All our rings are associative rings with identity and all our modules are left modules, unless specifically mentioned. U ( R ) U(R) U ( R ) denotes the set of all units of R R R , J ( R ) J(R) J ( R ) denotes the Jacobson radical, T ( a ) T(a) T ( a ) denotes the set of all outer or pseudo-inverses of a a a , and Idem ( R ) \operatorname{Idem}(R) Idem ( R ) denotes the set of idempotents of the ring R R R . By regular we mean von Neumann regular. By sr ( R ) \operatorname{sr}(R) sr ( R ) we denote the stable range of the ring R R R ; and by sr R ( a ) \operatorname{sr}_{R}(a) sr R ( a ) , we denote the (left) stable range of the element a a a in the ring R R R .
An element a a a in the skew corner e R f eRf e R f is invertible if there is some a ′ ∈ f R e a^{\prime} \in fRe a ′ ∈ f R e which satisfies a a ′ = e aa^{\prime} = e a a ′ = e and a ′ a = f a^{\prime}a = f a ′ a = f . Such elements appear in the literature in several guises. The idempotents e e e and f f f are algebraically Murray–von Neumann equivalent precisely when e R f eRf e R f contains an invertible element. This equivalence plays a fundamental role in nonstable algebraic K K K -theory. Two idempotents in M n ( R ) M_n(R) M n ( R ) are equivalent exactly when the corresponding modules are isomorphic. Then the equivalence classes of idempotents in M ∞ ( R ) = ⋃ n ≥ 1 M n ( R ) M_{\infty}(R) = \bigcup_{n \ge1} M_n(R) M ∞ ( R ) = ⋃ n ≥ 1 M n ( R ) , with addition induced by block-diagonal sums, form the monoid V ( R ) V(R) V ( R ) of isomorphism classes of finitely generated projective right R R R -modules [18 , 31 ] . From a semigroup-theoretic perspective, Green’s relations [13 ] compare elements of R R R through the one-sided ideals they generate: x R y x \mathrel{\mathcal{R}} y x R y if x R = y R xR = yR x R = y R , x L y x \mathrel{\mathcal{L}} y x L y if R x = R y Rx = Ry R x = R y , and H = R ∩ L \mathcal{H} = \mathcal{R} \cap\mathcal{L} H = R ∩ L . An element a a a is an invertible block of e R f eRf e R f exactly when a R = e R aR = eR a R = e R and R a = R f Ra = Rf R a = R f . Here the invertible blocks of e R f eRf e R f form a single H \mathcal{H} H -class (when there are any), and a ↦ a ′ a \mapsto a^{\prime} a ↦ a ′ maps them bijectively onto
In the sense of Loos [26 ] , the pair ( e R f , f R e ) (eRf,fRe) ( e R f , f R e ) is called an associative pair , modelled on pairs of rectangular matrices, and invertible pairs are invertible elements of this pair. Skew corners also enter the theory of stable range through Ara and Goodearl’s [2 ] reduction of problems of higher stable rank to stable rank one for skew corners, and the Q B QB QB -rings [3 ] which rest on quasi-invertible elements of skew corners, that is, elements x ∈ p R q x \in pRq x ∈ pR q with ( p − x y ) ⊥ ( q − y x ) (p-xy) \perp(q-yx) ( p − x y ) ⊥ ( q − y x ) for some y ∈ q R p y \in qRp y ∈ q R p [3 ] .
In the following sections we will systematically show how block invertibility in a skew corner e R f eRf e R f can be used to visualize Drazin’s ( b , c ) (b,c) ( b , c ) -inverse, which contains the Moore–Penrose, Drazin, group, core and dual core inverses, the Bott–Duffin inverse and the inverse along an element as special cases. We will envision the stable range condition using our notion of one-sided block invertibility. As applications we will provide key results in the case of Bott–Duffin inverses, von Neumann regularity, unit-regularity, stable range and special cleanness. For special-clean element, using block invertibility
2020 Mathematics Subject Classification. Primary 16U50, 16E50, 16D20.
Key words and phrases. skew-corners, stable-range, Bott-Duffin inverse, special-clean. we provide a characterization of when every unit-regular element is special clean, strengthening the theorem of Camillo and Khurana locally to elements.
