Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

∗†‡

A generalization of uniserial modules and rings

S. Shirzadia , R. Beyranvanda,§ and A. Moradzadeh-Dehkordib,c


arXiv:2311.10428v1 [math.RA] 17 Nov 2023

a
Department of Mathematics, Faculty of Science, Lorestan University, Khorramabad, Iran
b
Department of Science, Shahreza Campus, University of Isfahan, Iran
c
School of Mathematics, Institute for Research in Fundamental Sciences (IPM)
P.O.Box : 19395-5746, Tehran, Iran
beyranvand.r@lu.ac.ir
a.moradzadeh@shr.ui.ac.ir
shirzadi.sa@fs.lu.ac.ir

Abstract
We introduce and study a nontrivial generalization of uniserial modules and rings. A
module is called weakly uniserial if its submodules are comparable regarding embedding.
Also, a right (resp., left) weakly uniserial ring is a ring which is weakly uniserial as a right
(resp., left) module over itself. In this paper, in addition to providing the properties of weakly
uniserial modules and rings, we show that every right R-module is weakly uniserial if and
only if every 2-generated right R-module is weakly uniserial, if and only if R ∼ = Mn (D) where
D is a division ring. Then it is determined which torsion-free abelian groups of rank 1 are
weakly uniserial. Finally, when R is a commutative principal ideal domain, the structure of
finitely generated weakly uniserial R-modules are completely determined.

1. Introduction
Recall that an R-module M is said to be uniserial if its submodules are linearly ordered by
inclusion. An R-module M is called serial if it is a direct sum of uniserial modules. A left
(resp., right) uniserial ring is a ring which is uniserial as a left (resp., right) module. Also, a
ring R is called uniserial (resp., serial) if it is both a left and a right uniserial (resp., serial) ring.
Studying serial rings and modules has been done since many years ago. Köthe was probably the
pioneer in this field [14]. He proved that over an Artinian principal ideal ring (a special case of
serial rings), every module is a direct sum of cyclic submodules. In addition, Nakayama in [17,
Theorem 17], showed that every left (right) R-module is a serial module provided that R is an
Artinian serial ring. Skornyakov in [18] proved the converse. To observe the major properties of

The research of the third author was in part supported by a grant from IPM (No. 1402160411). This research
is partially carried out in the IPM-Isfahan Branch.

Key Words: Weakly uniserial module, Weakly uniserial ring, Torsion-free abelian group

2020 Mathematics Subject Classification. 16D10, 16D70, 16D60, 16N20, 16S90.
§
Corresponding author.

1
serial rings and modules, we can refer to [19]. Recently, in [7, 8], Behboodi et al., introduced and
studied the notions of virtually uniserial modules and almost uniserial modules as generalizations
of uniserial modules. An R-module M is said to be almost uniserial if any two non-isomorphic
submodules of M are linearly ordered by inclusion. A virtually simple module is a module which
is isomorphic to each its nonzero submodule. Also an R-module M is called virtually uniserial
if for every finitely generated submodule 0 6= N ⊆ M , N/Rad(N ) is virtually simple.
In this paper, we introduce and study the concept of weakly uniserial modules as a nontrivial
common generalization of uniserial modules and almost uniserial modules. Let M be a right
R-module. We say that M is weakly uniserial if for any two submodules N and K of M , N
is embedded in K or K is embedded in N . A left (resp., right) weakly uniserial ring is a ring
which is weakly uniserial as a left (resp., right) R-module. As usual, a ring R is called weakly
uniserial if it is both a left and a right weakly uniserial ring. We note that every uniserial (almost
uniserial) module is weakly uniserial, but the converse is not true in general. For instance, every
vector space with dimension at least 3 over a division ring is a weakly uniserial module, but it is
not almost uniserial.
Among the other basic properties of weakly uniserial modules and rings, we study the fol-
lowing questions:
Question 1: Which rings R have the property that every (2-generated or injective) right R-
module is weakly uniserial?
Question 2: When is a torsion-free abelian group of rank 1 a weakly uniserial group?
Throughout this paper, all rings have identity elements and all modules are unitary right
modules. For a ring R, the Jacobson radical and the right singular ideal of R are denoted by
J(R) and Z(RR ), respectively. For a module M , the socle, the injective hull and the singular
submodule of M are denoted by Soc(M ), E(M ) and Z(M ), respectively. Also we denote the set
of associated primes of M by Ass(M ). For a subset X of R, the right (resp., left) annihilator of
X in R is denoted by r.AnnR (X) (resp., l.AnnR (X)). By K ⊆ M we usually mean that K is a
submodule of M and the notation N ⊆e M means that N is an essential submodule of M . The
cardinal number of a set X is denoted by |X| and for any two R-modules M, N if there exists
an R-monomorphism from M to N , we write M ֌ N otherwise we write M 6֌ N .
The paper is organized as follows. In Section 2, we give some basic properties of weakly
uniserial modules and rings. It is shown that being weakly uniserial is a Morita invariant property
(see Proposition 2.7). Like of uniform modules, every weakly uniserial module such as M has
|Ass(M )| ≤ 1, while the class of uniform modules and the class of weakly uniserial modules
are independent (see Proposition 2.8). In Proposition 2.12, we prove that the class of weakly
uniserial rings is a straightforward generalization of domains. For a commutative principal ideal
domain R with the field of fractions Q, we show that if R is local, then Q is a (weakly) uniserial
R-module and if R is not local, then Q is not a weakly uniserial R-module (see Proposition 2.15).
In particular, Q (the field of rational numbers) is not weakly uniserial as a Z-module. One of the
important results in this section is Theorem 2.22 which says that if M is a weakly uniserial module
such that Soc(M ) is Dedekind-finite, then N ⊆ Soc(M ) or Soc(M ) ⊆ N , for any submodule N
of M . In Section 3, we answer Question 1 and completely determine the structure of rings R

2
for which every (2-generated or injective) right R-module is weakly uniserial (see Theorem 3.5).
In Section 4, we answer Question 2 and prove that if A is a torsion-free abelian group of rank
1, then A is weakly uniserial if and only if type(A) = [(αp )], where αp = 0 for all but a finite
number p and there is at most one p such that αp = ∞ (see Theorem 4.6). In Section 5, it
is completely determined the structure of finitely generated weakly uniserial R-modules, where
R is a commutative principal ideal domain (see Theorem 5.6). In particular, it is shown that
a finitely generated Z-module M is weakly uniserial if and only if M ∼ = Zn or M ∼ = ⊕n Zp or

M = Zp , where p is a prime number and n ≥ 0 is an integer. Also the structure of weakly
n

uniserial Z-modules with nonzero socle is given in Proposition 5.8.

2. Weakly uniserial modules and rings


In this section, we give some basic properties of weakly uniserial modules and rings.

Definitions 2.1. Let M be a right R-module. We say that M is weakly uniserial if for any two
submodules N and K of M , N ֌ K or K ֌ N . A left (resp., right) weakly uniserial ring is a
ring which is weakly uniserial as a left (resp., right) module. As susal, a ring R is called weakly
uniserial if it is both a left and a right weakly uniserial ring.

First of all, the following examples show that the notions of weakly uniserial rings and modules
are nontrivial generalizations of uniserial rings and modules.

Example 2.2. (1) For any prime number p, the Z-module Zp ⊕ Zp is weakly uniserial but is not
uniserial.
(2) Since each nonzero Z-submodule of Z is isomorphic to Z, the ring Z is weakly uniserial
but is not uniserial.

Recall that an R-module M is said to be almost uniserial if any two non-isomorphic sub-
modules of M are linearly ordered by inclusion (see [7]). Clearly uniserial and almost uniserial
right R-modules are always weakly uniserial, but the following example shows that the converse
is not true in general.

Example 2.3. Every vector space with dimension at least 3 over a division ring is a weakly
uniserial module, but it is not almost uniserial.

Recall that an R-module M is said to be virtually simple if M 6= 0 and M ∼ = N for every


nonzero submodule N of M (see [5, 6]). Also, an R-module M is called virtually uniserial if for
every finitely generated submodule 0 6= N ⊆ M , N/Rad(N ) is virtually simple (see [8]). The
following example shows that the class of weakly uniserial modules and the class of virtually
uniserial modules are independent.

Example 2.4. (1) As stated in Example 2.3, every vector space with dimension at least 3 over
a division ring is weakly uniserial. But, it is not virtually uniserial by [8, Proposition 2.4].
(2) By [8, Example 2.1(1)], Q is a virtually uniserial Z-module, but it is not weakly uniserial
(see Corollary 2.16(a)).

3
Consequently, we have the following relationships:

{Uniserial modules} ( {Almost uniserial modules} ( {Weakly uniserial modules},

*
{Weakly uniserial modules} + {Virtually uniserial modules}.

In the following we give some basic properties of weakly uniserial modules and rings.

Remark 2.5. (1) Every submodule of a weakly uniserial module is weakly uniserial.
(2) Every quotient of a weakly uniserial module is not necessarily weakly uniserial. For
instance, Z is a weakly uniserial Z-module, but Z/6Z is not a weakly uniserial Z-module.
(3) The notion of weakly uniserial module is preserved under isomorphism.

