Minus Partial Order in Regular modules
B. Ungor
Burcu Ungor, Department of Mathematics, Ankara University, Ankara, Turkey
[email protected]
,
S. Halicioglu
Sait Halıcıoglu, Department of Mathematics, Ankara University, Ankara, Turkey
[email protected]
,
A. Harmanci
Abdullah Harmanci, Department of Mathematics, Hacettepe University, Ankara, Turkey
[email protected]
and
J. Marovt
Janko Marovt, Faculty of Economics and Business, University of Maribor,
Razlagova 14, SI-2000 Maribor, Slovenia
[email protected]
Abstract.
The minus partial order is already known for sets of
matrices over a field and bounded linear operators on
arbitrary Hilbert spaces. Recently, this partial order has
been studied on Rickart rings. In this paper, we extend the concept of the minus
relation to the module theoretic setting and prove that this
relation is a partial order when the module is regular. Moreover,
various characterizations of the minus partial order in regular
modules are presented and some well-known results are also
generalized.
Key words and phrases:
Minus partial order,
regular module
2010 Mathematics Subject Classification:
06A06, 06F25, 06F99, 16B99
1. Introduction
Let S be a semigroup and
a∈S. Any solution x=a− to the equation axa=a
is called *an inner generalized inverse *of a. If such
a− exists, then a is called regular, and if every
element in a semigroup S is regular, then
S is called a regular semigroup. Hartwig
[7] introduced the minus partial order ≤− on
regular semigroups using generalized inverses. For a regular
semigroup S and a,b∈S, we write
[TABLE]
for some inner generalized inverse a− of a.
Let B(H) be
the algebra of all bounded linear operators on a Hilbert space H. For
an operator A∈B(H), the symbols Ker A
and Im A will denote the kernel and the image of A,
respectively. It is known that A∈B(H) is regular if and only if
ImA=ImA, i.e. the image of A is
closed (see for example [12]).
Šemrl studied in [14] the minus partial order on
B(H). He did not want to restrict himself only to
operators on B(H) with closed images so he defined a
new order ≤S on B(H) in the following way: For
A,B∈B(H) we write A≤SB if there exist
idempotent operators P,Q∈B(H) such that
ImP=ImA,
KerA=KerQ, PA=PB, and
AQ=BQ. Šemrl called this order the minus partial order on B(H) and proved that this is indeed a partial order on
B(H) for a general Hilbert space H. He
also showed that the partial order ≤S is the same as
Hartwig’s minus partial order ≤− when H is
finite dimensional.
For a subset A of a ring R, lR(A) and rR(A) will
denote the left annihilator and the right annihilator of A in
R, respectively. If the subset A is a singleton, say
A={a}, then again we simply write lR(a) and rR(a),
respectively. A ring R is called a Rickart ring if for every a∈R
there exist idempotent elements p,q∈R such that rR(a)=p⋅R
and lR(a)=R⋅q. Note that every
Rickart ring R has the (multiplicative) identity and that the class of
Rickart rings includes von Neumann algebras and rings with no proper zero divisors
(see [2]
or [9]).
Following Šemrl’s approach, the authors further generalized in
[4] the minus partial order to Rickart
rings. A new relation was introduced on a ring with identity: Let
R be a ring with the identity 1R and a,b∈R. Then we
write a≤−b if there exist idempotent elements p,q∈R
such that
[TABLE]
It was proved that this relation ≤− is indeed a partial
order when R is a Rickart ring and that definitions
(1) and (2) are
equivalent when R is a ring in which every element is regular,
i.e. R is a von Neumann regular ring.
In this paper, we will study the minus partial order in a more
general setting. The goal of this paper is to extend the notion of
the minus partial order to modules using their endomorphism rings.
We show that this minus relation is a partial order when the
module is regular. We also present various characterizations of
the minus partial order in regular modules and generalize some
well-known results.
Throughout this paper R denotes an associative ring
with identity 1R and modules are unitary right R-modules. For a
right R-module MR=M, S= EndR(M) is the ring of all right R-module
endomorphisms of M. It is well known that M is a left S and
right R-bimodule. In this work, for the (S,R)-bimodule M,
lS(.) and rR(.) stand for the left annihilator of a subset
of M in S and the right annihilator of a subset of M in R,
respectively. If the subset is a singleton, say {m}, then we
simply write lS(m) and rR(m), respectively.
2. Minus partial order in modules
Let M be a right R-module with S= EndR(M). For the sake
of brevity, in the sequel, S will stand for the endomorphism
ring of the module M considered. We will denote the identity map
in S by 1S. An element m∈M is called (Zelmanowitz) regular if
[TABLE]
for some φ∈M∗ where M∗= HomR(M,R) denotes the dual of M.