Introduction
Following Ara and Goodearl [2 ] , for idempotents e e e , f ∈ R f \in R f ∈ R the subset e R f eRf e R f , naturally regarded as an ( e R e , f R f ) (eRe, fRf) ( e R e , f R f ) -bimodule, is called the skew corner (also rectangular corner ) of R R R determined by the pair ( e , f ) (e, f) ( e , f ) . Likewise f R e fRe f R e , an ( f R f , e R e ) (fRf, eRe) ( f R f , e R e ) -bimodule, is the skew corner determined by ( f , e ) (f, e) ( f , e ) . When e = f e=f e = f , the skew corner e R e eRe e R e is the classical corner ring of R R R , a ring with identity e e e . Every a ∈ R a \in R a ∈ R , compressed on either side by e e e and f f f , produces the element e a f ∈ e R f eaf \in eRf e a f ∈ e R f , which we call the compression or rectangular block (or simply the block ) of a a a in e R f eRf e R f .
Before analyzing this relation, we fix some terminology for the subsets in R R R in which our elements are associated with.
Definition 1.1. Let e e e , f ∈ R f \in R f ∈ R be arbitrary idempotents. The skew-corner pairs e R f eRf e R f and f R e fRe f R e are called associate skew-corners . A compression block x ∈ e R f x \in eRf x ∈ e R f is called block invertible in e R f eRf e R f if there exists a compression block y y y in the associate skew-corner f R e fRe f R e such that
x y = e and y x = f . xy=e \qquad\text{and}\qquad yx=f. x y = e and y x = f . In this case, x x x and y y y are called block inverses of each other.
Kaplansky [17 ] called idempotents e e e and f f f equivalent , written e ∼ f e \sim f e ∼ f , if there exists a pair of block inverses x ∈ e R f x \in eRf x ∈ e R f and y ∈ f R e y \in fRe y ∈ f R e . Equivalent idempotents are also called Murray–von Neumann equivalent or Kaplansky equivalent . Block invertibility is positional: whether an element is block invertible depends on the skew-corner in which it is considered and its inverse in the corresponding associate corner. For this reason, we sometimes record the skew-corners explicitly and speak of a block invertible pair ( x , y ) (x,y) ( x , y ) in ( e R f , f R e ) (eRf, fRe) ( e R f , f R e ) . This notion of invertible blocks in the skew corners also appears as ( e , f ) (e,f) ( e , f ) -inverse, or ( e , f ) (e,f) ( e , f ) -invertibility; see Rakić and Djordjević [34 ] and Mary [28 ] .
We use the notation e ∼ x f e \overset{x}{\sim} f e ∼ x f to specify the invertible block x = e x f ∈ e R f x=exf \in eRf x = e x f ∈ e R f . In this language, Kaplansky’s definition reads: e ∼ f e \sim f e ∼ f if and only if the skew corner e R f eRf e R f contains an invertible block x x x . Invertible blocks have exactly one unique block inverse.
Representing an idempotent equivalence e ∼ f e \sim f e ∼ f with e ∼ x f e \overset{x}{\sim} f e ∼ x f allows us to use the chosen invertible block x x x to serve as a basis for the entire skew-corner in the form e R f = x ( f R f ) eRf=x(fRf) e R f = x ( f R f ) . This is because
e ∼ x f ⟹ ∃ y ∈ f R e such that x y = e and y x = f . e \overset{x}{\sim} f \Longrightarrow\exists y \in fRe\ \text{such that}\ xy=e\ \text{and}\ yx=f. e ∼ x f ⟹ ∃ y ∈ f R e such that x y = e and y x = f .