Proposition 2.6. For a ring R, the following statements hold:


(a) If R is right (left) weakly uniserial, then R has no nontrivial central idempotent.
Q
(b) If R ∼
= i∈I Ri , where |I| ≥ 2 and each Ri is a ring, then R is neither a right nor a left
weakly uniserial ring.

Proof. (a). Assume that e is a central idempotent of R. Thus R = eR ⊕ (1 − e)R and since R is
right weakly uniserial, eR ֌ (1−e)R or (1−e)R ֌ eR. If eR ֌ (1−e)R, then r.AnnR (1−e)R ⊆
r.AnnR (eR) and so e = 0. If (1 − e)R ֌ eR, then r.AnnR (eR) ⊆ r.AnnR (1 − e)R and so e = 1.
The left is similarly.
(b) follows from (a).

The following result shows that the weakly uniserial property is Morita invariant.

Proposition 2.7. Being weakly uniserial is a Morita invariant property.

Proof. Assume that R and S are Morita equivalent rings via inverse equivalences F : M odR →
M odS , G : M odS → M odR , and let η : F G → 1M odS , ξ : GF → 1M odR be the corre-
sponding natural isomorphisms. Suppose that M is a weakly uniserial right R-module. We
show that F (M ) is a weakly uniserial right S-module. Assume that A and B are submod-
ules of F (M ). Consider the inclusions iA : A → F (M ) and iB : B → F (M ). Then by [1,
G(iA ) G(iB )
Proposition 21.2], G(A) −→ GF (M ) and G(B) −→ GF (M ) are monomorphisms and so
G(iA ) ξ G(iB ) ξ
we have the monomorphisms G(A) −→ GF (M ) −→ M , G(B) −→ GF (M ) −→ M . Thus
G(A) ∼ = ξG(iA )(G(A)) ⊆ M and G(B) ∼ = ξG(iB )(G(B)) ⊆ M and since M is weakly unise-
rial, we assume that h : ξG(iA )(G(A)) → ξG(iB )(G(B)) is a monomorphism. Again, by [1,
Proposition 21.2], we have the following monomorphism from A to B:
η −1 F (h) η
A −→ F G(A) ∼
= F (ξG(iA )(G(A))) −→ F (ξG(iB )(G(B))) ∼
= F (G(B)) −→ B.

4
A right R-module M is called prime if M 6= 0 and AnnR (M ) = AnnR (N ), for any 0 6= N ⊆
M . A two sided ideal P of R is called an associated prime of M if there exists N ⊆ M such that
P = AnnR (N ) and N is a prime R-module. The set of associated primes of M is denoted by
Ass(M ). Also M is called semiprime (weakly prime) if AnnR (N ) is a semiprime (prime) ideal of
R, for any nonzero submodule N of M . For more details see [13].

Proposition 2.8. For a ring R, the following statements hold:


(a) If M is a weakly uniserial right R-module, then |Ass(M )| ≤ 1. Moreover, if R satisfies
in the ascending chain condition on two sided ideals, then |Ass(M )| = 1.
(b) If M is a weakly uniserial right R-module and N ⊆ M is a prime R-module, then
AnnR (N ) is a maximal member in the set of {AnnR (K) | 0 6= K ⊆ M }.
(c) A weakly uniserial right R-module M is weakly prime if and only if it is semiprime.
(d) A right (left) weakly uniserial ring R is prime if and only if it is semiprime.

Proof. (a). Assume that P and Q are distinct associated primes of M . Then P = AnnR (N )
and Q = AnnR (K), for some prime submodules N and K of M . Since M is weakly uniserial
f
we can assume that N ֌ K and so that N ∼ = f (N ) ⊆ K. Since K is prime we conclude
that P = AnnR (N ) = AnnR (f (N )) = AnnR (K) = Q, a contradiction. The second part of this
statement follows from the first part and [16, Lemma 3.58].
(b). Assume that 0 6= K ⊆ M and AnnR (N ) ⊆ AnnR (K). Since M is weakly uniserial, N ֌
f
K or K ֌ N . If N ֌ K, then N ∼ = f (N ) ⊆ K and so AnnR (K) ⊆ AnnR (f (N )) = AnnR (N ).
g
Therefore, in this case AnnR (N ) = AnnR (K). If K ֌ N , then K ∼ = g(K) ⊆ N and since N is
prime we conclude that AnnR (K) = AnnR (g(K)) = AnnR (N ), as desired.
(c). Assume that M is semiprime and there exists N ⊆ M such that N IJ = 0, for some
ideals I and J of R. Note that N (JI)2 = N JIJI ⊆ N IJI = 0 and since M is semiprime we
conclude that N JI = 0. By symmetry, let N I ֌ N J. Then AnnR (N J) ⊆ AnnR (N I) and so
N I 2 = 0. Since M is semiprime, N I = 0, as desired. The converse is clear.
(d). By substituting R for N in the proof of (c), the result is obtained

Example 2.9. Since Z × Z is not a prime ring, by Proposition 2.8(d), it is not a right (left)
weakly uniserial ring.

Corollary 2.10. If R is a semiprime right weakly uniserial ring, then Soc(RR ) = 0 or Z(RR ) =
0 = J(RR ).

Proof. We may assume that Soc(RR ) 6= 0. By [16, Proposition 7.13], Soc(RR ) ∩ Z(RR ) =
0. Thus Soc(RR )Z(RR ) = 0 and so by Proposition 2.8(d), Z(RR ) = 0. On the other hand
Soc(RR )J(RR ) = 0 and again by Proposition 2.8(d), J(RR ) = 0.

Recall that an element a ∈ R is right regular if r.AnnR (a) = 0. Left regular elements are
defined similarly. As usual, a ∈ R is regular if it is both a left and a right regular element.
A ring R is said to be right hereditary if every right ideal of R is projective as a right R-
module. Also, a ring R is said to be local if R has a unique maximal right ideal. The following

5
proposition introduces classes and examples of weakly uniserial rings and modules. First we need
the following lemma.

Lemma 2.11. If F1 and F2 are free right R-modules, then F1 ֌ F2 or F2 ֌ F1 .

Proof. The proof is routine.

Proposition 2.12. For a ring R, the following statements hold:


(a) If every nonzero right ideal of R contains a right regular element, then R is right weakly
uniserial.
(b) Every domain is a right and left weakly uniserial ring.
(c) If R is either a principal right ideal domain or a right hereditary local ring, then every
projective right R-module is weakly uniserial.

Proof. (a). Assume that I and J are two nonzero right ideals of R and b is a right regular
element of J. Then the map f : I → J by f (a) = ba is an R-monomorphism and so R is a right
weakly uniserial ring.
(b) follows from (a).
(c). Let R be a principal right ideal domain and P be a projective right R-module. There
exist a free right R-module F and a right R-module K such that F ∼ = P ⊕ K. By [16, Corollary
2.27], P is a free right R-module. Then it is sufficient to prove that free modules over principal
right ideal domains are weakly uniserial. Let F be a free right R-module and F1 , F2 are two
nonzero submodules of F . By [16, Corollary 2.27], F1 and F2 are free right R-modules and by
Lemma 2.11, F1 ֌ F2 or F2 ֌ F1 . Therefore, F is weakly uniserial.
Let R be a right hereditary local ring and P be a projective right R-module. Assume that
P1 and P2 are nonzero submodules of P . Since R is right hereditary, by [16, Corollary 2.26], P1
and P2 are projective modules and since R is local, P1 and P2 are free right R-modules. Now,
by Lemma 2.11, P1 ֌ P2 or P2 ֌ P1 and so P is a weakly uniserial right R-module.

Remark 2.13. (a) Proposition 2.12(b) shows that the class of weakly uniserial rings is a straight-
forward generalization of domains.
(b) The right hereditary and local conditions are necessary in Proposition 2.12(c). For ex-
ample, Z4 is local but is not a hereditary ring and consider M = Z4 ⊕ Z4 as a Z4 -module. We
show that M is not weakly uniserial. N := (1, 1)Z4 and K := 2Z4 ⊕ 2Z4 are Z4 -submodules
of M that |N | = |K|. Since AnnZ4 (N ) 6= AnnZ4 (K), we conclude that N 6֌ K and K 6֌ N .
Therefore M is not weakly uniserial. Also, Z6 is a hereditary (semisimple) and non-local ring.
Z6 as a Z6 -module is projective, but is not weakly uniserial.

Let R be a commutative principal ideal domain and a, b ∈ R. Then a is said to divide b in


R, denoted by a|b, if b = ac for some c ∈ R. Two elements a and b are associates if b = au,
where u is a unit of R. Also an element p of R is called irreducible (prime) if it is neither zero
nor a unit and if its only divisors are units and associates of p, i.e., for any a, b ∈ R, if p|ab, then
we conclude that p|a or p|b. Elements a and b of R are called relatively prime if 1 is a greatest
common divisor of a and b. In this case we write gcd(a, b) = 1.

