ALL Metrics
-
Views
-
Downloads
Get PDF
Get XML
Cite
Export
Track
Research Article

Reverse order law for outer inverses and Moore-Penrose inverse in the context of star order

[version 1; peer review: 2 approved]
PUBLISHED 27 Jul 2022
Author details Author details
OPEN PEER REVIEW
REVIEWER STATUS

This article is included in the Manipal Academy of Higher Education gateway.

Abstract

The reverse order law for outer inverses and the Moore-Penrose inverse is discussed in the context of associative rings. A class of pairs of outer inverses that satisfy reverse order law is determined. The notions of left-star and right-star orders have been extended to the case of arbitrary associative rings with involution and many of their interesting properties are explored. The distinct behavior of projectors in association with the star, right-star, and left-star partial orders led to several equivalent conditions for the reverse order law for the Moore-Penrose inverse.

Keywords

Generalized inverses, Outer inverses, Moore-Penrose inverse, Reverse order law, Star order

Introduction

Given any invertible matrices A and B, it is well-known that (AB)−1 = B−1A−1. This property is often called reverse order law for the invertible matrices. This law is easily extended for the invertible elements from an associative ring with identity. This notion has been well studied in the literature and it is known that the analogue is not true in the case of generalized inverses, unless there are certain additional conditions. Several researchers came up with many necessary as well as sufficient conditions for the reverse order law to hold in the case of different generalized inverses. Ivan Erdelyi in 1966 discussed reverse order law for Moore-Penrose inverses of complex matrices in the context of partial isometries1. Greville, in his celebrated paper2, gave several equivalent conditions for the reverse order law for Moore-Penrose inverse of matrices over complex field. He proved that given two complex matrices A and B with AB defined, the reverse order law (AB) = BA holds if and only if the equations AABB*A* = BB*A* and BBA*AB = A*AB are satisfied, which is equivalent to the column space inclusions 𝒞 (BB*A*) ⊆ 𝒞 (A*) and 𝒞 (A*AB) ⊆ 𝒞 (B). Here, the notations and * denote the Moore-Penrose inverse and conjugate transpose, respectively. Motivated by this paper, Hartwig discussed equivalent conditions for triple reverse order law for Moore-Penrose inverses (see, 3) and Y. Tian discussed equivalent conditions for multiple reverse order law for Moore-Penrose inverses of complex matrices (see, 4).

The reverse order law for Moore-Penrose inverses in the context of elements from an associative ring with involution was studied previously5. Also, it has been extended to different generalized inverses including group inverse, Drazin inverse, core inverse, Drazin Moore-Penrose (DMP) inverse and pseudo core inverse, of matrices over a field, operators on Hilbert space, and elements of rings. Interested readers may refer to 4, 610 and the references therein.

Another notion of our concern is the star order, introduced by M. P. Drazin in 197811. In the beginning of 90’s, J. K. Baksalary and S. K. Mitra considered weaker conditions than those considered by Drazin and introduced left-star and right-star orders12. Benitez et al.,13 discussed some equivalent conditions for reverse order law for group inverse and Moore-Penrose inverse in terms of sharp and star partial orders, respectively, where most of his results assume that one of the matrices under consideration is Equal Projector (EP).

In the present paper, one of our main focuses is to study the reverse order law for Moore-Penrose inverses of arbitrary elements, which need not be EP, in terms of different partial orders. While exploring the extension of reverse order law for the class of outer inverses, it may be noted that an outer inverse of given element is not uniquely established, unlike in the cases of inverse, Moore-Penrose inverse and group inverses. So, a natural question arises is that whether every outer inverse (ab)= of ab could be written as gbga for some outer inverses gb of b and ga of a. In case the answer is affirmative, then verify if

{(ab)=}={b=a=}(reverseorderlaw?)

holds. While answering these questions, a few noteworthy properties of right-star, left-star and star partial orders are studied. Further, we give some equivalent conditions for the reverse order law for Moore-Penrose inverse in terms of different star partial orders.