Then for any element s ∈ e R f s \in eRf s ∈ e R f we have y s ∈ f R f ys \in fRf y s ∈ f R f and
s = e s = ( x y ) s = x ( y s ) s=es=(xy)s=x(ys) s = es = ( x y ) s = x ( y s )
This representation is unique because if x t 1 = x t 2 xt_1=xt_2 x t 1 = x t 2 with t 1 , t 2 ∈ f R f t_1,t_2 \in fRf t 1 , t 2 ∈ f R f , then
t 1 = f t 1 = y x t 1 = y x t 2 = f t 2 = t 2 t_1=ft_1=yxt_1=yxt_2=ft_2=t_2 t 1 = f t 1 = y x t 1 = y x t 2 = f t 2 = t 2
This gives us our bijection t ↦ x t t \mapsto xt t ↦ x t from f R f → e R f fRf \to eRf f R f → e R f with inverse s ↦ y s s \mapsto ys s ↦ y s . Including the invertible block in the equivalence notion also allows us to obtain the invertible block in a composition i.e.
e ∼ x f ∼ u g ⟹ e ∼ x u g e \overset{x}{\sim} f \overset{u}{\sim} g \Longrightarrow e \overset{xu}{\sim} g e ∼ x f ∼ u g ⟹ e ∼ xu g
Another observation is that we can represent the set of all invertible blocks in e R f eRf e R f as either x U ( f R f ) x \mathop{\mathrm{U}}(fRf) x U ( f R f ) or U ( e R e ) x \mathop{\mathrm{U}}(eRe)x U ( e R e ) x . We therefore have a one-to-one correspondence from U ( f R f ) \mathop{\mathrm{U}}(fRf) U ( f R f ) or U ( e R e ) \mathop{\mathrm{U}}(eRe) U ( e R e ) onto the invertible blocks of e R f eRf e R f .
Continuing in this way, it is natural to ask what else survives of the above notions when the skew corner e R f eRf e R f contains a one-sided inverse.
Definition 1.2. Let e e e , f ∈ R f \in R f ∈ R be idempotents and let x ∈ e R f x \in eRf x ∈ e R f . An element y y y of the associate skew-corner f R e fRe f R e is called a right block inverse of x x x if x y = e xy=e x y = e , and a left block inverse of x x x if y x = f yx=f y x = f . We say that x x x is right (resp. left ) block invertible in e R f eRf e R f if it admits a right (resp. left) block inverse in the associate skew corner f R e fRe f R e .
Note that x x x is block invertible precisely when it is both left and right block invertible. In this case if e ∼ f e \sim f e ∼ f , then every right (resp. left) invertible block of e R f eRf e R f is exactly x x x times the right (resp. left) invertible block elements of f R f fRf f R f . The same works for e R e eRe e R e but by post-multiplication by x x x . In particular, we can see that when e ∼ f e \sim f e ∼ f , every right or left invertible block of e R f eRf e R f (f R e fRe f R e ) is invertible if and only if either f R f fRf f R f or e R e eRe e R e is directly finite. This has interesting consequences that allows us to answer when left ( b , c ) (b,c) ( b , c ) invertibility is equivalent to right ( b , c ) (b,c) ( b , c ) invertibility [11 pg. 71] , [39 ] . We shall see that later in the next section when we visualize Drazin’s ( b , c ) (b,c) ( b , c ) inverse as a traditional left-right invertibility notion.
Now the relation symbol captured by one-sided block invertibility is subequivalence , for which we use the notation below. First recall that the minus partial order was defined by Hartwig[15 ] and Nambooripad [32 ] on von Neumann regular elements a , b ∈ R a,b \in R a , b ∈ R and given below:
a ≤ − b if and only if a = e b = b f for idempotents e 2 = e , f 2 = f ∈ R a \leq^{-} b \text{ if and only if } a=eb=bf \text{ for idempotents } e^{2}=e,\ f^{2}=f \in R a ≤ − b if and only if a = e b = b f for idempotents e 2 = e , f 2 = f ∈ R
Then the set of idempotents of R R R , Idem ( R ) \operatorname{Idem}(R) Idem ( R ) is partially ordered by the minus partial order as follows:
p ≤ − q ⟺ p = p q = q p . p \leq^{-} q \Longleftrightarrow p=pq=qp. p ≤ − q ⟺ p = pq = q p .
Definition 1.3 (Ara–Goodearl [2 ] ). Let e , f ∈ R e,f \in R e , f ∈ R be idempotents. We say e e e is subequivalent to f f f , written e ≲ f e \lesssim f e ≲ f , if there exists an idempotent g ∈ R g \in R g ∈ R such that
e ∼ g and g ≤ − f . e \sim g \qquad\text{and}\qquad g \leq^{-} f. e ∼ g and g ≤ − f .