6
Example 2.14. (1) For any index set I with |I| ≥ 2, the Z-module ⊕I Z is neither uniserial nor
almost uniserial but since Z is a principal ideal domain by the first part of Proposition 2.12(c),
it is weakly uniserial.
(2) Let Z(p) = {a/b ∈ Q | gcd(a, b) = gcd(b, p) = 1} be the localization of Z at the prime ideal
pZ. It is a hereditary (principal ideal domain) local ring. So by the second part (as well as the
first part) of Proposition 2.12(c), for any index set I the Z(p) -module ⊕I Z(p) is weakly uniserial.
(3) For a nontrivial automorphism σ : F → F of a field F , let R = F [[x; σ]] be a formal skew
power series ring. Then R is a right (and left) hereditary (principal right ideal domain) local
ring and so by the second part (and also the first part) of Proposition 2.12(c), for any index set
I the right R-module ⊕I R is right weakly uniserial.

In the following, we show that if R is a right weakly uniserial ring, then a localization of R
is not necessarily a weakly uniserial right R-module.

Proposition 2.15. Let R be a commutative principal ideal domain and Q be the field of fractions
of R. Then the following statements hold:
(a) If R is local, then Q is a (weakly) uniserial R-module.
(b) If R is not local, then Q is not a weakly uniserial R-module.

Proof. (a). Suppose that R is local with the maximal ideal M. Then M = pR where p is
the unique irreducible element of R (up to multiplication by units). Let I be a nonzero proper
R-submodule of Q. If I ⊆ R, then I = pn R for some n ∈ N ∪ {0}. Now if I * R, then there
exists m ∈ N such that 1/pm ∈ / I, because I 6= Q. We chose the smallest number n such that
n
1/p ∈ / I. Then 1/p n−1 ∈ I and so it is easy to see that I = (1/pn−1 )R. Thus the all of nonzero
proper R-submodules of Q are

· · · ⊆ p2 R ⊆ pR ⊆ R ⊆ (1/p)R ⊆ (1/p2 )R ⊆ · · · .

Therefore Q is uniserial as an R-module.

(b). Suppose that R is not local and p, q are non-associate irreducible elements of R. We claim
that HomR (R(p) , R(q) ) = 0 where

R(p) = { a/b ∈ Q | gcd(a, b) = gcd(b, p) = 1},

R(q) = { a/b ∈ Q | gcd(a, b) = gcd(b, q) = 1}.

Let f ∈ HomR (R(p) , R(q) ) and f (a/b) = c/d 6= 0, for some a/b ∈ R(p) and c/d ∈ R(q) . Since
every commutative principal ideal domain is a unique factorization domain, we can choose the
positive integer n such that q n ∤ c. Now let f (a/(bq n )) = x/y. Then c/d = f (a/b) = f ((aq n )/
(bq n )) = f (a/(bq n ))q n thus c/d = (x/y)q n . Hence cy = dxq n and so c/(dq n ) = x/y. Therefore
f (a/q n b) = c/(q n d) ∈
/ R(q) , a contradiction.

7
Let R be a ring and S be the set of all regular elements in a ring R. The ring R is called right
Ore if S is right permutable, i.e., aS ∩ sR 6= ∅, for any a ∈ R and s ∈ S. Left Ore ring is defined
similarly. As usual, R is called Ore if it is both left and right Ore. It is well-known that R is
right Ore if and only if for any right R-module M , t(M ) = {m ∈ M | ms = 0 for some regular
element s ∈ R} is an R-submodule of M (see [16, Exercise 10.19]). Let M be a right R-module
over a right Ore ring R. Then M is called torsion if t(M ) = M and torsion-free if t(M ) = 0. In
case R is right Ore, M/t(M ) is always torsion-free.

Corollary 2.16. The following statements hold:


(a) Q is not weakly uniserial as a Z-module.
(b) If M is not a torsion Z-module, then M(0) (the localization of M at the prime ideal (0))
is not weakly uniserial as a Z-module.

Proof. (a) follows from Proposition 2.15(b).


(b). Suppose that x ∈ M and nx 6= 0, for any 0 6= n ∈ Z. Then Z ∼ = xZ ⊆ M and so
Q = Z(0) ∼=Q (xZ) (0) ⊆ M (0) . Thus, Q = Z ∼
(0) = (xZ)(0) as a Z-module. Since by (a), QZ is not
weakly uniserial, (xZ)(0) is not weakly uniserial. This implies that M(0) is not weakly uniserial
as a Z-module.

In Section 4, a different proof for Corollary 2.16(a) is given. Recall that a semisimple right
R-module M is said to be homogeneous if every two simple submodules of M are isomorphic.

Remark 2.17. Every uniserial module is uniform, however the class of uniform modules and the
class of weakly uniserial modules are independent. For instance every not simple homogeneous
semisimple module is weakly uniserial, but it is not uniform. Also Q is a uniform Z-module that
is not weakly uniserial by Corollary 2.16(a).

Example 2.18. (1) Every


" homogeneous
# semisimple right R-module is weakly uniserial.
Z
Z 2Z
(2) The ring R = is neither a right nor a left weakly uniserial ring. To see this,
0 Z
" # " #
2Z 0 0 0
consider two ideals I := and J := of R. Since r.AnnR (I) * r.AnnR (J)
0 0 0 2Z
and r.AnnR (J) * r.AnnR (I), we conclude that J 6֌ I and I 6֌ J. Also, note that l.AnnR (I) *
l.AnnR (J) and l.AnnR (J) * l.AnnR (I).  
a 0 b
 
(3) Let D be a division ring and R = { 0 a c  | a, b, c ∈ D}. Then, R is both left
0 0 a
 
0 0 D
 
and right weakly uniserial. Note that the only proper right (left) ideals of R are  0 0 D ,
0 0 0
   
0 0 D 0 0 0
   
 0 0 0  and  0 0 D .
0 0 0 0 0 0

8
Lemma 2.19. A semisimple right R-module is weakly uniserial if and only if it is homogeneous
semisimple.

Proof. (⇒). Set M = ⊕i∈I Si , where Si is a simple right R-module for any i ∈ I. Since M is
weakly uniserial, for any i, j ∈ I, Si ֌ Sj or Sj ֌ Si . In any case, we conclude that Si ∼
= Sj , as
required. The converse is clear.

Recall that a ring R is said to be right (left) duo if every right (left) ideal of R is two-sided.
Also a ring R is called right (left) semi-duo if every maximal right (left) ideal of R is two-sided.

Corollary 2.20. A right semi-duo ring R is local if and only if every semisimple right R-module
is weakly uniserial.

Proof. Assume that every semisimple right R-module is weakly uniserial and M1 and M2 are
two distinct maximal right ideals of R. In right R-module R/M1 ⊕ R/M2 , by Lemma 2.19,
R/M1 ∼ = R/M2 . But since R is right semi-duo, M1 = AnnR (R/M1 ) = AnnR (R/M2 ) = M2
and hence R is local. The converse is clear by Lemma 2.19.

A right R-module M is called semi-Artinian if for every submodule N 6= M , Soc(M/N ) 6= 0.


A ring R is a right semi-Artinian ring, if R is a semi-Artinian right R-module. A ring R is right
semi-Artinian if and only if every nonzero right R-module has an essential socle, if and only if
every nonzero right R-module has a simple submodule.

Proposition 2.21. For a ring R, the following statements hold:


(a) If M is a weakly uniserial right R-module with nonzero socle, then Soc(M ) is a homoge-
neous semisimple essential submodule of M .
(b) If R is a right semi-Artinian, then the socle of any weakly uniserial right R-module M is
a homogeneous semisimple essential submodule of M .

Proof. (a). By Remark 2.5(1) and Lemma 2.19, Soc(M ) is homogeneous semisimple. Now,
suppose that S is a simple submodule of M . Then by hypothesis, for every nonzero submodule
N of M , N ֌ S or S ֌ N . In any case, N contains a simple submodule and so Soc(M ) ⊆e M .
(b) follows from (a).

The uniform dimension of a right R-module M (denoted by u.dim(M )) is the supremum of


the set {k | M contains a direct sum of k nonzero submodules}. Also M is called Dedekind-finite
if for every right R-module N , the relation M ∼
= M ⊕ N implies that N = 0. It is easy to check
that any right R-module with finite uniform dimension is Dedekind-finite.

Theorem 2.22. For a ring R, the following statements hold:


(a) If M is a weakly uniserial right R-module such that Soc(M ) is Dedekind-finite, then
N ⊆ Soc(M ) or Soc(M ) ⊆ N , for any submodule N of M . Moreover, Soc(M ) = 0 or M
is indecomposable or M is homogeneous semisimple.
(b) If R is a right weakly uniserial ring such that Soc(RR ) is Dedekind-finite, then J(R)2 = 0
or Soc2 (RR ) = 0. Moreover, Soc(RR ) = 0 or RR is indecomposable or R ∼ = Mn (D)
where D is a division ring.