For a ring R, let a∈R be a regular element (in the sense of
von Neumann) so that there exists b∈R such that a=aba.
Define the map φ:R→R with φ(r)=br,r∈R. Then φ∈R∗= EndR(R). We have
[TABLE]
which yields that a is regular in RR
(in the sense of Zelmanowitz, see also [15]).
Let now R be a ring with identity and suppose a is a regular element in RR (in the sense of Zelmanowitz).
Then there exists φ∈R∗ such that a=aφa.
Define a−=φ(1R)∈R. Then
[TABLE]
We may conclude that a∈R is regular if and only if a is regular in
RR (or, similarly, in the left-R module RR). A module M is called regular (in the
sense of Zelmanowitz) if every element of M is
regular.
Remark 2.1**.**
[13*, Lemma B.47]** *Let M be a module and m∈M regular,
say m=mφm where φ∈M∗. Then e=φm∈R
is an idempotent, mR≅eR is projective, and M=mR⊕N
where N={n∈M∣mφn=0}.
Remark 2.2**.**
Let M be a module. It is known that HomR(R,M)≅M.
Let m∈M be regular, say m=mφm where φ∈M∗. Then for the map mφ:M→M defined by
[TABLE]
we may conclude that mφ∈S and that mφ is an idempotent in S.
With the following definition we will introduce the concept of
minus order in the setting of modules.
Definition 2.3**.**
Let M be a module and m1,m2∈M.
We write m1≤−m2
if there exists φ∈M∗ such that m1=m1φm1,
m1φ=m2φ, and φm1=φm2. We call
the relation ≤− the minus order on M.**
We will prove that when the module M is regular, the relation ≤− is a partial order. First, let us present an auxiliary result and a new characterization of the minus order.
Proposition 2.4**.**
Let M be a module and m1,m2∈M.
If m1≤−m2, then m1R⊆m2R.
Proof.
Assume that m1≤−m2. Then there exists φ∈M∗
such that m1=m1φm1,
m1φ=m2φ, and φm1=φm2. Hence
m1=m1φm1=m2φm1=m2φm2∈m2R and thus
m1R⊆m2R.
∎
Theorem 2.5**.**
Let M be a module and m1,m2∈M
with m1 regular. Then m1≤−m2 if and only if there
exist f2=f∈S, a2=a∈R such that lS(m1)=lS(f),
rR(m1)=rR(a), fm1=fm2, and m1a=m2a.
Proof.
Let m1≤−m2. Then there exists φ∈M∗ such that m1=m1φm1,
m1φ=m2φ, and φm1=φm2. Let
f=m1φ and a=φm1. Then f2=f∈S and
a2=a∈R. Clearly, lS(m1)⊆lS(f). From
m1=m1φm1 we obtain (1S−m1φ)m1=0. So,
lS(f)=S(1S−m1φ)⊆lS(m1). Hence
lS(m1)=lS(f). Obviously, rR(m1)⊆rR(a).
Similarly, since m1=m1φm1, m1(1R−φm1)=0, it follows
rR(a)=(1R−φm1)R⊆rR(m1). Thus
rR(m1)=rR(a). Also, fm1=m1φm1=m1φm2=fm2 and m1a=m1φm1=m2φm1=m2a.
Conversely, assume that there exist f2=f∈S, a2=a∈R
such that lS(m1)=lS(f), rR(m1)=rR(a), fm1=fm2, and
m1a=m2a. Also, m1=m1φm1 for some φ∈M∗. Then m1=fm1=m1a. Let β=aφf∈M∗. Hence
m1βm1=m1aφfm1=m1φm1=m1. Also, we
have m1β=m1aφf=m2aφf=m2β and βm1=aφfm1=aφfm2=βm2. Therefore
m1≤−m2.
∎
We are now in position to prove that the minus order ≤− is indeed
a partial order when the module M is regular.
Theorem 2.6**.**
Let M be a regular module.
Then the relation ≤− is a partial order on M.
Proof.
Reflexivity: Obvious.
Antisymmetry: Let m1,m2∈M with m1≤−m2
and m2≤−m1. Then there exist φ,β∈M∗
such that
[TABLE]
and
[TABLE]
By Proposition
2.4, m1R=m2R. Remark 2.1 yields that
M=m1R⊕N1 where N1={n∈M∣m1φn=0}.
Since m1φ(m1−m2)=0, m1−m2∈N1, say m1−m2=n∈N1. Hence m1=n+m2, and also m2=m1r for some r∈R.