Preliminaries

Throughout our discussion, 𝒜 denotes an associative ring, need not be with unity unless indicated otherwise. We use the notations a, b, c, ... for the elements of a ring and A, B, C, ... for matrices under discussion. Let a𝒜. If there exists a g𝒜 satisfying aga = a, then a is said to be regular and g is called a generalized inverse (or simply a g-inverse) of a. An arbitrary g-inverse of a is denoted by a. If there exists a g𝒜 satisfying gag = g, then such a g is called an outer inverse of a. An arbitrary outer inverse of a is denoted by a=. A generalized inverse, which is also an outer inverse, is called a reflexive generalized inverse and an arbitrary reflexive generalized inverse of a is denoted by ar. A commuting reflexive generalized inverse of a is called the group inverse of a and it is denoted by a#. It is unique whenever it exists. The classes {a}, {a=}, and {ar} denote the class of all g-inverses, outer inverses and reflexive g-inverses of a, respectively. For basic notions of generalized inverses in the context of matrices we refer the readers to 14, 15, and 16.

An involution ‘*’ of a ring 𝒜 is an anti-automorphism whose square is identity; i.e.

(a*)*=a(a+b)*=a*+b*(ab)*=b*a*foralla,b𝒜.

Involution ‘*’ is said to be proper (see 11) if for all a, b𝒜,

a*a=a*b=b*a=b*ba=b.

A proper *-ring is a ring equipped with a proper involution ‘*’. A ring is said to be regular ring if all of its elements are regular. A *-regular ring is a regular ring with involution ‘*’ such that a*a = 0 ⇒ a = 0. Observe that the involution is proper if and only if a*a = 0 ⇒ a = 0 holds. An important consequence of this property of involution is the *-cancellation law.

An element a is said to satisfy left (right) *-cancellation law if a*ax = a*ayax = ay (xaa* = yaa* ⇒ xa = ya). Element a is said to be *-cancellable if it satisfies both left and right *-cancellation laws.

For an associative ring 𝒜 with involution '*' and a𝒜, consider the following four equations:

axa=a(1)
xax=x(2)
(ax)*=ax(3)
(xa)*=xa.(4)

An element g satisfying the equations {i, j, ...} ⊆ {1, 2, 3, 4}, is called {i, j, ...}-inverse of a and the class of all {i, j, ...}-inverses of a is denoted by {a{i, j,...}}. An element g𝒜 satisfying all of (1), (2), (3) and (4), if exists, is unique and is called the Moore-Penrose inverse of a. The existence of Moore-Penrose inverse for the elements of a *-regular ring was proved by Kaplansky17, almost at the same time that Penrose18 proved the existence in the case of matrices. The Moore-Penrose inverse of an element a is denoted by a. We will use the notation 𝒜 to denote the set of all elements of 𝒜 having Moore-Penrose inverse. Note that the elements of 𝒜 are *-cancellable.

An element a is Hermitian if a* = a, and it is an idempotent if a2 = a. Hermitian idempotents are called projectors. If a is the Moore-Penrose inverse of a, then both aa and aa are projectors.

The right annihilator ao of an element a𝒜 is the set {x𝒜 : ax = 0} and the left annihilator oa is the set {y𝒜 : ya = 0}. If 1 ∉ 𝒜, then the right annihilator (1 − a)o is defined as {x𝒜 : x = ax} and the left annihilator o(1 − a) = {y𝒜 : y = ya} (see 19). For any a, b𝒜, we write acsb (arsb) if the principal right (left) ideals generated by a and b coincide. We say that a and b are space equivalent if both acsb and arsb are true. In such a case we write aspb.

The sum a + b is written as ab (⊕ refers to direct sum) if a𝒜b𝒜 = {0} and 𝒜a𝒜b = {0}. If so is the case, then we say that a and b are disjoint. We refer to 19, 20, and 21 for more details on the above notions.

The minus partial order is helpful in discussing many of the properties of star partial order. For the initial works on minus partial order, one may refer to 2123.

Definition 1 (Minus Partial Order21,22). The minus partial order is a relation, denoted by, defined on set of all regular elements of 𝒜 by a b if there exists a g-inverse g of a such that ag = bg and ga = gb.

For the proof of the following theorem, see 24 and 25.

Theorem 2 (Right-left Symmetry24,25). Let 𝒜 be an associative ring with unity and let a, b, c𝒜 such that a = b + c is regular. Then the following statements are equivalent.