9
Proof. (a). Assume that N is a nonzero submodule of M . Since M is weakly uniserial, N ֌
Soc(M ) or Soc(M ) ֌ N . If N ֌ Soc(M ), then N is semisimple and so N ⊆ Soc(M ). If
f
Soc(M ) ֌ N , then Soc(M ) ∼
= f (Soc(M )) ⊆ N and so there exists a submodule K of M such
that f (Soc(M )) ⊕ K = Soc(M ). Thus Soc(M ) ⊕ K ∼
= Soc(M ) and since Soc(M ) is Dedekind-
finite we conclude that K = 0. Therefore, Soc(M ) = f (Soc(M )) ⊆ N .
For the second part of this statement, assume that M is not indecomposable. Then there
exist two nonzero submodules M1 and M2 of M such that M = M1 ⊕ M2 . If Soc(M ) = 0, we
are done but if Soc(M ) 6= 0, then by Proposition 2.21(a), Soc(M ) ⊆e M and so by the first part
of the proof we conclude that Mi ⊆ Soc(M ), for i = 1, 2. Then M = M1 ⊕ M2 ⊆ Soc(M ) and so
by Lemma 2.19, M is homogeneous semisimple.
(b). By (a), Soc(RR ) ⊆ J(R) or J(R) ⊆ Soc(RR ). On the other hand, Soc(RR )J(R) = 0 and
so J(R) ⊆ Soc(RR ) implies that J(R)2 = 0 and Soc(RR ) ⊆ J(R) implies that Soc2 (RR ) = 0. The
second part follows from (a).

Artinian principal ideal rings were studied in papers of G. Köthe and K. Asano, where it
was proved that any Artinian principal right ideal ring is right uniserial (see [3, 4]). In fact,
K. Asano proved that an Artinian ring is uniserial if and only if each ideal is a principal right
ideal and a principal left ideal. In the following result we give the structure of some right weakly
uniserial rings such as the right Artinian principal right ideal rings. A ring R is called left perfect
if R/J(R) is semisimple and J(R) is left T-nilpotent.

Corollary 2.23. For a ring R, the following statements hold:


(a) Let R be a left perfect ring such that Soc(RR ) is Dedekind-finite. If R is right weakly
uniserial, then R is local or R ∼
= Mn (D) where D is a division ring.
(b) If R is a right Artinian right weakly uniserial ring, then R is local or R ∼
= Mn (D) where
D is a division ring.
(c) If R is a right Artinian principal right ideal ring, then R is right weakly uniserial if and
only if R is right uniserial or R ∼
= Mn (D) where D is a division ring.

Proof. (a). Since R is left perfect, by [15, Theorem 23.20], Soc(RR ) 6= 0. So by Theorem 2.22(b),
if R ≇ Mn (D), then RR is indecomposable and so R has no non-trivial idempotents. Therefore
by [15, Corollary 21.29], R is a local ring.
(b) follows from (a).
(c). Assume that R is right weakly uniserial. Since R is right Artinian, u.dim(Soc(RR )) < ∞
and so Soc(RR ) is Dedekind-finite. Hence by (a) if R ≇ Mn (D), then R is a local ring and so by
[8, Lemma 2.13], R is a right uniserial ring. Conversely, suppose that R ∼ = Mn (D) where D is
a division ring. Then RR is homogeneous semisimple and so by Example 2.18(1), RR is weakly
uniserial. Also clearly right uniserial rings are right weakly uniserial.
" #
D D
Example 2.24. Let D be a division ring and R = . Clearly R is a right (and left)
0 D
" # " #
D D 0 D
Artinian ring. Since and are two maximal right (and left) ideals of R,
0 0 0 D

10
" #
0 D
then R is not a local ring. Also J(Mn (D)) = 0 and J(R) = 6= 0, hence R ≇ Mn (D).
0 0
Therefore by Corollary 2.23(b), R is neither a right nor a left weakly uniserial ring.

A ring R is called right Kasch if every simple right R-module can be emmbedded in RR . Left
Kasch ring is defined similarly. As usual, R is called a Kasch ring if it is both right and left
Kasch.

Proposition 2.25. For a ring R, the following statements hold:


(a) If R is a commutative (semi) prime weakly uniserial ring, then R is a field or Soc(R) = 0.
(b) Every commutative Kasch weakly uniserial ring is local.
(c) Every commutative Artinian weakly uniserial ring is local.

Proof. (a). Assume that Soc(R) 6= 0. Then there exists a nonzero minimal ideal I of R. By
[15, Brauer’s Lemma 10.22], I 2 = 0 or I = eR, for some idempotent e ∈ R. If I 2 = 0, then since
R is semiprime we conclude that I = 0, a contradiction. Hence I = eR, but since R is weakly
uniserial, we have e = 0 or e = 1. If e = 0, then I = 0, a contradiction. Thus e = 1, and so
Soc(R) = R. Now by Lemma 2.19, RR is homogeneous semisimple and so R ∼ = Mn (D) where D
is a division ring. Now, since R is commutative we conclude that R is a field.
(b). Suppose that R is a commutative weakly uniserial Kasch ring and M1 , M2 are maximal
ideals of it. Since R is Kasch, R/M1 and R/M2 are embedded in R and since R is weakly
uniserial, R/M1 ֌ R/M2 or R/M2 ֌ R/M1 . In both cases we have R/M1 ∼ = R/M2 and so
M1 = AnnR (R/M1 ) = AnnR (R/M2 ) = M2 , as desired.
(c) follows from Corollary 2.23(b).

Let N be a submodule of a right R-module M . A submodule C ⊆ M is said to be a


complement to N (in M ) if C is maximal with respect to the property that C ∩ N = 0. A
submodule C of a right R-module M is called a complement (in M ), if there exists a submodule
N of M such that C is a complement to N (in M ). A right R-module M is called CS, if every
complement in M is a direct summand of M . Also M is called cocyclic if it has an essential
simple submodule. Note that CS modules are not necessarily weakly uniserial. For instance QZ
is a CS module that is not weakly uniserial (see Corollary 2.16(a)).

Proposition 2.26. If M is a weakly uniserial CS right R-module such that Soc(M ) is Dedekind-
finite, then Soc(M ) = 0 or M is cocyclic or M is homogeneous semisimple.

Proof. Assume that Soc(M ) 6= 0. If Soc(M ) = S where S is a simple submodule of M , then by


Proposition 2.21(a), S ⊆e M and so M is cocyclic. Now suppose that Soc(M ) = ⊕I Si , where
|I| ≥ 2 and every Si is a simple submodule of M . First, we show that every Si is a complement
in M . Let t ∈ I and N be a submodule of M such that St ⊆ N and N ∩ (⊕t6=i∈I Si ) = 0.
Since Soc(M ) * N , by Theorem 2.22(a), we conclude that N ⊆ Soc(M ). Suppose that x ∈ N .
P
Then x = xt + nj=1 xij , where xij ∈ Sij , xt ∈ St and ij ∈ I \ {t}, for j = 1, . . . , n. Thus
Pn
x − xt = j=1 xij and so x = xt . It follows that N = St and hence St is a complement in M .
Since M is a CS module, there exists a nonzero submodule L of M such that St ⊕ L = M . So, M

11
is decomposable and Soc(M ) 6= 0. Therefore, Theorem 2.22(a) implies that M is a homogeneous
semisimple right R-module.

Corollary 2.27. Every weakly uniserial uniform right R-module with nonzero socle is cocyclic.

3 . Rings over which every module is weakly uniserial


In this section, we characterize rings over which every (fnitely generated or injective) module
is weakly uniserial. A ring is called quasi-Frobenius (briefly, QF ) if it is right Noetherian and
right self-injective. The following example shows that projective and injective modules are not
necessarily weakly uniserial.

Example 3.1. Let R = Zp2 and M = R⊕ R, where p is a prime number. Since M is a projective
R-module and R is QF, M is an injective R-module. Now, N := (1, 1)R and K := pR ⊕ pR
are two R-submodules of M that |N | = |K|. But since AnnR (N ) 6= AnnR (K), we conclude that
N 6֌ K and K 6֌ N . Therefore, M is not a weakly uniserial R-module.

In the following we show that cyclic modules are not necessarily weakly uniserial.

Example 3.2. (1) As stated in Proposition 2.6(a), a ring with nontrivial central idempotents is
not weakly uniserial. For instance consider the ring Z/6Z.
(2) Let K be a field and R be the K-algebra with generators x and y such that x3 = y 3 =
xy = 0 (i.e., R ∼= K[x, y]/hx3 , y 3 , xyi). R is an Artinian local ring with the maximal ideal
M = xR ⊕ yR. Also, Spec(R) = {M}. If xR ֌ yR, then AnnR (yR) ⊆ AnnR (xR) and so
x2 = 0, a contradiction. If yR ֌ xR, then similarly y 2 = 0, a contradiction. Therefore neither
M is weakly uniserial, nor RR .

The following lemma plays a key role in the subsequent results.

Lemma 3.3. For a ring R that every 2-generated right R-module is weakly uniserial, the fol-
lowing statements hold:
(a) All simple right R-modules are isomorphic.
(b) R is a right semi-Artinian right Kasch ring.
(c) R has only one prime ideal.
(d) All two-sided ideals are comparable.