Thus m1=n+m1r, so n=m1−m1r∈N1∩m1R={0}.
This implies that m1=m1r. Therefore m1=m2.
Transitivity: Let m1,m2,m3∈M with m1≤−m2 and m2≤−m3. By Theorem 2.5, there exist
f12=f1∈S, a12=a1∈R such that lS(m1)=lS(f1),
rR(m1)=rR(a1), f1m1=f1m2, m1a1=m2a1; and
f22=f2∈S, a22=a2∈R such that lS(m2)=lS(f2),
rR(m2)=rR(a2), f2m2=f2m3, m2a2=m3a2. From
(1S−f1)m1=0 and m1(1R−a1)=0 it follows
[TABLE]
Let f=f1+f1f2(1S−f1) and
a=a1+(1R−a1)a2a1. Then f2=f∈S and a2=a∈R. We
claim that
lS(m1)=lS(f), rR(m1)=rR(a),
fm1=fm3, and m1a=m3a.
- (i)
Clearly, lS(m1)=lS(f1)⊆lS(f). Let g∈lS(f). Then gff1=0. It follows gf1=0 and so g∈lS(f1)=lS(m1).
Hence lS(m1)=lS(f).
2. (ii)
It is obvious that rR(m1)=rR(a1)⊆rR(a).
Let x∈rR(a). Then a1ax=0. This yields a1x=0, i.e.
x∈rR(a1)=rR(m1), and thus rR(m1)=rR(a).
3. (iii)
We will now prove fm1=fm3. By (3), fm1=m1.
Similarly, fm2=m2.
From the definition of f we obtain, fm3=f1m3+f1f2m3−f1f2f1m3.
Note that f1f2m3=f1f2m2=f1m2=f1m1=m1.
Then
[TABLE]
We claim that f1(1S−f2)f1=0, that is, f1(1−f2)∈lS(f1)=lS(m1).
By f1f2m1=f1f2m2a1=f1m2a1=f1m1, we obtain f1(1S−f2)m1=0 and
hence f1(1S−f2)f1=0, as claimed. Thus fm3=m1 and
therefore fm1=fm3.
4. (iv)
We assert that m1a=m3a. We have m1a1=m1 due to
m1(1R−a1)=0. Similarly, m2a2=m2. From
m3a=m3a1+m3a2a1−m3a1a2a1 and since
m3a2a1=m2a2a1=m2a1=m1a1=m1, we obtain
[TABLE]
We now claim that a1(1R−a2)a1=0, i.e.
(1R−a2)a1∈rR(a1)=rR(m1). By
m1a2a1=f1m2a2a1=f1m2a1=m1a1 it follows that
m1(1R−a2)a1=0. Then a1(1R−a2)a1=0, as required. Hence
m3a=m1 and thus m1a=m3a.
By Theorem 2.5, m1≤−m3 and so the relation
≤− is a partial order on M.
∎
We close this section by giving the star analogues of the minus
order on modules. A ring equipped with an involution ∗ will
be called a ∗-ring. A ∗-ring R is called a
Rickart ∗-ring if the left annihilator lR(a) of
any element a∈R is generated by a projection (i.e. a
self-adjoint idempotent), equivalently, the right annihilator
rR(a) of any element a∈R is generated by a projection.
Recall that Drazin [6] defined the star partial order in
a general setting of proper ∗-semigroups and note that
natural special cases of proper ∗-semigroups are all
proper ∗-rings, with “properness” defined via
aa∗=0 implying a=0. Recall also (see e.g. [2])
that any Rickart ∗-ring has the identity and is a proper
∗-ring.
For Rickart ∗-rings, the following characterization of the
star partial order ∗≤ was given in
[10]. Let R be a Rickart ∗-ring. For
a,b∈R we have a∗≤b if and only if there
exist projections p,q∈R such that
[TABLE]
Observe here (see e.g. [4, Lemma 2.1])
that for any idempotent p in a ring R with identity,
[TABLE]
Two orders that are closely related to the star and the minus
partial orders are the left-star and the
right-star partial orders. These orders were introduced by
Baksalary and Mitra in [1] on the set of
m×n complex matrices Mm,n(C). The left-star
partial order is defined as follows. For A,B∈Mm,n(C),
[TABLE]
The right star partial order ≤∗ is
defined symmetrically: A≤∗B if and
only if AA∗=BA∗ and
ImA∗⊆ImB∗. These two
orders were generalized in [5] from the
set of all n×n complex matrices to B(H) where
H is an arbitrary complex Hilbert space. Using
annihilators, authors further generalized in
[10] the left and the right-star partial
orders to Rickart ∗-rings.