In this case e R eR e R is isomorphic to a direct summand of f R fR f R and this occurs precisely when there exists elements x ∈ e R f x \in eRf x ∈ e R f and y ∈ f R e y \in fRe y ∈ f R e such that x y = e xy=e x y = e .
We will use e ≲ x f e \mathrel{\overset{x}{\lesssim}} f e ≲ x f to denote that the compression block e x f ∈ e R f exf \in eRf e x f ∈ e R f being right invertible (i.e. there exists a w ∈ f R e w \in fRe w ∈ f R e such that ( e x f ) w = e (exf)w=e ( e x f ) w = e ). Dually, for y ∈ R y \in R y ∈ R we write e ≲ y f e \mathrel{\lesssim^{y}} f e ≲ y f if the compression f y e ∈ f R e fye \in fRe f y e ∈ f R e is left invertible, that is, z ( f y e ) = e z(fye)=e z ( f y e ) = e for some z ∈ e R f z \in eRf z ∈ e R f . The two notations describe the same relation e ≲ f e \lesssim f e ≲ f from its two sides i.e. if x ∈ e R f x \in eRf x ∈ e R f and y ∈ f R e y \in fRe y ∈ f R e satisfy x y = e xy=e x y = e , then
e ≲ x f and e ≲ y f e \mathrel{\overset{x}{\lesssim}} f \qquad\text{and}\qquad e \mathrel{\lesssim^{y}} f e ≲ x f and e ≲ y f
both hold, and we record the relation with whichever of the two blocks is at hand. In each notation the block is written next to its left identity: x = e x x=ex x = e x beside e e e , and y = f y y=fy y = f y beside f f f .
We see that the relation ≲ \lesssim ≲ is a preorder on Idem ( R ) \operatorname{Idem}(R) Idem ( R ) : it is reflexive (p ≲ p p \lesssim p p ≲ p ) and transitive. Indeed if p ≲ q p \lesssim q p ≲ q and q ≲ s q \lesssim s q ≲ s for idempotents p , q , s ∈ Idem ( R ) p,q,s \in\operatorname{Idem}(R) p , q , s ∈ Idem ( R ) then there exists elements x 1 ∈ p R q x_{1} \in pRq x 1 ∈ pR q , y 1 ∈ q R p y_{1} \in qRp y 1 ∈ q R p and x 2 ∈ q R s x_{2} \in qRs x 2 ∈ q R s and y 2 ∈ s R q y_{2} \in sRq y 2 ∈ s R q such that x 1 y 1 = p x_{1}y_{1}=p x 1 y 1 = p and x 2 y 2 = q x_{2}y_{2}=q x 2 y 2 = q . Consequently,
( x 1 x 2 ) ( y 2 y 1 ) = x 1 ( x 2 y 2 ) y 1 = x 1 ( q ) y 1 = x 1 y 1 = p (x_{1}x_{2})(y_{2}y_{1})=x_{1}(x_{2}y_{2})y_{1}=x_{1}(q)y_{1}=x_{1}y_{1}=p ( x 1 x 2 ) ( y 2 y 1 ) = x 1 ( x 2 y 2 ) y 1 = x 1 ( q ) y 1 = x 1 y 1 = p
where ( x 1 x 2 ) ∈ p R s (x_{1}x_{2}) \in pRs ( x 1 x 2 ) ∈ pR s and ( y 2 y 1 ) ∈ s R p (y_{2}y_{1}) \in sRp ( y 2 y 1 ) ∈ s R p and therefore, p ≲ s p \lesssim s p ≲ s .