  • (i) b a

  • (ii) b𝒜c𝒜 = a𝒜

  • (iii) b𝒜c𝒜 = {0} = 𝒜b𝒜c

  • (iv) 𝒜b𝒜c = 𝒜a

  • (v) c a

  • (vi) ba𝒜𝒜a and {a} ⊆ {b}

  • (vii) ca𝒜𝒜a and {a} ⊆ {c}.

Note that the theorem holds even in the case of an associative ring without unity, provided a, b, c are regular.

Definition 3 (Star Order11). Let 𝒜 be an associative ring with involution ‘*’ and a, b𝒜. The relation ≤* defined by a ≤* b if a*a = a*b and aa* = ba* is called star order on 𝒜. If a ≤* b, then we say that the element a is dominated by b under star order.

Drazin, in 11 noticed that the relation star order defined on a *-semigroup is a partial order. He pointed out that if the Moore-Penrose inverses of a and b exist, then,

a*baa=abandba=aa(5)

and

a*baa=baandaa=ab.(6)

He also noted that

a*ba*b.(7)

Definition 4. A partial order2 on a set S2 is said to be dominated by a partial order1 defined on set S1 if S2S1 and for a, bS2,

b2ab1a.

Note that the minus partial order dominates the star partial order (as seen from (5)).

Reverse order law for outer inverses

In this section we probe if the reverse order law mentioned in the earlier section holds. In the following lemma, we observe that the inclusion {(ab)=} ⊆ {b=a=} always holds.

Lemma 5. For a, b𝒜, we have

{(ab)=}{b=a=}.(8)

Proof. Given an outer inverse gab of ab, note that gaba is an outer inverse of b and bgab is an outer inverse of a. Further, we have gab = gab(ab)gab = (gaba)(bgab) ∈ {b=a=}, proving (8). Hence the lemma.

Unfortunately, the reverse of inclusion appearing in (8) need not be true. We have a counter example in the matrix case. For A=[1236] and B=[2211], consider GA=[1000] and GB=[0100], which are outer inverses of A and B, respectively. Note that GBGA=[0100] is not an outer inverse of AB=[81648], as we have

GBGAABGBGA=[0100][81648][0100]=[0400][0100].

This proves that {B=A=} ≠ {(AB)=}.

In 26 and 27, the authors presented necessary and sufficient conditions for {B{1,2,3}A{1,2,3}} ⊆ {(AB){1,2,3}} in the case of matrices and bounded linear operators on Hilbert spaces, respectively. An equivalent condition for the inclusion {B{1,2}A{1,2}} ⊆ {(AB){1,2}} is given in 28 for bounded linear operators on complex Hilbert spaces.

Now, our interest in this section is to characterize the subset of {b=a=}, which equals to {(ab)=}, where a, b𝒜.

Theorem 6. Let 𝒜 be any associative ring and let a, b𝒜. Then,

{(ab)=}={gbga:ga{a=},gb{b=},suchthat𝒜gb=𝒜a1orga𝒜=b1𝒜},(9)

where a1 = agaa a and b1 = bgbb b.

Proof. Note that outer inverse of any element is a regular element in the ring, and if x is any regular element in 𝒜, then xx𝒜𝒜x.

Now consider any gab ∈ {(ab)=}. As noted earlier in the Lemma 5, we have gab = gababgab in the form of gbga, where gb = gaba ∈ {b=} and ga = bgab ∈ {a=}. Now,

𝒜gb=𝒜gaba=𝒜abgababecausegababgab=gab=𝒜agaa=𝒜a1.

The proof of ga𝒜 = b1𝒜 is similar. From the definition of a1, it is verified that ga is a generalized inverse of a1. Further, we get a1ga = aga and gaa1 = gaa, proving that a1 a. b1 b is similarly proved.

Now to prove (9) and the theorem, consider ga ∈ {a=} and gb ∈ {b=} satisfying 𝒜gb = 𝒜a1 or ga𝒜 = b1𝒜. Now consider the case of 𝒜gb = 𝒜a1, whereas the other case similar. Since gb and a1 are regular elements, we have gb, a1𝒜gb = 𝒜a1 and therefore, a1bgb = a1 and gaabgb = gaa. Now, using the same we get

(gbga)(ab)(gbga)=gb(gaabgb)ga=gb(gaaga)=gbga,

proving that gbga ∈ {(ab)=}.