Proof. (a). Set M = S1 ⊕ S2 where S1 and S2 are simple right R-modules. Since M is weakly
uniserial, by Lemma 2.19, S1 ∼
= S2 .
(b). Set M = (R/M) ⊕ C where M is a maximal right ideal of R and C is a cyclic right
R-module. Since M is weakly uniserial, R/M ֌ C or C ֌ R/M. In any case Soc(C) 6= 0 and
so every nonzero right R-module has a simple submodule. Therefore, R is a right semi-Artinian
ring. Also, if we set M = R ⊕ S where S is a simple right R-module, then since M is weakly
uniserial, R ֌ S or S ֌ R. In any case, S is embedded in R and so R is a right Kasch ring.

12
(c). Assume that P1 and P2 are distinct prime ideals of R and set M = (R/P1 ) ⊕ (R/P2 ).
Then Ass(M ) = Ass(R/P1 ) ∪ Ass(R/P2 ) = {P1 , P2 }. Thus by Proposition 2.8(a), M is not
weakly uniserial, a contradiction.
(d). Set M = R/I ⊕ R/J where I and J are two-sided ideals of R. Since M is weakly
uniserial, R/I ֌ R/J or R/J ֌ R/I. Therefore, J = r.AnnR (R/J) ⊆ r.AnnR (R/I) = I or
I = r.AnnR (R/I) ⊆ r.AnnR (R/J) = J.

Lemma 3.4. Let R be a ring that all simple right R-modules are isomorphic. Then R is homo-
geneous semisimple or Soc2 (RR ) = 0.

Proof. We may assume Soc(RR ) 6= 0. It is well-known that every simple right R-module is
either singular or projective, but not both. By hypothesis, all simple right R-modules are either
singular or projective. If all simple right R-modules are projective, then it is easy to see that
R is semisimple and by hypothesis, R is homogeneous semisimple (see [9, p.63 Theorem 1.1]).
Thus assume that all simple right R-modules are singular. Then any simple right ideal I of R is
singular and by [16, Lemma 7.2], I(Soc(RR )) = 0 and hence Soc2 (RR ) = 0.

Let µ be an ordinal and A = (Aα | α ≤ µ) be a sequence of modules. A is called continuous


S
chain provided that A0 = 0, Aα ⊆ Aα+1 for all α < µ and Aα = β<α Aβ for all limit ordinals
α ≤ µ. Let M be a module and C be a class of modules. M is C–filtered, if there are an ordinal
κ and a continuous chain of modules, (Mα | α ≤ κ), consisting of submodules of M such that
M = Mκ , and each of the modules Mα+1 /Mα (α < κ) is isomorphic to an element of C. In this
case the chain (Mα | α ≤ κ) is called a C–filtration of M . If R is a ring, M is a semi-Artinian
right R-module and C is the class of all simple right R-modules, then it is easy to see that M is
C–filtered. Also, a ring R is called right V -ring if every simple right R-module is injective. The
following theorem answers Question 1.

Theorem 3.5. For a ring R, the following statements are equivalent:


(a) R ∼
= Mn (D) where D is a division ring.
(b) Every right R-module is weakly uniserial.
(c) Every finitely generated right R-module is weakly uniserial.
(d) Every 2-generated right R-module is weakly uniserial.
(e) Every injective right R-module is weakly uniserial.
(f) The left-right symmetric of (b), (c), (d) and (e).

Proof. (a) ⇒ (b) ⇒ (c) ⇒ (d) are clear.


(d) ⇒ (a). Assume that every 2-generated right R-module is weakly uniserial. By Lemma 3.3(a),
there is only one (up to isomorphic) simple right R-module, say S. We claim that every short
f g
exact sequence of the form 0 −→ S −→ X −→ S −→ 0 splits where X is a right R-module.
For if not, it is shown that X is a cyclic module and X ⊕ S is not weakly uniserial, which
contradicts (d). First we show that X is a module of length 2. By exactness of above short exact
sequence, we have Imf both a maximal and minimal submodule of X. Then 0 ( Imf ( X is a
composition series for X and so l(X) = 2 (the length of X).
Suppose that X1 and X2 are nonzero proper submodules of X. We show that X1 ∩ X2 6= 0. For

13
if not, X1 ∩ X2 = 0 and so 0 ( X1 ( X1 ⊕ X2 ⊆ X. Since l(X) = 2, we have X = X1 ⊕ X2
and there exists a maximal submodule T of X such that 0 ( X1 ⊆ T ( X. Hence T = X1 and
X/X1 ∼ = X2 implies that X2 is a simple submodule of X and so X2 ∼= S. Similarly, X1 ∼
= S and

so X = X1 ⊕X2 = S ⊕S, a contradiction. Thus X1 ∩X2 6= 0 and we have 0 ( X1 ∩X2 ⊆ X1 ( X.
But since l(X) = 2, X1 ∩ X2 = X1 and so X1 ⊆ X2 . Similarly, X2 ⊆ X1 and therefore X1 = X2 .
This clearly shows that X is cyclic.
Now to show that X ⊕ S is not weakly uniserial, we consider two submodules X and N ⊕ S of
X ⊕ S where N is a maximal (minimal) submodule of X and proceed by cases:
ϕ
Case 1: N ⊕ S ֌ X. Then N ⊕ S ∼ = ϕ(N ⊕ S) = ϕ(N ) ⊕ ϕ(S) ⊆ X. Hence 0 ( ϕ(S) ( ϕ(N ) ⊕
ϕ(S) ⊆ X and so we have ϕ(N )⊕ϕ(S) = X, since l(X) = 2. Therefore X = ϕ(N )⊕ϕ(S) ∼ = S⊕S,
a contradiction.
ψ
Case 2: X ֌ N ⊕ S. Then X ∼ = ψ(X) ⊆ N ⊕ S. But l(N ) = 1 and so N ∼ = S. Thus X is a

semisimple module of length 2, i.e., X = S ⊕ S, a contradiction.
Therefore, every short exact sequence of the form 0 −→ S −→ X −→ S −→ 0 splits. This
implies that Ext1R (S, S) = 0. Now assume that M is a right R-module. By Lemma 3.3(b), M
is semi-Artinian and so it has an S-filtration, that is, there exist an ordinal κ and an increasing
continuous chain of submodules 0 = M0 ⊆ M1 ⊆ M2 ⊆ · · · ⊆ Mκ = M such that Mi+1 /
Mi ∼ = S for every i < κ. Thus Ext1R (Mi+1 /Mi , S) = 0 and so, by Eklof’s Lemma [11, Lemma
3.1.2], Ext1R (M, S) = 0. Hence, by [16, Lemma 5.49], S is an injective right R-module and so
R is a right V -ring. Now by [12, Theorem 6.2], Soc2 (RR ) = Soc(RR ). Since Soc(RR ) 6= 0, by
Lemma 3.4, R is homogeneous semisimple and hence R ∼ = Mn (D) where D is a division ring.
(d) ⇔ (e) is clear, since every module is a submodule of its injective hull.

Corollary 3.6. For a right semi-duo ring R, the following statements are equivalent:
(a) R is a division ring.
(b) Every finitely generated free right R-module is weakly uniserial and Soc(RR ) is a minimal
ideal of R.
(c) R ⊕ R is a weakly uniserial right R-module and Soc(RR ) is a minimal ideal of R.
(d) Every (2-generated) right R-module is weakly uniserial.

Proof. (a) ⇒ (b) ⇒ (c) and (a) ⇒ (d) are clear.


(c) ⇒ (a). Let K = (0)⊕R and N = I ⊕I be submodules of R⊕R where I is the minimal right
ideal of R (i.e., I = Soc(RR )). Since R ⊕ R is weakly uniserial, N ֌ K or K ֌ N . In the first
f
case, I ⊕I ֌ R and so f (I)⊕f (I) = f (I ⊕I) = f (Soc(I ⊕I)) ⊆ Soc(RR ), a contradiction. Thus,
R ֌ I ⊕ I and so r.AnnR (I) = r.AnnR (I) ∩ r.AnnR (I) = r.AnnR (I ⊕ I) ⊆ r.AnnR (R) = 0. But
I∼= R/M where M is a maximal right ideal of R and since R is right semi-duo, M = r.AnnR (R/
M) = r.AnnR (I) = 0. This implies that R is a division ring.
(d) ⇒ (a). Assume that every (2-generated) right R-module is weakly uniserial. By Theorem 3.5,

R = Mn (D) where D is a division ring. On the other hand since R is right semi-duo, similar to
the proof of Corollary 2.20, R is local. Therefore, R is a division ring.

Recall that a ring R is called right p.p-ring if each principal right ideal of R is projective.

14
Proposition 3.7. For a ring R, the following statements are equivalent:
(a) R ∼= Mn (D) where D is a division ring.
(b) R is a right p.p-ring such that R ⊕ S is a weakly uniserial right R-module for any simple
right R-module S.
(c) Soc(RR ) is nonzero Dedekind-finite and R ⊕ R is a weakly uniserial right R-module.

Proof. (a) ⇒ (b) is clear.


