isomorphism types T ( M ) \mathcal{T}(M) T ( M ) has both the ascending and descending chain conditions.
Then M M M is both isonoetherian and isoartinian.
Proof. Let N 1 ⊆ N 2 ⊆ ⋯ N_1\subseteq N_2\subseteq\cdots N 1 ⊆ N 2 ⊆ ⋯ be an ascending chain of submodules of M M M . Since M M M is an iso-multiplication module, for every i i i there is a two-sided ideal B i ◃ A B_i\triangleleft A B i ◃ A such that N i ≅ M B i N_i\cong MB_i N i ≅ M B i . The inclusion N i ↪ N i + 1 N_i\hookrightarrow N_{i+1} N i ↪ N i + 1 , together with these isomorphisms, gives an embedding M B i ↪ M B i + 1 MB_i\hookrightarrow MB_{i+1} M B i ↪ M B i + 1 . Consequently, [ M B 1 ] ≤ [ M B 2 ] ≤ ⋯ [MB_1]\leq[MB_2]\leq\cdots [ M B 1 ] ≤ [ M B 2 ] ≤ ⋯ . By condition (ii), this ascending chain of isomorphism
classes eventually stabilizes. Hence there is an integer n 0 n_0 n 0 such that [ M B i ] = [ M B j ] [MB_i] = [MB_j] [ M B i ] = [ M B j ] for all i , j ≥ n 0 i,j \ge n_0 i , j ≥ n 0 . It follows that N i ≅ M B i ≅ M B j ≅ N j N_i \cong MB_i \cong MB_j \cong N_j N i ≅ M B i ≅ M B j ≅ N j for all i , j ≥ n 0 i,j \ge n_0 i , j ≥ n 0 . Thus M M M is isonoetherian. The descending case is analogous. If N 1 ⊇ N 2 ⊇ ⋯ N_1 \supseteq N_2 \supseteq\cdots N 1 ⊇ N 2 ⊇ ⋯ , then the corresponding ideal multiples give [ M B 1 ] ⪰ [ M B 2 ] ⪰ ⋯ [MB_1] \succeq[MB_2] \succeq\cdots [ M B 1 ] ⪰ [ M B 2 ] ⪰ ⋯ . The descending chain condition in (ii) implies eventual stabilization of these isomorphism classes. Therefore M M M is isoartinian.
Remark 7.10. This criterion is structural but conditional. It does not assert that the iso-multiplication condition alone imposes a chain condition. The mutual-embedding hypothesis cannot be omitted just by calling ≤ \leq ≤ a partial order. The A ( V ) A(V) A ( V ) examples show why explicit controls on ideal multiples are required.
Examples and limitations
Table 3. Comparison of multiplication modules and iso-multiplication modules
Theorem 3.3 applies to all modules over T 2 ( k ) T_2(k) T 2 ( k ) , but not to arbitrary Artinian rings. In general, the iso-multiplication property passes neither to submodules (Example 2.8 ) nor to direct sums (Example 2.9 ). The identity N = M ( N : M ) N = M(N : M) N = M ( N : M ) for multiplication modules also fails for iso-multiplication modules (Example 2.4).
Connections with previous results
Tuganbaev’s definition of multiplication modules over noncommutative rings requires the equality N = M B N = MB N = M B with B B B a two-sided ideal [6 , 7 ] . The present paper replaces equality by an A A A -linear isomorphism, and Example 2.4 shows that the relaxation is strict. Tuganbaev’s results on distributivity, localization and flatness are used only with their original hypotheses [6 , 8 , 9 ] ; our Goldie theorem is formulated for semiprime right Goldie rings and gives a necessary rational obstruction rather than a converse criterion.
The normal form for right modules over T 2 ( k ) T_2(k) T 2 ( k ) is the familiar representation theory of the one-arrow quiver A 2 A_2 A 2 . The assertion specific to this paper is the ideal-multiple recognition in Theorem 3.3 ; the quiver decomposition is not claimed as new. Likewise, Theorem 5.4 uses established Goldie theory and the standard exactness of classical localization as inputs; the iso-multiplication obstruction on its localized multiplicities is the asserted contribution. This distinction does not itself establish that no earlier formulation of the same obstruction exists. The matrix-amplification result uses standard Morita correspondence [3 ] . Alsuraiheed-Bavula [10 ] study direct sums of classical multiplication modules, whereas our examples concern the weaker iso-multiplication property.
Facchini-Nazemian introduced iso-Noetherian and iso-Artinian chain conditions [11 , 12 ] . The implication in Corollary 7.6 uses their published classification under semiprime, right-Noetherian and right-isoartinian hypotheses, with matrix-factor elimination as an additional deduction. The chain conditions are not built into our definition. For commutative locally cyclic projective modules, the work of Ali [1 ] and other classical classifications [2 ] explain why we do not present an extended commutative classification here. The results above use these standard decomposition and equivalence theorems with the hypotheses stated in their original forms.
Conclusion
We have constructed iso-multiplication modules which are not multiplication modules and classified all such modules over T 2 ( k ) T_2(k) T 2 ( k ) . We have also proved that the classical localization of an iso-multiplication module over a semiprime right Goldie ring is multiplicity-free. Further results concern composition length, local rings, Morita equivalence, and chain conditions up to isomorphism. The localization condition is not sufficient in general.
Declaration of generative AI and AI-assisted technologies in the manuscript preparation process.
During the preparation of this manuscript, the authors used ChatGPT (OpenAI) to assist with language editing and the presentation of the manuscript. The authors reviewed and edited the resulting text and take full responsibility for the content of the manuscript.