It is known [35 Ex. 2.2] that the relation e ≲ f ≲ e e \lesssim f \lesssim e e ≲ f ≲ e need not give e ∼ f e \sim f e ∼ f . However, if the left and right block inverses of x x x coincide then it is true i.e. ,
e ≲ x f and f ≲ x e ⟺ e ∼ x f e \mathrel{\overset{x}{\lesssim}} f \text{ and } f \mathrel{\overset{x}{\lesssim}} e \Longleftrightarrow e \mathrel{\overset{x}{\sim}} f e ≲ x f and f ≲ x e ⟺ e ∼ x f
In general, a left or right invertible block in a skew corner need not be block invertible, even under stable range one assumption of the ambient ring. Take the following example of R = M 2 ( k ) R=M_{2}(k) R = M 2 ( k ) with f = I 2 f=I_{2} f = I 2 and e = ( 1 0 0 0 ) e=\begin{pmatrix}1&0\\0&0\end{pmatrix} e = ( 1 0 0 0 ) which has stable range one. Then e ≲ f e f e \mathrel{\overset{fe}{\lesssim}} f e ≲ f e f since ( e f ) ( f e ) = e (ef)(fe)=e ( e f ) ( f e ) = e but f ≲ r e f \mathrel{\overset{r}{\lesssim}} e f ≲ r e is impossible for any r ∈ f R e r \in fRe r ∈ f R e as that would force r s = I 2 rs=I_{2} r s = I 2 for some s ∈ e R f s \in eRf s ∈ e R f . We will see in Sec. 3 that it requires a stronger form of each relation to enforce subequivalence into equivalence.
The following lemma shows that replacing either idempotent in e ≲ f e \lesssim f e ≲ f by an equivalent one preserves the subequivalence. In particular, ≲ \lesssim ≲ induces a well-defined preorder on the equivalence classes [ e ] [e] [ e ] of idempotents, given by [ e ] ≤ [ f ] [e] \leq[f] [ e ] ≤ [ f ] if e ≲ f e \lesssim f e ≲ f .
Lemma 1.4. For idempotents e , f , g ∈ Idem ( R ) e,f,g \in\operatorname{Idem}(R) e , f , g ∈ Idem ( R ) , the following hold.
e ≲ f ∼ g ⟹ e ≲ g e \lesssim f \sim g \Longrightarrow e \lesssim g e ≲ f ∼ g ⟹ e ≲ g
g ∼ e ≲ f ⟹ g ≲ f g \sim e \lesssim f \Longrightarrow g \lesssim f g ∼ e ≲ f ⟹ g ≲ f
Proof. (1) : Let e , f ∈ Idem ( R ) e,f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) such that
Then by definition, there exists an idempotent p ∈ Idem ( R ) p \in\operatorname{Idem}(R) p ∈ Idem ( R ) such that p ∼ e p \sim e p ∼ e and we have p ≤ f ⟺ p = f p = p f p \leq f \Longleftrightarrow p=fp=pf p ≤ f ⟺ p = f p = p f .
Now p ∼ e ⟹ p \sim e \Longrightarrow p ∼ e ⟹ there exists blocks a ∈ p R e a \in pRe a ∈ pR e and b ∈ e R p b \in eRp b ∈ e R p with
a b = p and b a = e ab=p \quad\text{and} \quad ba=e ab = p and ba = e Now f ∼ g ⟹ f \sim g \Longrightarrow f ∼ g ⟹ there exists blocks x ∈ f R g x \in fRg x ∈ f R g and y ∈ g R f y \in gRf y ∈ g R f with
x y = f and y x = g xy=f \quad\text{and} \quad yx=g x y = f and y x = g Now set u = b x ∈ e R g u=bx \in eRg u = b x ∈ e R g , v = y a ∈ g R e v=ya \in gRe v = y a ∈ g R e . Then,
u v = b ( x y ) a = b p ( x y ) p a = b ( p f p ) a = b a = e uv=b(xy)a=bp(xy)pa=b(pfp)a=ba=e uv = b ( x y ) a = b p ( x y ) p a = b ( p f p ) a = ba = e Therefore e ≲ g e \lesssim g e ≲ g .
(2) : The proof works similarly as above.
Block Invertibility and Generalized Inverses
Many generalized inverses are defined by a system of equations together with conditions on ranges and kernels. In the language of invertible blocks, where an invertible block is the ordinary notion of inverse read inside a skew corner, they become genuine two-sided inverses of a suitable compression of the element. We will first illustrate this case with Drazin’s ( b , c ) (b,c) ( b , c ) -inverse [10 ] for arbitrary b b b and c c c . Recall the following definition of ( b , c ) (b,c) ( b , c ) inverse of an element.