Remark 1. Note that the set {gbga : ga ∈ {a=}, gb ∈ {b=}, such that 𝒜gb = 𝒜a1 or ga𝒜 = b1𝒜} does not determine exact class of pairs of outer inverses (ga, gb) of a and b whose product determine {(ab)=}. It may be noted in each of the cases (i) ga𝒜b1𝒜 (ii) b1𝒜 ⊂ ga𝒜 (iii) 𝒜a1 ⊂ 𝒜gb (iv) 𝒜gb ⊂ 𝒜a1, we get

gbgaabgbga=gbga.

Therefore, determining the exact class of (ga, gb) determining {(ab)=} remains to be a problem to probe.

Let 𝒜 be an associative ring with involution ‘*’. Then, we have the following corollary.

Corollary 7. Given elements a, b𝒜, if ga and gb are reflexive g-inverses of a and b, respectively, satisfying any of the cases (a) ga𝒜b𝒜 (b) b𝒜ga𝒜 (c) 𝒜a𝒜gb (d) 𝒜gb𝒜a, then gbga{(ab)r}. Further, we have the following.

  • (i)  In the case of (a) or (c) and if ga is a {1, 2, 3}-inverse of a, then we have

    gbga{(ab){1,2,3}}.

  • (ii)  In the case of (b) or (d) and if gb is a {1, 2, 4}-inverse of b, then we have

    gbga{(ab){1,2,4}}.

  • (iii)  If a, b exists and a*𝒜b𝒜, then we have

    ba{(ab){1,2,3}}.

  • (iv)  If a, b exists and b𝒜a*𝒜, then we have

    ba{(ab){1,2,4}}.

  • (v)  If a, b exist and a*csb, then (ab) exists, in which case

    (ab)=ba.

Proof. In the cases of (a), (b), (c), and (d) we have bgbga = ga, gaab = b, abgb = a, and gbgaa = gb, respectively. Now, by direct substitution it is easily verified that gbga is a reflexive generalized inverse of ab.

If gaa{1,2,3} and the condition (a) or (c) is satisfied, then abgbga = aga proving that gbga ∈ {(ab){1,2,3}}. Hence, (i) is proved.

Similarly, if gba{1,2,4} and the condition (b) or (d) is satisfied, then we have gbgaab = gbb proving gbga ∈ {(ab){1,2,4}} and (ii).

Proof of (iii) and (iv) are consequences of (i) and (ii), respectively, since aspa*.

Proof of (v) is an immediate consequence of (iii) and (iv).

Let A and B be any two complex matrices with AB defined. According to Corollary 7 (v), we have (AB) = BA if (A) = (B*), where (X ) denotes the row space of X. However, the converse need not be true. For example, if

A=[1000]andB=[0110],

then (AB)=[0100] = BA. But (B*) ≠ (A).

Left-star and right-star orders

In the present section, we extend the notions of left-star and right-star orders to the case of elements from an associative ring with involution. These notions were originally defined in the matrix case by Baksalary and Mitra in 12. As noted earlier, Drazin proved that the star order is a partial order on a proper *-semigroup and hence on a proper *-ring (see 11).

Following the matrix case, we shall now define the left-star and right-star orders in the context of associative rings with involution.

Definition 8. For a, b𝒜, we say that a is dominated by b under left-star order, and denote a l* b if a*a = a*b, aa𝒜 and a𝒜b𝒜. We say that a is dominated by b under right-star order, and denote a r* b if aa* = ba*, a𝒜a and 𝒜a𝒜b.

Baksalary and Mitra in 12 showed that the right-star and left-star orders are partial orders on the class of matrices over complex field. In the case of rings, clearly these relations are reflexive. Further, if a l* b and b l* c, then we have a*a = a*b = a*c, since a* = xb* for some x and b*b = b*c. Thus, the the relation l* is transitive and preorder on ring 𝒜. Additionally, if a or b is left *-cancellable and a l* b and b l* a, then a = b. For the proof, consider the case of a being left *-cacellable. Then, a*a = a*ba = b, since b = ax for some x and aa𝒜. Hence, we have the following theorem.