(b) ⇒ (a). Assume that R is a right p.p-ring and S is a simple right R-module. By hypothesis,
f
R ֌ S or S ֌ R. In the first case R is a simple right R-module and we are done. If S ֌ R,
then S ∼ = f (S) ⊆ R and so f (S) is a principal right ideal of R and S must be projective. Now,
it is easy to see that R is a semisimple ring and since it is right weakly uniserial, Lemma 2.19
implies that R ∼ = Mn (D) where D is a division ring.
(a) ⇒ (c). Follows from Theorem 3.5.
(c) ⇒ (a). Let R be a ring such that Soc(RR ) is nonzero Dedekind-finite and set M = R ⊕ R.
We consider two submodules (0) ⊕ R and I ⊕ I of M , where I is a minimal right ideal of R. Since
M is weakly uniserial, (0) ⊕ R ֌ I ⊕ I or I ⊕ I ֌ (0) ⊕ R. In the first case R is semisimple. In
f
the latter case I ⊕ I ֌ R and so I ⊕ I ∼= f (I ⊕ I) = f (I) ⊕ f (I) ⊆ R. Since f is monomorphism,
f (I) is a minimal right ideal of R. Set I1 = f (I) and again since M is weakly uniserial,
(0) ⊕ R ֌ (I1 ⊕ I1 ) ⊕ I1 or (I1 ⊕ I1 ) ⊕ I1 ֌ (0) ⊕ R. In the first case R is semisimple. In the
f
second case (I1 ⊕ I1 ) ⊕ I1 ֌ R and so (I1 ⊕ I1 ) ⊕ I1 ∼
= f (I1 ⊕ I1 ⊕ I1 ) = f (I1 ) ⊕ f (I1 ) ⊕ f (I1 ) ⊆ R.
If we set I2 = f (I1 ), then R contains I2 ⊕ I2 ⊕ I2 , where I2 is a minimal right ideal of R.
By continuing this process if R is not semisimple, then ⊕∞ i=1 Ii ⊆ Soc(RR ) where every Ii is
a minimal right ideal of R and Ii = Ij for every i, j ≥ 1. Since ⊕∞ ∼ ∞
i=1 Ii = ⊕i=2 Ii ⊕ I1 and
⊕∞ ∼ ∞ ∞
i=1 Ii = ⊕i=2 Ii , we conclude that ⊕i=1 Ii is not Dedekind-finite. On the other hand it is easy
to see that every direct summand of Dedekind-finite modules is also Dedekind-finite. Now since
Soc(RR ) is semisimple and ⊕∞ ∞
i=1 Ii is a direct summand of it, we conclude that ⊕i=1 Ii is Dedekind-
finite, a contradiction. Therefore, R is a semisimple ring and since R is right weakly uniserial,
by Lemma 2.19, R ∼ = Mn (D) where D is a division ring.

It is easy to see that every cyclic right R-module is uniserial if and only if R is a right uniserial
ring. But this is not necessarily true for weakly uniserial modules. For example Z is a weakly
uniserial ring, but Z6 is a cyclic Z-module that is not weakly uniserial. This fact and above
results motivate the following question:

Question 3.8. Which rings R have the property that every cyclic right R-module is weakly
uniserial?

4. Torsion-free abelian groups of rank 1 which are weakly uniserial


In this section, we answer Question 2. In fact, it is determined which torsion-free abelian groups
of rank 1 are weakly uniserial. For this purpose, we need some definitions and results that will be
presented below. A height sequence (αp )p∈π is a sequence of non-negative integers together with

15
∞, indexed by the elements of π (the set of primes of Z). Let A be a torsion-free abelian group,
a ∈ A and p is a prime number. Recall that p-height a in A, denoted by hp (a), is a non-negative
integer n with a ∈ pn A\pn+1 A and ∞ if no such n exists. Also hA (a) (the height sequence a in
A) is the height sequence (hp (a))p∈π . For simplicity, we write (αp ) instead of the height sequence
(αp )p∈π . Two height sequences (αp ) and (βp ) are equivalent if αp = βp for all but a finite number
p and αp = βp if either αp = ∞ or βp = ∞. It is easy to see that this relation is an equivalence
relation. An equivalence class τ of height sequences is called a type, written τ = [α] for some
height sequence α. Now if A is a torsion-free abelian group and a ∈ A, then we define the type
a in A, to be typeA (a) = [hA (a)]. If any two nonzero elements of A have the same type, then
the common value being denoted by type(A). In this case A is called homogeneous, (for more
details see [2]).

Remark 4.1. Every subgroup of Q is homogeneous. To see this, let A ≤ Q and a, b ∈ A. Then
there exist m, n ∈ Z such that ma = nb. It is easy to see that hp (ma) = hp (a) + hp (m) and
hp (nb) = hp (b) + hp (n). On the other hand, hp (m) = hp (n) = 0 for all but a finite number
p ∈ π. Thus typeA (a) = [hA (a)] = [(hp (a))p ] = [(hp (a) + hp (m))p ] = [(hp (ma))p ] = [(hp (nb))p ] =
[(hp (b) + hp (n))p ] = [(hp (b))p ] = [hA (b)] = typeA (b).

Example 4.2. (1) type(Z) = [hZ (1)] = [(αp )], where αp = 0 for each prime number p.
(2) type(Q) = [hQ (1)] = [(βp )], where βp = ∞ for each prime number p.
(3) For A = h 21n , 31m | n, m ∈ Ni, the subgroup of Q generated by 21n and 31m where n, m ∈ N,
type(A) = [hA (1)] = [(∞, ∞, 0, 0, 0, . . .)].
(4) For the subgroup A = h 12 , 13 , 51 , . . .i of Q, we have type(A) = [hA (1)] = [(1, 1, 1, . . .)].

The set of height sequences has a partial ordering given by (αp ) ≤ (βp ) if αp ≤ βp for each
p ∈ π. Now let type(A) = [(αp )] and type(B) = [(βp )], where A and B are homogeneous
torsion-free abelian groups. Then we define type(A) ≤ type(B) if there exist (α′p ) ∈ [(αp )]
and (βp′ ) ∈ [(βp )] such that (α′p ) ≤ (βp′ ). It is easy to see that if type(A) ≤ type(B) and
type(B) ≤ type(A), then type(A) = type(B). Let A be a torsion-free abelian group. Recall that
the rank of A, denoted by rank(A), is defined dimQ (A ⊗Z Q).

Remark 4.3. A torsion-free abelian group A has rank 1 if and only if A is isomorphic to a
subgroup of Q. For if rank(A) = 1, then A ⊗Z Q ∼=Q Q and since A is flat, A is isomorphic to
a subgroup of Q. On the other hand, if A is a subgroup of Q, then since QZ is flat, A ⊗Z Q is
embedded in Q ⊗Z Q ∼ =Q Q. Thus dimQ (A ⊗Z Q) = 1 and so rank(A) = 1.

Lemma 4.4. (see [2, Theorem 1.1]) Let A and B be torsion-free abelian groups of rank 1. Then
A and B are isomorphic if and only if type(A) = type(B).

We note that if α = (αp )p∈π is a height sequence, then there exists a torsion-free abelian
group A of rank 1 with type(A) = [α], say A = h p1n | p ∈ π, n = αp when αp 6= ∞ and n ∈
N when αp = ∞i. Clearly type(A) = [hA (1)] = [α].

Lemma 4.5. (see [2, Proposition 1.2]) Let A and B be torsion-free abelian groups of rank 1.
Then the following statements are equivalent:

16
(a) HomZ (A, B) 6= 0.
(b) HomZ (A, B) contains a monomorphism.
(c) type(A) ≤ type(B).

In the following theorem, we determine which torsion-free abelian groups of rank 1 are weakly
uniserial.

Theorem 4.6. Let A be a torsion-free abelian group of rank 1. Then A is weakly uniserial if
and only if type(A) = [(αp )], where αp = 0 for all but a finite number p and there is at most one
p such that αp = ∞.

Proof. First assume that A is weakly uniserial. By Remark 4.3, we may assume that A is a
subgroup of Q and so, by Remark 4.1, we can consider the following cases:
Case 1: type(A) = [(0, 0, 0, . . . , 0, m1 , m2 , m3 , . . .)], where n ∈ N and mi 6= 0 for any i ≥ 1.
| {z }
n−times
Consider the following two height sequences:
α = (0, 0, 0, . . . , 0, m1 , 0, m3 , 0, m5 , . . .), β = (0, 0, 0, . . . , 0, 0, m2 , 0, m4 , 0, . . .).
| {z } | {z }
n−times n−times