Definition 2.1 (Drazin [10 , 11 ] ). Let a , b , c ∈ R a,b,c \in R a , b , c ∈ R . An element y ∈ R y \in R y ∈ R is a ( b , c ) (b,c) ( b , c ) -inverse of a a a if
y ∈ b R y ∩ y R c , y a b = b , c a y = c . y \in bRy \cap yRc, \qquad yab=b, \qquad cay=c. y ∈ b R y ∩ y R c , y ab = b , c a y = c . Such a y y y is unique when it exists, and is usually denoted as a ( b , c ) a^{(b,c)} a ( b , c ) .
y y y is a left ( b , c ) (b,c) ( b , c ) -inverse of a a a if R y ⊆ R c Ry \subseteq Rc R y ⊆ R c and y a b = b yab=b y ab = b , and a right ( b , c ) (b,c) ( b , c ) -inverse of a a a if y R ⊆ b R yR \subseteq bR y R ⊆ b R and c a y = c cay=c c a y = c . If a a a is left and right ( b , c ) (b,c) ( b , c ) invertible then a a a is ( b , c ) (b,c) ( b , c ) invertible.
When a a a has a ( b , c ) (b,c) ( b , c ) inverse (say y y y ) then b R ≅ c R bR \cong cR b R ≅ c R (and R b ≅ R c Rb \cong Rc R b ≅ R c ). Here b , c b,c b , c is automatically von Neumann regular since we can write b = ( y ) a b = ( b u y ) a b b=(y)ab=(buy)ab b = ( y ) ab = ( b u y ) ab and c = c a ( y ) = c a ( y v c ) c=ca(y)=ca(yvc) c = c a ( y ) = c a ( y v c ) for some u , v ∈ R u,v \in R u , v ∈ R .
Hence, given a ∈ R a \in R a ∈ R we can determine from below what kind of values b , c ∈ R b,c \in R b , c ∈ R can be chosen to force a a a to be ( b , c ) (b,c) ( b , c ) invertible.
If R R R is an exchange ring, then
(i) a a a is right ( b , c ) (b,c) ( b , c ) -invertible if and only if c ∈ R y c \in Ry c ∈ R y for some y ∈ T ( a b ) y \in\mathrm{T}(ab) y ∈ T ( ab ) , and in this case ( b y ) (by) ( b y ) is a right ( b , c ) (b,c) ( b , c ) -inverse of a a a . (ii) a a a is left ( b , c ) (b,c) ( b , c ) -invertible if and only if b ∈ y R b \in yR b ∈ y R for some y ∈ T ( c a ) y \in T(ca) y ∈ T ( c a ) , in which case ( y c ) (yc) ( y c ) is a left ( b , c ) (b,c) ( b , c ) -inverse of a a a .
In the two-sided case not only are b , c b,c b , c von Neumann regular but also a b , c a ab,ca ab , c a as can be easily seen. We can then replace b R bR b R and R c Rc R c by e R eR e R and R f Rf R f with e , f ∈ Idem ( R ) e,f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) . We assume it in the following result in our block invertibility characterization of ( b , c ) (b,c) ( b , c ) invertibility.
Proposition 2.3. Let a , b , c ∈ R a,b,c \in R a , b , c ∈ R with b , c b,c b , c regular. Then there exists e , f ∈ Idem ( R ) e,f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) satisfying e R = b R eR=bR e R = b R and R f = R c Rf=Rc R f = R c . Then
(i) a a a is ( b , c ) (b,c) ( b , c ) -invertible if and only if f ∼ f a e e f \underset{fae}{\sim} e f f a e ∼ e ;
(ii) a a a is left ( b , c ) (b,c) ( b , c ) -invertible if and only if e ⪯ f a e f e \underset{fae}{\preceq} f e f a e ⪯ f ;
(iii) a a a is right ( b , c ) (b,c) ( b , c ) -invertible if and only if f ⪯ f a e e f \underset{fae}{\preceq} e f f a e ⪯ e .