Motivated (4) and (5), and by
generalizations of the left-star and the right-star orders to
Rickart ∗-rings in [10], we will now
extend the notion of these orders to the module theoretic setting.
Definition 2.7**.**
Let M be a module and m1,m2∈M. We write
m1≤∗ m2 if there
exist f2=f∈S, a2=a=a∗∈R such that lS(m1)=lS(f),
rR(m1)=rR(a), fm1=fm2, and m1a=m2a, where R is a
∗-ring. We call the relation
≤∗ the right-star order on M.
m1 ∗≤ m2 if there
exist f2=f=f∗∈S, a2=a∈R such that lS(m1)=lS(f),
rR(m1)=rR(a), fm1=fm2, and m1a=m2a, where S is a
∗-ring. We call the relation
∗≤ the left-star order on M.
m1∗≤m2 if there
exist f2=f=f∗∈S, a2=a=a∗∈R such that
lS(m1)=lS(f), rR(m1)=rR(a), fm1=fm2 and m1a=m2a,
where R, and S are ∗-rings. We call the relation
∗≤ the star order on M.
We will prove that the relations introduced with Definition
2.7 are partial orders when the module M over a Rickart
∗-ring is regular.
Theorem 2.8**.**
Let M be a regular module. Then the
following hold.
- (1)
If R is a Rickart ∗-ring, then the relation
≤∗, introduced
with Definition 2.7, is a partial order on M.
2. (2)
If S is a Rickart ∗-ring, then the relation
∗≤, introduced
with Definition 2.7, is a partial order on M.
3. (3)
If R and S are Rickart ∗-rings, then the
relation ∗≤, introduced
with Definition 2.7, is a partial order on M.
Proof.
We will only prove (1). The proofs of (2) and (3) are similar.
Reflexivity: Let m∈M. Since M is regular, there exists
φ∈M∗ such that m=mφm. Let
f=mφ and b=φm. Then f2=f∈S and b2=b∈R. Clearly, lS(m)⊆lS(f). From m=mφm we have
m=fm. Thus lS(f)⊆lS(m) and hence lS(m)=lS(f).
Similarly, rR(m)=rR(b). By assumption R is a Rickart
∗-ring so since b∈R, there exists (the unique) a∈R
with a=a2=a∗ and rR(a)=rR(b). To sum up, there exists
f2=f∈S, a2=a=a∗∈R such that lS(m)=lS(f),
rR(m)=rR(a). We may conclude that m≤∗ m
for every m∈M.
Antisymmetry: Let m1,m2∈M with
m1≤∗ m2 and m2≤∗ m1. Then there exist f12=f1∈S, a12=a1=a1∗∈R
such that lS(m1)=lS(f1), rR(m1)=rR(a1),
f1m1=f1m2, m1a1=m2a1; and f22=f2∈S,
a22=a2=a2∗∈R such that lS(m2)=lS(f2),
rR(m2)=rR(a2), f2m2=f2m1, m2a2=m1a2. By
(5), lS(f1)=S(1S−f1) and
rR(a1)=(1R−a1)R which imply by assumption that
(1S−f1)m1=0 and m1(1R−a1)=0. So,
m1=f1m1=m1a1=m2a1. Similarly, m2=f2m2=f2m1. Hence
[TABLE]
Transitivity: Let m1,m2,m3∈M with
m1≤∗ m2 and m2≤∗ m3. Then there exist f12=f1∈S, a12=a1=a1∗∈R
such that lS(m1)=lS(f1), rR(m1)=rR(a1),
f1m1=f1m2, m1a1=m2a1; and f22=f2∈S,
a22=a2=a2∗∈R such that lS(m2)=lS(f2),
rR(m2)=rR(a2), f2m2=f2m3, m2a2=m3a2. From
(1S−f1)m1=0 and m1(1R−a1)=0 we obtain
[TABLE]
Let f=f1f2 and a=a1. To
conclude the proof we will show that f2=f∈S and
lS(m1)=lS(f),
fm1=fm3, and m1a=m3a.
From rR(m2)=rR(a2) we have m2=m2a2 and thus
m1a1a2=f1m2a2=f1m2=m1. So m1(a1a2−1R)=0. Since
rR(m1)=rR(a1), it follows that a1(a1a2−1R)=0 and hence
a1a2=a1. Recall that a1 and a2 are self-adjoined. So,
a1=a2a1 which yields
[TABLE]
Since lS(m2)=lS(f2), we have m2=f2m2 which implies
f2m1=f2m2a1=m2a1=m1. So,
(1S−f1f2)m1=m1−f1f2m1=m1−f1m1=m1−m1=0. Since
lS(m1)=lS(f1), (1−f1f2)f1=0, so f1=f1f2f1. Now
fm1=f1f2m1=m1 and fm3=f1f2m3=f1f2m2=f1m2=m1. So
[TABLE]
Also f2=(f1f2)2=f1f2f1f2=f1f2. Finally,
since f=f1f2 and f1=f1f2f1=ff1 we get,
lS(m1)=lS(f1)=lS(f).