It is easy to see that [α]  [β] and [β]  [α]. Thus if G1 and G2 are subgroups of A, such that
type(G1 ) = [α] and type(G2 ) = [β], then by Lemma 4.5, HomZ (G1 , G2 ) = 0 = HomZ (G2 , G1 ).
Therefore A is not weakly uniserial, a contradiction.
Case 2: type(A) = [(m1 , m2 , . . . , mn , 0, 0, 0, . . .)], where n ∈ N. If there exist 1 ≤ i, j ≤ n with
i < j and mi = mj = ∞, then we consider the following height sequences:
α = (m1 , m2 . . . , mi−1 , ∞, mi+1 , . . . , mj−1 , 0, mj+1 , . . . , mn , 0, 0, 0, . . .),
β = (m1 , m2 . . . , mi−1 , 0, mi+1 , . . . , mj−1 , ∞, mj+1 , . . . , mn , 0, 0, 0, . . .).
Clearly [α]  [β] and [β]  [α]. Now suppose that G1 and G2 are subgroups of A such that
type(G1 ) = [α] and type(G2 ) = [β]. Then by Lemma 4.5, HomZ (G1 , G2 ) = 0 = HomZ (G2 , G1 )
and so A is not weakly uniserial, a contradiction. Therefore, at most one of the mi ’s is equal to
∞, as desired.
Conversely, suppose that type(A) = [(m1 , m2 , . . . , mn , 0, 0, 0, . . .)], where n ∈ N, mi ∈ N ∪ {∞}
and at most one of the mi ’s is equal to ∞. We show that A is weakly uniserial. Let G1
and G2 be two subgroups of A. Clearly, HomZ (G1 , A) 6= 0 and HomZ (G2 , A) 6= 0. Thus by
Lemma 4.5, type(G1 ) ≤ type(A) and type(G2 ) ≤ type(A) and we may assume type(G1 ) =
[(k1 , k2 , . . . , kn , 0, 0, 0, . . .)] and type(G2 ) = [(l1 , l2 , . . . , ln , 0, 0, 0, . . .)], where ki ≤ mi and li ≤ mi
for 1 ≤ i ≤ n. Consider the following cases:
Case 1: mi ∈ N, for 1 ≤ i ≤ n. Then it is clear that type(G1 ) = type(G2 ) and so by Lemma 4.4,
G1 ∼= G2 .
Case 2: Without loss of generality, we may assume that m1 = ∞. In this case, either k1 =
l1 = ∞ or k1 , l1 ∈ N or only one of them is equal to ∞. Therefore either type(G1 ) = type(G2 )
or type(G1 ) ≤ type(G2 ) or type(G2 ) ≤ type(G1 ). Then by Lemma 4.4 and Lemma 4.5, either
G1 ∼= G2 or one of them is embedded in the other.

17
Now, by Theorem 4.6, we have the following example.

Example 4.7. (1) QZ is not weakly uniserial, because type(Q) = [(∞, ∞, ∞, . . .)], see also
Corollary 2.16.
(2) ZZ is weakly uniserial, because type(Z) = [(0, 0, 0, . . .)].
(3) If A = h 21n , 13 , 15 | n ∈ Ni, then type(A) = [(∞, 1, 1, 0, 0, 0, . . .)]. Thus A is a weakly
uniserial group.
(4) If A = h 21n , 31m , 51 | n, m ∈ Ni, then type(A) = [(∞, ∞, 1, 0, 0, 0, . . .)]. Thus A is not
weakly uniserial.
(5) If A = h 21 , 13 , 15 , . . .i, then type(A) = [(1, 1, 1, . . .)]. Thus A is not weakly uniserial.

5. Finitely generated weakly uniserial modules over commutative


principal ideal domains
In this section, we completely determine the structure of finitely generated weakly uniserial R-
modules, where R is a commutative principal ideal domain. In particular, it is shown that a
finitely generated Z-module M is weakly uniserial if and only if M ∼ = Zn or M ∼ = ⊕n Zp or

M = Zpn , where p is a prime number and n ≥ 0 is an integer. Also the structure of weakly
uniserial Z-modules with nonzero socle is given.
Let x be an element of a Z-module M . The order of x is the smallest integer n such that nx = 0,
if no such n exists, the order x is called infinite. The order of x is denoted by ord(x).

Proposition 5.1. Let M be a finite right R-module and |M | = pα1 1 . . . pαt t , where t ≥ 2, αi ∈ N
and pi ’s are distinct prime numbers for each 1 ≤ i ≤ t. Then M is not weakly uniserial.

Proof. By fundamental theorem of finite abelian groups we have M = M1 ⊕ · · · ⊕ Mt , where


|Mi | = pαi i , αi ∈ N and pi ’s are distinct prime numbers for each 1 ≤ i ≤ t. First we show that
every Mi is a right R-submodule of M . Assume that s ∈ {1, . . . , t} and y ∈ Ms . Then for
every r ∈ R we have yr = y1 + · · · + ys + · · · + yt , where yi ∈ Mi for each 1 ≤ i ≤ t. Hence
pαs s yr = pαs s y1 + · · · + psαs ys + · · · + pαs s yt and so pαs s y1 + · · · + pαs s ys−1 + pαs s ys+1 + · · · + pαs s yt = 0.
It follows that pαs s y1 = · · · = pαs s ys−1 = pαs s ys+1 = · · · = pαs s yt = 0. On the other hand for
every 0 6= yi , we have ord(yi ) = pβi i where 1 ≤ βi ≤ αi . Thus for every i 6= s, we conclude that
pβi i | pαs s and so pi | ps , a contradiction. Therefore for every i 6= s, yi = 0 and so yr = ys ∈ Ms .
This means that Ms is an R-submodule of M . Now, if M is a weakly uniserial right R-module,
then Mi ֌ Mj or Mj ֌ Mi , where i 6= j and 1 ≤ i, j ≤ t. Therefore, pi | pj or pj | pi , a
contradiction.

Corollary 5.2. If M is a finite weakly uniserial right R-module, then |M | = pα where p is a


prime number and α ≥ 0 is an integer.

Remark 5.3. The converse of Corollary 5.2 is not necessarily true. For example suppose that
M ∼= Zpα1 ⊕ Zpα2 ⊕ · · · ⊕ Zpαt as Z-modules, where p is a prime number and αi ∈ N, for
each 1 ≤ i ≤ t. Assume that there exists s ∈ {1, . . . , t} such that αs ≥ 2. Then M1 =

18
(0) ⊕ · · · ⊕ (0) ⊕ Zpαs ⊕ (0) ⊕ · · · ⊕ (0) and M2 = pα1 −1 Zpα1 ⊕ (0) ⊕ · · · ⊕ (0) ⊕ pZpαs ⊕ (0) ⊕ · · · ⊕ (0)
are submodules of M that |M1 | = |M2 |, but since M1 is cyclic and M2 is not cyclic we conclude
that M1 ≇ M2 . Therefore, M1 6֌ M2 and M2 6֌ M1 .

It is well-known that being uniserial is not a symmetric property. For example if R is the left
skew polynomial ring K[x; σ]/(x2 ), where K is a field and σ : K → K is a ring homomorphism
which is not an automorphism, then R is a left uniserial ring that is not right uniserial. In [10,
Lemma 1], W. E. Clark and D. A. Drake showed that if R is a finite ring, then R is right uniserial
if and only if it is left uniserial. We could not show that being weakly uniserial is a symmetric
property, but the following result shows that under certain conditions a right weakly uniserial
ring is also a left weakly uniserial ring.

Proposition 5.4. Let R be a finite ring with a principal maximal right ideal. If R is a right
weakly uniserial ring, then either R ∼
= Mn (D) where D is a field or R is left and right uniserial.

Proof. Since R is a right Artinian right weakly uniserial ring, by Corollary 2.23(b), R is local
or R ∼ = Mn (D) where D is a field. If R ∼ = Mn (D), then we are done. So assume that R is a
local ring and M is the maximal right ideal of R. By hypothesis there exists x ∈ R such that
M = xR. Note that x ∈ / M2 , for if not, we conclude that M = M2 and since M is nilpotent
we see that M = 0, a contradiction. Since R is a finite ring, R/M is a field and hence M/M2
is a right R/M-vector space. Thus M/M2 ∼ = R/M and so |M/M2 | = |R/M|. Note that M is
a two-sided ideal, so that by the following definition M/M2 is a left R/M-vector space
(r + M)(y + M2 ) = ry + M2 ,
for any r ∈ R and any y ∈ M. Suppose that dimR/M (M/M2 ) = n < ∞. Then |M/M2 | = |R/
M|n and so n = 1. Hence M/M2 is a left R/M-vector space of dimension 1 and so M =
Rx + M2 . Let t be the index of nilpotency of M. Then M = Rx + M2 = Rx + M3 = . . . =
Rx + Mt = Rx. Thus M = xR = Rx and Mn = xn R = Rxn , for any n ∈ N. Now by
Nakayama’s Lemma we have the following proper chain
0 = xt R ( xt−1 R ( · · · ( x2 R ( xR ( R.
Assume that 0 6= a ∈ R and i is the smallest number that a ∈ xi R\xi+1 R. If a is unit, then
aR = R. If a is not unit, then a ∈ M = xR and since a ∈ xi R\xi+1 R, then aR ⊆ xi R.
We show that xi R ⊆ aR. Since aR ⊆ xi R, there exists r ∈ R such that a = xi r. If r is
unit, then xi = ar −1 ∈ aR and hence xi R ⊆ aR. If r is not unit, then r ∈ M = xR and so
a = xi r ∈ xi RxR = xi+1 R, a contradiction. Hence aR = xi R and we conclude that R is a right
uniserial ring. Now by [10, Lemma 1], R is also a left uniserial ring.

Lemma 5.5. If M is a weakly uniserial right R-module, then M does not include nonzero torsion
submodules and torsion-free submodules simultaneously.