Proof. (i) ( ⟹ ) (\Longrightarrow) ( ⟹ ) : Let y y y be the Bott-Duffin ( b , c ) (b,c) ( b , c ) inverses of a a a . Since e R = b R eR=bR e R = b R and R f = R c Rf=Rc R f = R c , set e = b r e=br e = b r and f = s c f=sc f = sc for r , s ∈ R r,s \in R r , s ∈ R . Then clearly, y a b = b ⟹ y a e = e yab=b \Longrightarrow yae=e y ab = b ⟹ y a e = e and c = c a y ⟹ f = f a y c=cay \Longrightarrow f=fay c = c a y ⟹ f = f a y . Also y ∈ b R y ∩ y R c ⟹ y = e y = y f ⟹ y ∈ e R f y \in bRy \cap yRc \Longrightarrow y=ey=yf \Longrightarrow y \in eRf y ∈ b R y ∩ y R c ⟹ y = ey = y f ⟹ y ∈ e R f .
Set θ = f a e \theta=fae θ = f a e and θ ′ = e y f \theta'=eyf θ ′ = ey f . Then f ∼ f a e e f \underset{fae}{\sim} e f f a e ∼ e since
θ θ ′ = ( f a e ) ( e y f ) = f a ( y ) = f and θ ′ θ = ( e y f ) ( f a e ) = y a e = e \theta\theta'=(fae)(eyf)=fa(y)=f \quad\text{and}\quad\theta'\theta=(eyf)(fae)=yae=e θ θ ′ = ( f a e ) ( ey f ) = f a ( y ) = f and θ ′ θ = ( ey f ) ( f a e ) = y a e = e
( ⟸ ) (\Longleftarrow) ( ⟸ ) : Let f ∼ f a e e f \underset{fae}{\sim} e f f a e ∼ e . Then there exists y = e y f ∈ e R f y=eyf \in eRf y = ey f ∈ e R f such that ( f a e ) y = f (fae)y=f ( f a e ) y = f and y ( f a e ) = e y(fae)=e y ( f a e ) = e . Then using y ( f a e ) y = y y(fae)y=y y ( f a e ) y = y and e R = b R eR=bR e R = b R and R f = R c Rf=Rc R f = R c gives y ∈ b R y ∩ y R c y \in bRy \cap yRc y ∈ b R y ∩ y R c . Set b = e r b=er b = er and c = w f c=wf c = w f . Then clearly b = ( e ) r = ( y a e ) r = y a b b=(e)r=(yae)r=yab b = ( e ) r = ( y a e ) r = y ab . Similarly, c = w f = c a y c=wf=cay c = w f = c a y . Hence y y y is a ( b , c ) (b,c) ( b , c ) inverse of a a a .
(ii) ( ⟹ ) (\Longrightarrow) ( ⟹ ) : Let y y y be the left ( b , c ) (b,c) ( b , c ) inverses of a a a . Then R y ⊆ R c = R f ⟹ y f = y Ry \subseteq Rc=Rf \Longrightarrow yf=y R y ⊆ R c = R f ⟹ y f = y . Also, y a b = b yab=b y ab = b and b R = e R bR=eR b R = e R gives y a e = e yae=e y a e = e .
Take θ = f a e \theta=fae θ = f a e and θ ′ = e y \theta'=ey θ ′ = ey . Then e ⪯ f a e f e \underset{fae}{\preceq} f e f a e ⪯ f
θ ′ θ = ( e y ) ( f a e ) = e y a e = e \theta'\theta=(ey)(fae)=eyae=e θ ′ θ = ( ey ) ( f a e ) = ey a e = e
( ⟸ ) (\Longleftarrow) ( ⟸ ) : Suppose e ⪯ f a e f e \underset{fae}{\preceq} f e f a e ⪯ f . Then there exists z ∈ e R f z \in eRf z ∈ e R f such that z ( f a e ) = e z(fae)=e z ( f a e ) = e . Using b R = e R bR=eR b R = e R we get z f a e = e ⟹ z a b = b zfae=e \Longrightarrow zab=b z f a e = e ⟹ z ab = b . Also z f = z ⟹ z ∈ R c zf=z \Longrightarrow z \in Rc z f = z ⟹ z ∈ R c . Thus a a a has a left ( b , c ) (b,c) ( b , c ) -invertible z z z .