Theorem 9. Let 𝒜 be an associative ring with involution ‘*. Then the relation l* (r*) is a partial order on the class of left *-cancellable (right *-cancellable) elements. In particular, if 𝒜 is a proper *-ring, then l* andr* are partial orders on 𝒜.

Now, we shall proceed to prove some of the basic properties of left-star ordering. Analogous results for the right-star order can be obtained in the similar manner.

Theorem 10. Let 𝒜 be an associative ring and let a𝒜 be left *-cancellable. Then the following statements hold.

  • (i)  a l* ba* r* b*

  • (ii)  If p is a projector and a l* p, then a is also a projector.

  • (iii)  If p and q are projectors, then the following are equivalent.

    • (a)  p l* q

    • (b)  p𝒜q𝒜

    • (c)  𝒜p𝒜q

    • (d)  p r* q

    • (e)  p ≤* q

    • (f)  p q

  • (iv)  If q is a projector, then a l* qa r* qa ≤* qa q.

Proof. Note that (i) follows from the properties of involution ‘*’.

From the hypothesis of (ii), we have a*a = a*p = pa and aa𝒜p𝒜. Therefore, a = pa = a*a proving that a is a projector. This proves (ii).

The equivalences of (a)-(f) in (iii) are easily verified using the properties of projectors.

Proof of (iv) follows from the statements (ii) and (iii).

Now, for a, b, c𝒜, we make the following observations.

  • (i)  a l* ba l* b and a r* ba r* b: If a l* b, then a*a = a*b, aa𝒜 and a𝒜b𝒜. So, a*a = b*a and 𝒜a = 𝒜a* ⊆ 𝒜b* = 𝒜b. Then, we have bba = a and abb = a. Thus, a𝒜. Also, aa = ab = b*(a)* and hence,

    a𝒜b*𝒜=b𝒜(10)

    Further,

    (a)*b=(a)*aab=(a)*abb=(a)*a(11)

    Thus, a l* b. Converse part follows by the uniqueness of Moore-Penrose inverse and the fact that (a) = a. A similar argument will prove the second equivalence.

  • (ii)  a l* baa = ab (a r* baa = ba): Follows by noting that aa = ab is equivalent to (10) and (11).

  • (iii)  If a l* b, then b is a {1,3}-inverse of a (If a r* b, then b is a {1,4}-inverse of a): From (ii) above, it follows that ab is Hermitian. Also, postmultiplying aa = ab by a, we get that b is a g-inverse of a.

  • (iv)  If a l* b (or a r* b), then (bba) = abb and (abb) = bba: If a l* b, then we have a𝒜b𝒜, 𝒜a𝒜b, a𝒜b𝒜 (refer (i) above) and hence, 𝒜a𝒜b. From the first two inclusions, we get that bba = a and abb = a proving that (bba) = abb. The last two inclusions imply the equations bba = a and abb = a proving that (abb) = bba.

  • (v)  l* and r* are partial orders on 𝒜: Evidently, l* is reflexive. Suppose that a l* b and b l* c. Then a = (a)*a*a = (a)*a*b = aab = b as it follows from b𝒜a𝒜 that aab = b. Hence the anti-symmetry. Now for some a.b, c𝒜, suppose that a l* b and b l* c. Then a*a = a*b and b*b = b*c with a𝒜b𝒜c𝒜. Now,

    a*a=a*b=a*(b)*b*b=a*(b)*b*c=(bba)*c=a*c

    as it follows from a𝒜b𝒜 that bba = a. Hence, l* is transitive, which proves that l* is a partial order.

  • (vi)  The minus order dominates the left-star and right-star orders: Suppose that a l* b. Then aa = ab and a𝒜b𝒜. So, a = aab and a = bba for any g-inverse b of b. Hence, ab𝒜𝒜b. Also,

    aba=aaaba=aabba=aaa=a

    and hence {b} ⊆ {a}. Now, from Theorem 2 it follows that a b.

  • (vii)  If a ≤* b, then b is a {1,3,4}-inverse of a and bab = a: The proof that b is a {1,3,4}-inverse of a follows from the equivalence given in (6). Further, using the same equivalence, we get that bab = baa = aaa = a.