We may conclude that m1≤∗ m3.
∎
3. Some Characterizations of the Minus order on Modules
In this section we investigate some properties and
characterizations of the minus partial order on modules.
Proposition 3.1**.**
Let M be a regular module and
m1,m2∈M. Then we have the following.
- (1)
m1≤−m2 if and only if gm1≤−gm2
for every invertible element g∈S.
2. (2)
m1≤−m2 if and only if m1b≤−m2b
for every invertible element b∈R.
Proof.
(1) Assume first that m1≤−m2. By Theorem 2.5, there exist
f2=f∈S, a2=a∈R such that lS(m1)=lS(f),
rR(m1)=rR(a), fm1=fm2, and m1a=m2a. Let g∈S be invertible. Note first that gm1 is a regular element in M since M is a regular module. Denote p=gfg−1. Then p∈S is an idempotent. We will show that lS(gm1)=lS(p),
rR(gm1)=rR(a), pgm1=pgm2, and gm1a=gm2a.
- (i)
If h∈lS(gm1), then hgm1=0 and so hgfm1=0. Hence
hgf∈lS(m1)=lS(f). Thus hgf=0 and so hgfg−1=0.
Therefore lS(gm1)⊆lS(p). Conversely, if x∈lS(p), then xp=xgfg−1=0. Hence
xgm1=xgfm1=xgfg−1gm1=0 which yields lS(p)⊆lS(gm1) and therefore lS(gm1)=lS(p).
2. (ii)
If x∈rR(gm1), then gm1x=0 and so
m1x=0. Hence x∈rR(m1)=rR(a). Thus rR(gm1)⊆rR(a). Conversely, if x∈rR(a), then m1x=0 and so
gm1x=0. Hence x∈rR(gm1) and thus rR(a)⊆rR(gm1). So, rR(gm1)=rR(a).
3. (iii)
pgm1=gfg−1gm1=gfm1=gfm2=gfg−1gm2=pgm2.
4. (iv)
Since m1a=m2a, gm1a=gm2a.
Therefore gm1≤−gm2.
If gm1≤−gm2 for an invertible g∈S, then g−1gm1≤−g−1gm2, i.e. m1≤−m2.
(2) Assume first, that
m1≤−m2. By Theorem 2.5, there exist f2=f∈S, a2=a∈R such that lS(m1)=lS(f), rR(m1)=rR(a),
fm1=fm2, and m1a=m2a. For an invertible element b∈R,
let q=b−1ab. Note again that m1b is a regular element in M since M is a regular module. Then q∈R is an idempotent. We will show
that lS(m1b)=lS(f), rR(m1b)=rR(q), m1bq=m2bq, and
fm1b=fm2b.
- (i)
If h∈lS(m1b), then hm1b=0 and so hm1bb−1=0.
Hence hm1=0 and so h∈lS(m1)=lS(f). Thus
lS(m1b)⊆lS(f). Conversely, if h∈lS(f)=lS(m1), then hm1=0 and so hm1b=0. Hence h∈lS(m1b), i.e. lS(f)⊆lS(m1b), and therefore
lS(m1b)=lS(f).
2. (ii)
If x∈rR(q), then qx=0 and thus
b−1abx=0. Hence bb−1abx=0 and so abx=0. From
bx∈rR(a)=rR(m1) we obtain m1bx=0 and therefore rR(q)⊆rR(m1b). Conversely, if x∈rR(m1b), then m1bx=0 which yields
bx∈rR(m1)=rR(a). Hence abx=0 and so b−1abx=0.
Thus qx=0, i.e. x∈rR(q). It follows that rR(m1b)⊆rR(q) and therefore rR(m1b)=rR(q).
3. (iii)
m1bq=m1bb−1ab=m1ab=m2ab=m2bb−1ab=m2bq.
4. (iv)
Since fm1=fm2, fm1b=fm2b.
We conclude that m1b≤−m2b.
If m1b≤−m2b for an invertible b∈R, then m1bb−1≤−m2bb−1, i.e. m1≤−m2.
∎
We will next prove that if we somewhat relax conditions
on left and right annihilators in Theorem 2.5, we still
obtain the minus partial order on regular modules.