Proof. Assume that N and K are nonzero torsion-free and torsion submodules of M , respec-
f
tively. Since M is weakly uniserial, N ֌ K or K ֌ N . If N ֌ K, then for every n ∈ N there
exists a regular element r of R such that 0 = f (n)r = f (nr). Hence nr = 0 and so n = 0, a

19
f
contradiction. If K ֌ N , then for every k ∈ K there exists a regular element r in R such that
kr = 0. Hence f (k)r = f (kr) = 0, and so f (k) = 0. This means that k = 0, a contradiction.

Theorem 5.6. Let R be a commutative principal ideal domain and M be a finitely generated R-
module. Then M is weakly uniserial if and only if M ∼
= Rn or M ∼
= ⊕n (R/pR) or M ∼ = R/pn R,
where p is a prime element of R and n ≥ 0 is an integer.

Proof. Assume that M is a nonzero weakly uniserial R-module. By the fundamental structure
theorem for finitely generated modules over a principal ideal domain, M ∼ = Rn ⊕ R/pα1 1 R ⊕ · · · ⊕
R/pαr r R, where n ≥ 0 is an integer, αi ≥ 0 for 1 ≤ i ≤ r and each pi is a prime element in R.
Since R is torsion-free and for any i, R/pαi i R is torsion, by Lemma 5.5, this decomposition does
not include R and R/pαi i R simultaneously. Thus M ∼ = Rn or M ∼ = R/pα1 1 R ⊕ · · · ⊕ R/pαr r R.
Suppose that M ∼ = R/pα1 1 R ⊕ · · · ⊕ R/pαr r R. We proceed by cases:
Case 1: There exist pi , pj ∈ {p1 , . . . , pr } such that pi and pj are not associates. In this case we
α
consider R-module R/pαi i R ⊕ R/pj j R. Since M is weakly uniserial, without loss of generality,
α α
we can assume that R/pαi i R ֌ R/pj j R. It follows that AnnR (R/pj j R) ⊆ AnnR (R/pαi i R) and
α α
so pj j = pαi i r, for some r ∈ R. Thus pi |pj j and hence pi |pj , i.e., pi and pj are associates, a
contradiction.
Case 2: For every 1 ≤ i, j ≤ r, pi and pj are associates. In this case we can write M ∼ =
R/pα1 R ⊕ · · · ⊕ R/pαr R, for some prime element p of R. We show that if r > 1, then for any
1 ≤ i ≤ r, αi = 1. Suppose that αi > αj ≥ 1, for some 1 ≤ i, j ≤ r. It is easy to check that
R/pαi R is uniform (uniserial). Thus (pαi −1 R/pαi R) ⊕ (pαj −1 R/pαj R) 6֌ R/pαi R. Since M is
weakly uniserial, we must have R/pαi R ֌ (pαi −1 R/pαi R) ⊕ (pαj −1 R/pαj R). But it follows that
p ∈ AnnR (R/pαi R), a contradiction. Therefore for any i, αi = 1 and the proof of one direction
is complete.
Convesely, if M ∼ = R/pn R, then M is uniserial and so it is weakly uniserial. If M ∼ = ⊕n (R/pR),
then by Example 2.18(1), M is weakly uniserial. Finally, if M ∼ = R n , then by Proposition 2.12(c),

M is weakly uniserial.

Corollary 5.7. A finitely generated Z-module M is weakly uniserial if and only if M ∼


= Zn or
M∼= ⊕n Zp or M ∼= Zpn , where p is a prime number and n ≥ 0 is an integer.

We conclude the paper with the following result, which gives the structure of weakly uniserial
Z-modules with nonzero socle.

Proposition 5.8. Every Z-module M is weakly uniserial with Soc(M ) 6= 0 if and only if M ∼
=
Zpn or M ∼
= Zp∞ or M ∼= ⊕I Zp , where p is a prime number, n ∈ N, and I is an index set.

Proof. Suppose that M is a weakly uniserial Z-module with Soc(M ) 6= 0. Then by Proposition 2.21(a),
Soc(M ) ⊆e M . If Soc(M ) is simple, then M is cocyclic and so by [20, Theorem 2.6], there exists
a prime number p such that M ֌ HomZ (Z, Zp∞ ) ∼ = Zp∞ . Hence M ∼ = Zp∞ or M ∼ = Zpn for
some n ≥ 0. Now assume that Soc(M ) is not simple and S is a simple submodule of M . Then
S = xZ for some x ∈ S. If ord(x) = ∞, then xZ ∼ = Z, a contradiction. Thus ord(x) < ∞ and
we conclude that |xZ| = |S| = p for some prime number p. Now we claim that ord(m) = p for

20
every 0 6= m ∈ M . If there exists m ∈ M such that ord(m) = ∞, then mZ ∼ = Z and since M
is weakly uniserial, mZ ֌ S or S ֌ mZ, which in any case we have a contradiction. Thus
ord(m) < ∞ and so mZ is a finite weakly uniserial Z-module. By Corollary 5.2, |mZ| = q n
where q is a prime number and n ∈ N. We show that n = 1 and q = p. If n ≥ 2, then by the
first Sylow’s Theorem mZ contains a subgroup N of order q 2 . Since Soc(M ) is not simple, there
exists S1 ⊕S2 ⊆ Soc(M ), where S1 and S2 are isomorphic simple submodules and |S1 | = |S2 | = p.
Thus S1 ⊕ S2 ֌ N or N ֌ S1 ⊕ S2 , a contradiction. Hence ord(m) = q for every 0 6= m ∈ M .
Again since M is weakly uniserial, mZ ֌ S or S ֌ mZ and so p = q. Therefore, M is a vector
space over Zp by multiplication cx = cx for every c ∈ Zp and any x ∈ M . Then M ∼ = ⊕I Zp
where I is an index set. The converse is clear.

Acknowledgment. The research of the third author was in part supported by a grant from
IPM (No. 1400160414). This research is partially carried out in the IPM-Isfahan Branch.

References
[1] F. W. Anderson and K. R. Fuller, Rings and categories of modules, 2nd ed., vol. 13, Springer-
Verlag, New York, 1992.

[2] D. M. Arnold, Finite rank torsion free abelian groups and rings, Lecture Notes in Math.,
vol. 931, Springer-Verlag, Berlin, Heidelberg and New York, 1982.

[3] K. Asano, Über verallgemeinerte ablsche gruppe mit hyperkomplexen operatorenring und ihre
anwedungen, Jpn. J. Math. 15 (1939) 231–253.

[4] K. Asano, Über hauptidealringe mit kettensatz, Osaka J. Math. 1 (1949) 52–61.

[5] M. Behboodi, A. Daneshvar and M. R. Vedadi, Several generalizations of the Wedderburn-


Artin theorem with applications, Algebras and Representation Theory 21 (2018) 1333–1342.

[6] M. Behboodi, A. Daneshvar and M. R. Vedadi, Virtually semisimple modules and a gener-
alization of the Wedderburn-Artin theorem, Comm. Algebra 46 (2018) 2384–2395.

[7] M. Behboodi and S. Roointan-Isfahani, Almost uniserial rings and modules, J. Algebra 446
(2016) 176–187.

[8] M. Behboodi, A. Moradzadeh-Dehkordi and M. Qourchi Nejadi, Virtually uniserial modules


and rings, J. Algebra 549 (2020) 365–385.

[9] J. L. Chen, N. Q. Ding and H. Marubayashi, Advances in ring theory, World Scientific
Publishing Company, 2005.

[10] W. E. Clark and D. A. Drake, Finite chain rings, Abh. Math. Sem. Univ. Hamburg 39
(1973) 147–153.

[11] R. Göbel and J. Trlifaj, Approximations and endomorphism algebras of modules, de Gruyter
Expositions in Mathematics 41, Walter de Gruyter, Berlin and New York, 2006.

21
[12] S. K. Jain, A. K. Srivastava and A. A. Tuganbaev, Cyclic modules and the structure of
rings, Oxford University Press, USA, 2012.

[13] P. Karimi-Beiranvand and R. Beyranvand, Almost prime and weakly prime submodules, J.
Algebra Appl. 18 (2019) 1950129, 14 pp.

[14] G. Köthe, Verallgemeinerte ablesche gruppen mit hyperkomplexem operatorenring, Math. Z.


39 (1935) 31–44.

[15] T. Y. Lam, A first course in noncommutative rings, Graduate Texts in Mathematics, 131,
Springer-Verlag, New York, 1991.

[16] T. Y. Lam, Lectures on modules and rings, Graduate Texts in Mathematics, vol. 189,
Springer-Verlag, New York, 1999.

[17] T. Nakayama, On Frobeniusean algebras. II, Ann. of Math. 42 (1941) 1–21.

[18] L. A. Skornyakov, When are all modules serial, Mat. Zametki 5 (1969) 173–182.

[19] R. Wisbauer, Foundations of module and ring theory, Gordon and Branch Reading, 1991.

[20] S. M. Yahya, On cocyclic modules, Acta Mathematica Academiae Scientiarum Hungaricae


36 (1980) 147–153.

22

You might also like