    From (7), we have a ≤* b. In fact, if g is a {1,3,4}-inverse of a, then

    (a)*a=(a)*a(a)*a*g*a*=(a)*ag*a*=(a)*aag=(a)*g

    and similarly, a(a)* = g(a)*. Hence, a ≤* g. Therefore, we have

    a=min{g:aga=a,(ag)*=ag,(ga)*=ga},

    where the minimum is with respect to the star order (see Theorem 211).

  • (viii) If c ≤* a, then c is a {2,3,4}-inverse of a: The proof follows by (5). From (7), we have c ≤* a. In fact, it is true that h ≤* a for any {2,3,4}-inverse h of a. Thus, we have

    a=max{h:hah=h,(ah)*=ah,(ha)*=ha},

    where the maximum is with respect to the star order (Theorem 211).

  • (ix) The following implication follows from (5),

    a*bab.(12)

    If a ≤* b and c = ba, then by (12) and Theorem 2 it follows that b = ac. Moreover, b exists if and only if a and c exist, in which case,

    b=ac(13)

    This can be seen as follows: If a ≤* b, then a*c = a*ba*a = 0, which implies ac = 0. Similarly, ca = 0, ac = 0, and ca = 0.

The reverse order law for Moore-Penrose inverse

The projectors play a prominent role in obtaining several equivalent conditions for the reverse order law for Moore-Penrose inverse. They have distinct properties associated with the star order, a few of which are discussed in 29. In this section, we shall note a few more interesting behaviors of projectors associated with star, left-star and right-star orders. An important observation in proving our main theorem is the following fact: For a projector p, we have pa r* a (ap l* a) if and only if pa ≤* a (ap ≤* a).

Lemma 11. Let 𝒜 be an associative ring with involution ‘*’ and let p be a projector. For any a𝒜, the following are equivalent.

  • (i) (pa)(pa)* = a(pa)*

  • (ii) paa* ∈ o(p − 1)

  • (iii) p commutes with aa*

  • (iv) paa* is Hermitian

  • (v) pa ≤* a

In particular, if a𝒜a, then

par*aifandonlyifpa*a.(14)

Proof. Note that (i)⇒(ii), (ii)⇒(iii), and (iii)⇒(iv) are obvious.

Evidently, (pa)*(pa) = (pa)*a. Now, pa(pa)* = p(pa)a* = paa* = a(pa)*. This proves (iv)⇒(v).

Now, (14) follows from the definition of star order, equivalence of (i) and (v), and by noting that a𝒜a implies that pa𝒜pa, as p is idempotent.

Symmetrically, we have the following result.

Lemma 12. Let 𝒜 be an associative ring with involution ‘*’ and let p be a projector. For any a𝒜, the following are equivalent.

  • (i) (ap)*(ap) = (ap)*a

  • (ii) a*ap ∈ (1 − p)o

  • (iii) p commutes with a*a

  • (iv) a*ap is Hermitian

  • (v) ap ≤* a

In particular, if aa𝒜, then

apl*aifandonlyifap*a.(15)

Lemma 13. Let 𝒜 be an associative ring with involution ‘*’ and let a𝒜. If p is a projector commuting with a, then pa ≤* a and ap ≤* a.

Proof. Follows by noting that the hypothesis implies statement (iii) of Lemma 11 as well as Lemma 12.

Now we prove the main theorem of the present paper. The equivalence of statements (i) and (vii) is discussed in 2 and in 21 and the equivalence of (i) and (vi) is proved in 30. However, our proof is in the general setting of Lemma 11 and Lemma 12, which use the properties of projectors.

Theorem 14. Let 𝒜 be a proper *-ring. For a, b𝒜, the following are equivalent.

  • (i) (ab) = ba

  • (ii) (aab) = baa and (abb) = bba.

  • (iii) aab r* b and bb l* a

  • (iv) (aab)(aab)* = b(aab)* and (abb)*(abb) = (abb)*a

  • (v) aabb* ∈ o(aa − 1) and a*abb ∈ (1 − bb)o

  • (vi) aabb* = bb*aa and a*abb = bba*a

  • (vii) aabb* and a*abb are Hermitian

  • (viii) aab ≤* b and abb ≤* a

Proof. (i)⇒(ii) can be verified by direct substitution. Conversely, premultiplying aab(baa)aab = aab by a and postmultiplying (baa)aab(baa) = baa by a, we get abbaab = ab and baabba = ba, respectively. Also, we have abba = abb(bba) and baab = (baa)aab. Hence, abba and baab are Hermitian. Thus, we have (ab) = ba, proving (ii)⇒(i). This proves (i)⇔(ii).