Theorem 3.2**.**
Let M be a regular module and m1,m2∈M.
Then m1≤−m2 if and only if there exist f2=f∈S,
a2=a∈R such that lS(f)⊆lS(m1), rR(a)⊆rR(m1), fm1=fm2, and m1a=m2a.
Proof.
Suppose first that there exist f2=f∈S, a2=a∈R such
that lS(f)⊆lS(m1), rR(a)⊆rR(m1),
fm1=fm2, and m1a=m2a. Since M is a regular module,
for m1∈M, there exists φ∈M∗ such that m1=m1φm1. Let
[TABLE]
Clearly, f12=f1 and a12=a1. We will first show that
lS(f1)=lS(m1) and rR(a1)=rR(m1). Let g∈lS(m1). Then gm1=0 and so gf1=gm1φ=0. Hence
g∈lS(f1) and thus lS(m1)⊆lS(f1).
Conversely, if g∈lS(f1), then gf1=0 and so gm1=gm1φm1=gf1m1=0. It follows that g∈lS(m1), i.e. lS(f1)⊆lS(m1), and thus
lS(f1)=lS(m1). Now let b∈rR(m1). Then m1b=0
and so a1b=φm1b=0. Hence b∈rR(a1) and so
rR(m1)⊆rR(a1). For the reverse inclusion, let
b∈rR(a1). Then m1b=m1φm1b=m1a1b=0 and
hence b∈rR(m1). So, rR(m1)=rR(a1).
Next we will prove that
lS(f)⊆lS(f1) and rR(a)⊆rR(a1). Let g∈lS(f). Then gf=0 and so gf1=gm1φ=gfm1φ=0. Hence g∈lS(f1), i.e.
lS(f)⊆lS(f1). For the other inclusion, let b∈rR(a). Then ab=0 and so a1b=φm1b=φm1ab=0. It follows that b∈rR(a1), i.e. rR(a)⊆rR(a1). Since then (1S−f)f1=0 and a1(1R−a)=0 and since f1 and a1 are idempotents, we have f1=ff1=f1ff1 and a1=a1a=a1aa1. Note that f1f∈S and aa1∈R are idempotents.
We will now prove that lS(m1)=lS(f1f),
rR(m1)=rR(aa1).
Clearly, lS(m1)=lS(f1)⊆lS(f1f) and
rR(m1)=rR(a1)⊆rR(aa1). For any x∈lS(f1f) and any y∈rR(aa1), we have xf1=xf1ff1=0
and a1y=a1aa1y=0. This yields x∈lS(f1)=lS(m1) and
y∈rR(a1)=rR(m1). It follows that
lS(m1)=lS(f1f) and rR(m1)=rR(aa1), as desired.
Finally, from fm1=fm2 and m1a=m2a, we establish
that f1fm1=f1fm2 and m1aa1=m2aa1. Therefore
m1≤−m2.
The converse implication follows directly by Theorem 2.5.
∎
Corollary 3.3**.**
Let M be a regular module and m1,m2∈M.
Then m1≤−m2 if and only if there exist f2=f∈S,
a2=a∈R such that m1R⊆fM, Sm1⊆Ma,
fm1=fm2, and m1a=m2a.
Proof.
Note that for any m∈M, f2=f∈S and a2=a∈R,
- (1)
lS(f)⊆lS(m) if and only if mR⊆fM,
2. (2)
rR(a)⊆rR(m) if and only if Sm⊆Ma.
The result is thus an immediate consequence of Theorem 3.2.
∎
There are many definitions that are equivalent to Hartwig’s
definition of the minus partial order. One of the most commonly
used is the following which is due to Jones (see for example
[8]). Let S be a semigroup. Then
[TABLE]
where p,q are idempotent elements in S1 and
S1 denotes S if S has the
identity, and S with the identity adjoined in the
other case. Jones introduced this relation for the setting of an
arbitrary semigroup. If S is a regular semigroup, then
it is known (see [11]) that ≤J is a partial order
on S. We will now introduce a new relation on a module M
that is analogous to (6) and then show that when M is
a regular module this new relation is the minus partial order on
M.
Definition 3.4**.**
Let M be a module and m1,m2∈M. We write m1≤Jm2
if there exist idempotent elements f∈S and a∈R such
that m1=fm2=m2a.
Lemma 3.5**.**
Let M be a module and m1,m2∈M. If m1≤−m2,
then there exist f2=f∈S, a2=a∈R such that
m1=fm1=fm2=m1a=m2a.
Proof.
If m1≤−m2, then there exist f2=f∈S, a2=a∈R such that
lS(m1)=lS(f), rR(m1)=rR(a), fm1=fm2, and m1a=m2a.
Hence (1S−f)m1=0 and m1(1R−a)=0, and therefore
m1=fm1=fm2=m1a=m2a.
∎
Theorem 3.6**.**
Let M be a regular module and m1,m2∈M.
Then m1≤Jm2 if and only if m1≤−m2.
Proof.
If m1≤−m2, then m1≤Jm2 by Lemma 3.5. Conversely,
let m1≤Jm2. Then m1=fm2=m2a for some f2=f∈S, a2=a∈R. Clearly, fm1=fm2 and m1a=m2a. From
m1=fm2 we get lS(f)⊆lS(m1) and by
m1=m2a we obtain rR(a)⊆rR(m1). Therefore an
application of Theorem 3.2 completes the proof.
∎
In [11], Mitsch generalized Hartwig’s definition of the
minus partial order to arbitrary semigroups. The definition
follows: Suppose a,b are two elements of an arbitrary semigroup
S. Then we write
[TABLE]
for some elements x,y∈S1. Mitsch proved that
≤M is indeed a partial order for any semigroup
S and that for a,b∈S,
[TABLE]
for some elements x,y∈S1. Following the above
ideas, we now introduce a new relation ≤M on a module M.
Definition 3.7**.**
Let M be a module and
m1,m2∈M. We write m1≤Mm2 if there exist f∈S and a∈R such that m1=m2a=fm2 and m1=fm1.
We will show that when M is a regular module, ≤M is again
the minus partial order on M. First, we present an auxiliary result which was inspired by (8).
Lemma 3.8**.**
Let M be a module and m1,m2∈M.
Then m1≤Mm2 if and only if there exist f∈S and a∈R such that
[TABLE]
Proof.
Let m1≤Mm2. Then there exist f∈S and a∈R such that m1=m2a=fm2 and m1=fm1. Hence
m1a=(fm2)a=f(m2a)=fm1=m1. The converse is obvious.
∎
Theorem 3.9**.**
Let M be a regular module and m1,m2∈M. Then
m1≤−m2 if and only if m1≤Mm2.
Proof.
Let m1≤−m2. Then by Theorem 3.6 there exist
f2=f∈S and a2=a∈R with m1=fm2=m2a. It follows,
(1S−f)m1=0 and so m1=fm1. Hence m1≤Mm2.
Conversely, let m1≤Mm2. Then there exist f∈S and
a∈R with m1=fm2=m2a and m1=fm1. Since M is a
regular module, there exists φ∈M∗ such that
m1=m1φm1. Let g=m1φ∈S and b=φm1∈R. Obviously, g2=g and b2=b. Also, clearly,
lS(m1)=lS(g) and rR(m1)=rR(b). Equation m1=fm1
implies (1S−f)m1=0. Therefore (1S−f)g=0 and
hence g=fg. It follows (gf)2=gf. Also, m1a=m1 by Lemma
3.8. Then m1(1R−a)=0 and so b(1R−a)=0
which implies b=ba. Hence (ab)2=ab. We now take into
account the following items:
- (i)
Let x∈lS(gf). Then 0=xgf=xm1φf and so
[TABLE]
Hence x∈lS(m1) and thus lS(gf)⊆lS(m1).
2. (ii)
Let x∈rR(ab). Then 0=abx=aϕm1x and so
[TABLE]
Hence x∈rR(m1) and thus rR(ab)⊆rR(m1).
3. (iii)
gfm1=gm1=gfm2.
4. (iv)
m1ab=m1b=m2ab.
By Theorem 3.2, we may conclude that m1≤−m2.
∎
Corollary 3.10**.**
Let M be a regular module and m1,m2∈M.
The following statements are then equivalent:
- (i)
m1≤−m2;
2. (ii)
There exist g∈S and b∈R such that m1=m2b=gm2 and
m1=m1b.
Proof.
(i) ⇒ (ii) Clear by Theorem 3.9 and Lemma 3.8.
(ii) ⇒ (i) There exist g∈S and b∈R such
that m1=m2b=gm2 and m1=m1b. We have
[TABLE]
and thus (i) holds.
∎
With the next result, which follows directly by Lemma 3.5, we will present a relation between annihilators of
elements m1 and m2 of a module M where m1 and m2 are connected by the
minus partial order.
Proposition 3.11**.**
Let M be a module and m1,m2∈M. If
m1≤−m2, then lS(m2)⊆lS(m1) and
rR(m2)⊆rR(m1).
The following example shows that the converse statement of
Proposition 3.11 need not hold in general.