The equivalence of the statements (iii)-(viii) will follow from the Lemma 11 and Lemma 12 noting that aa and bb are projectors. It remains to prove that (i)(viii).

(i)⇔(viii): Evidently, (aab)*aab = b*a*(a)*a ab = (b*aa)b = (aab)*b.

Premultiplying b*a* = baab(ab)* by abb*b, we get

abb*bb*a*=abb*bbaab(ab)*=abb*aabb*a*.

Taking u = bb*a* and v = aabb*a*, we note that u*u = u*v = v*u = v*v = abb*aabb*a* and hence u = v, since ‘*’ is a proper involution. That is, bb*a* = aabb*a*. Therefore, we get

aab(aab)*=aabb*a*(a)*=bb*a*(a)*=b(aab)*

which proves that aab ≤* b. Proof of other inequality is symmetric.

Conversely, we have that b and a are {1,3,4}-inverses of aab and abb, respectively. Hence, baab and abba are Hermitian. Also, premultiplying aabbaab = aab by a, we get abbaab = ab.

Further, 𝒜 (aab)* ⊆ 𝒜b* = 𝒜b, gives

(aab)*bb=(aab)*i.e.,b*aabb=b*aa

Premultiplying by b(b)* and post-multiplying by a, we get

baabba=ba

proving (i).

From (i)⇔ (ii) of Theorem 14, we have (aab) = baa and (abb) = bba if and only if the reverse order law for Moore-Penrose inverse holds. Further from (i)⇔(viii) and (7), we have (ab) = ba if and only if (aab) ≤* b and (abb) ≤* a. Hence, we have the following corollary.

Corollary 15. Let 𝒜 be a proper *-ring and let a, b𝒜. Then the following are equivalent.

  • (i) (ab) = ba

  • (ii) baa l* b and bba r* a

  • (iii) (baa)*(baa) = (baa)*b and (bba)(bba)* = a(bba)*

  • (iv) (b)*baa ∈ (aa − 1)o and bba(a)* ∈o (1 − bb)

  • (v) aa(b)*b = (b)*baa and a(a)*bb = bba(a)*

  • (vi) aa(b)*band a(a)*bbare Hermitian

  • (vii) baa ≤* b and bba ≤* a

From Theorem 14, we have that aab r* b and abb l* a if and only if the reverse order law (ab) = ba holds. The equivalence fails to exist if we interchange the left-star and right star orders. However, one of the implications is true and the same is proved in the following theorem.

Theorem 16. Let 𝒜 be a proper *-ring. Let a, b𝒜 be such that a and b exist. Then,

(ab)=ba{aabl*babbr*a.

Proof. We have, (aab)*(aab) = b*a*(a)*aab = b*aab = (aab)*b. Further, taking u = aab and v = bbaab, we get u*u = u*v = v*u = v*v = b*aab.

Thus,

aab=bbaabandhenceaab𝒜b𝒜

which proves that aab l* b. Proof of other inequality is symmetric.

Remark 2. The converse of the above theorem need not be true in general. For example, if

A=[10120]andB=[1012],

then

A=45[11200],B=B1=12[2011]

so that

AAB=45[11200]andABB=[10120].

Clearly, AAB l* B and ABB r* A. But

(AB)=[1000][11500]=BA.

From Lemma 11 and Lemma 12, we have that aab r* b implies aab ≤* b and abb l* a implies abb ≤* a. It is not true in general that aab ≤* b whenever aab l* b and that abb ≤* a whenever abb r* a. The following lemma gives a sufficient condition for this result to hold.

Theorem 17. Let 𝒜 be a proper *-ring and let a, b𝒜. Suppose (ab) = ba. Then, aab l* b implies aab ≤* b and abb r* a implies abb ≤* a.

Proof. Since aab𝒜b𝒜 , we have

aab(aab)*=bbaab(aab)*=bbaab(ab)*(a)*=b(ab)*((ab))*(ab)*(a)*=b(ab)*(a)*=b(aab)*

as desired. The second implication can be proved similarly.

The proof of the following theorem is immediate from Theorem 14 and Lemma 13, noting that aa and bb are projectors commuting with b and a, respectively. This theorem shows that only two commutative relations out of four, given by Erdelyi in (Theorem 11) are sufficient for the reverse order law.

Theorem 18. Let 𝒜 be a proper *-ring and let a, b𝒜 be such that a and b exist. If both a(bb) = (bb)a and b(aa) = (aa)b are true, then (ab) = ba.

Conclusions

In the present paper, the reverse order law for outer inverses is discussed. The notions of left-star and right-star orderings are extended to the case of associative rings with involution and several properties of the same are discussed. Also, necessary and sufficient conditions—in terms of star, left-star and right-star orders—for the reverse order law for Moore-Penrose inverse are given. The discussion on left-star and right-star partial orderings appears to help in further study of relation between star partial order and column space decomposition of matrices. Further, it has been noted in Theorem 10 that, in the class of all projectors the star, left-star, right-star and minus orders are all equivalent. The classification of elements of associative rings for which any two or more orders are equivalent seems to be an interesting problem to explore and the same will be presented in the subsequent papers.

Data availability

Underlying data

All data underlying the results are available as part of the article and no additional source data are required.

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 27 Jul 2022
Comment
Author details Author details
Competing interests
Grant information
Copyright
Download
 
Export To
metrics
Views Downloads
F1000Research - -
PubMed Central
Data from PMC are received and updated monthly.
- -
Citations
CITE
how to cite this article
Kelathaya U and Karantha MP. Reverse order law for outer inverses and Moore-Penrose inverse in the context of star order [version 1; peer review: 2 approved]. F1000Research 2022, 11:843 (https://doi.org/10.12688/f1000research.123411.1)
NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.
track
receive updates on this article
Track an article to receive email alerts on any updates to this article.

Open Peer Review

Current Reviewer Status: ?
Key to Reviewer Statuses VIEW
ApprovedThe paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approvedFundamental flaws in the paper seriously undermine the findings and conclusions
Version 1
VERSION 1
PUBLISHED 27 Jul 2022
Views
7
Cite
Reviewer Report 25 Apr 2023
Mehsin Jabel Atteya, Department of Mathematics, Al- Mustansiriyah University, Falastin St, Baghdad, Iraq 
Approved
VIEWS 7
At the beginning, the counter examples and examples which are used in this article assist in understanding the specific points. During this paper, the authors provide background information about this subject in the introduction with preliminary results which are used ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Atteya MJ. Reviewer Report For: Reverse order law for outer inverses and Moore-Penrose inverse in the context of star order [version 1; peer review: 2 approved]. F1000Research 2022, 11:843 (https://doi.org/10.5256/f1000research.135513.r169472)
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.
Views
21
Cite
Reviewer Report 21 Oct 2022
Balaji Ramamurthy, Department of Mathematics, Indian Institute of Technology Madras, Chennai, Tamil Nadu, India 
Approved
VIEWS 21
The reverse order law in the case of matrix inverses i.e., (AB) −1 = B−1A−1 is well known in the literature. But the same is not true when the inverses are replaced by generalized inverses. However, in the case of ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Ramamurthy B. Reviewer Report For: Reverse order law for outer inverses and Moore-Penrose inverse in the context of star order [version 1; peer review: 2 approved]. F1000Research 2022, 11:843 (https://doi.org/10.5256/f1000research.135513.r146946)
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 27 Jul 2022
Comment
Alongside their report, reviewers assign a status to the article:
Approved - the paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations - A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approved - fundamental flaws in the paper seriously undermine the findings and conclusions
Sign In
If you've forgotten your password, please enter your email address below and we'll send you instructions on how to reset your password.

The email address should be the one you originally registered with F1000.

Email address not valid, please try again

You registered with F1000 via Google, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Google account password, please click here.

You registered with F1000 via Facebook, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Facebook account password, please click here.

Code not correct, please try again
Email us for further assistance.
Server error, please try again.