Example 3.12**.**
Let Z10 denote the ring of integers modulo 10. It
is known that EndZ(Z10)≅Z10. Since
Z10 is a semisimple Z-module, it is a
regular module. Note that the nontrivial idempotents of
Z10 are 5 and 6. Consider
the elements m1=2 and m2=6 of
Z10.
Then, on the one hand, lZ10(m1)=lZ10(m2)={0,5} and
rZ(m1)=rZ(m2)=5Z. On the
other hand, there are no idempotents f∈ EndZ(Z10) and a∈Z such that m1=fm2=m2a. Thus,
m1≤−m2. **
In [3], Blackwood at al. defined a partial order
≤⊕ in the following way:
[TABLE]
and called it the direct sum partial
order. The direct sum partial order is known to be
equivalent to the minus partial order on von Neumann regular
rings (see [3, Lemma 3]). Motivated by the notion of the direct sum partial order on (regular) rings, we will conclude the paper by introducing a new relation on a module M and then show that this relation is the minus partial order when the module M is regular.
Definition 3.13**.**
Let M be a module and m1,m2∈M.
We write
[TABLE]
Theorem 3.14**.**
Let M be a module and m1,m2∈M with m1 and m2 regular.
Then the following statements are equivalent:
- (i)
m1≤−m2;
2. (ii)
m1≤⊕m2.
Proof.
(1)⇒(2): Let m1≤−m2.
There exists φ∈M∗ such that m1=m1φm1. By Remark 2.2, we have M=m1R⊕T where
T={x∈M ∣ m1φx=0}. To prove m1≤⊕m2, we will show that m2R=m1R⊕(m2−m1)R. Let first x∈m2R. Then x=m2r for some r∈R and therefore
[TABLE]
By Proposition 2.4, m1R⊆m2R which yields (m2−m1)R⊆m2R. Thus,
[TABLE]
Let a∈R. Observe that
[TABLE]
and therefore
(m2−m1)R⊆T. Since M=m1R⊕T, it follows that m1R∩(m2−m1)R={0}. Thus, m2R=m1R⊕(m2−m1)R.
(2) ⇒ (1): Let m1≤⊕m2, i.e. m2R=m1R⊕(m2−m1)R. There exist φ2∈M∗ and r∈R
with m2=m2φ2m2 and m1=m2r.
Then m2φ2m1=m2φ2m2r=m2r=m1 and hence
[TABLE]
Thus m1φ2m1−m1=−(m2−m1)φ2m1∈m1R∩(m2−m1)R. Since m1R∩(m2−m1)R={0}, it follows m1φ2m1=m1. Therefore m2φ2m1=m1=m1φ2m1. Let x=φ2m1φ2∈M∗.
Then m1xm1=m1. Note that m2x=m2φ2m1φ2=m1φ2 and m1x=m1φ2m1φ2=m1φ2 and hence m1x=m2x. Also,
m2φ2(m2−m1)=m2φ2m2−m2φ2m1=m2−m1, so
[TABLE]
and this
implies that m1φ2(m2−m1)=(m2−m1)−(m2−m1)φ2(m2−m1)∈m1R∩(m2−m1)R. Since m1R∩(m2−m1)R=0, we obtain m1φ2m2=m1φ2m1 and m2−m1=(m2−m1)φ2(m2−m1). From m1φ2m1=m1, it follows m1φ2m2=m1. Now, xm2=φ2m1φ2m2=φ2m1 and xm1=φ2m1φ2m1=φ2m1. Thus xm1=xm2 and
therefore m1≤−m2.
∎
The following result follows directly from Theorem 3.14.
Corollary 3.15**.**
If M is a regular module, then Definitions 2.3 and 3.13 are equivalent, i.e. the relation
≤⊕ is the minus partial order on M.
It can be seen that for nonzero elements, the minus order is the
identity relation on vector spaces. But this is not the case if
the module is regular as the following example shows.
Example 3.16**.**
Consider M=Z6 as an R=Z30-module.
On the one hand, since R≅M⊕Z5, M is a
projective R-module. On the other hand, M is semisimple as an
R-module. Hence M is a regular R-module. Let
m1=2, m2=5∈M. Then m2R=m1R⊕(m2−m1)R, i.e., 5Z30=M=2Z30⊕3Z30. This implies
m1≤⊕m2. Therefore m1≤−m2 by Theorem
3.14.**
Acknowledgment. The work is partially supported by
Slovene Research Agency, ARRS (Slovene-Turkish Grant
BI-TR/17-19-004) and by The Scientific and Technological Research
Council of Turkey, TUBITAK (Grant TUBITAK-116F435). The authors
wish to thank ARRS and TUBITAK for financial support.