(iii) ( ⟺ ) (\Longleftrightarrow) ( ⟺ ) : Similar to (ii). □
Remark 2.4. In the case when a a a is ( b , c ) (b,c) ( b , c ) invertible, the inverse y y y is also von Neumann regular and is an outer inverse of a a a since y ∈ b R y y \in bRy y ∈ b R y and y a b = b ⟹ y a y = y yab=b \Longrightarrow yay=y y ab = b ⟹ y a y = y . Furthermore, the idempotents y a ya y a and a y ay a y generate the right and left principal ideals b R bR b R and R c Rc R c respectively. This is because b = y a b ∈ ( y a ) R b=yab \in(ya)R b = y ab ∈ ( y a ) R and y a ∈ b R ( y a ) ⊆ b R ya \in bR(ya) \subseteq bR y a ∈ b R ( y a ) ⊆ b R . Same for R c Rc R c .
Thus we have the following proposition.
Proposition 2.5. Let R R R be a ring and fix arbitrary b , c ∈ R b,c \in R b , c ∈ R . Let a ∈ R a \in R a ∈ R . Then the following are equivalent:
(i) a a a is ( b , c ) (b,c) ( b , c ) -invertible
(ii) There exists e , f ∈ Idem ( R ) e,f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) with e R = b R eR=bR e R = b R and R f = R c Rf=Rc R f = R c such that f ∼ f a e e f \underset{fae}{\sim} e f f a e ∼ e
(iii) There exists y ∈ T ( a ) y \in T(a) y ∈ T ( a ) such that ( y a ) R = b R (ya)R=bR ( y a ) R = b R and R ( a y ) = R c R(ay)=Rc R ( a y ) = R c .
In this case the ( b , c ) (b,c) ( b , c ) inverse of a a a is indeed the inverse of ( f a e ) (fae) ( f a e ) denoted as ( f a e ) ′ ∈ e R f (fae)' \in eRf ( f a e ) ′ ∈ e R f .
Proof. Straightforward from Remark 2.4 and Proposition 2.3. For (iii) note that a y ∼ a y a y a ay \underset{aya}{\sim} ya a y a y a ∼ y a always holds for any y ∈ T ( a ) y \in T(a) y ∈ T ( a ) .
Many other well-known inverses of Von Neumann regular elements are particular instances of Drazin’s ( b , c ) (b,c) ( b , c ) inverses [10 ] . We note down the block-invertible equivalency of some of those inverses. Here I ( a ) I(a) I ( a ) denotes the set of all inner inverses of a a a .
Group inverse: The group inverse, if it exists, is the unique y ∈ I ( a ) y \in I(a) y ∈ I ( a ) with a y a = a aya=a a y a = a , y a y = y yay=y y a y = y and a y = y a ay=ya a y = y a . It is denoted by a # a^{\#} a # and is the ( a , a ) (a,a) ( a , a ) -inverse of a a a . Using Prop. 2.5 (iii), a # a^{\#} a # exists if and only if y a R = a R yaR=aR y a R = a R and R a y = R a Ray=Ra R a y = R a for some y ∈ T ( a ) y \in T(a) y ∈ T ( a ) . Since for every y ∈ I ( a ) y \in I(a) y ∈ I ( a ) we have ( a y ) R = a R (ay)R=aR ( a y ) R = a R and R ( y a ) = R a R(ya)=Ra R ( y a ) = R a , then by Prop. 2.5 (ii) a # a^{\#} a # exists if and only if y a ∼ ( y a ) a ( a y ) a y ya \mathrel{\overset{(ya)a(ay)}{\sim}} ay y a ∼ ( y a ) a ( a y ) a y for some y ∈ I ( a ) y \in I(a) y ∈ I ( a ) , and then a # = ( ( y a ) a ( a y ) ) ′ a^{\#}=((ya)a(ay))' a # = (( y a ) a ( a y ) ) ′ .
Moore-Penrose Inverse: In a ring with involution ∗ * ∗ , the Moore-Penrose inverse of a a a is the unique x x x with a x a = a axa=a a x a = a , x a x = x xax=x x a x = x , ( a x ) ∗ = a x (ax)^*=ax ( a x ) ∗ = a x and ( x a ) ∗ = x a (xa)^*=xa ( x a ) ∗ = x a . It is denoted as a † a^{\dagger} a † and is the ( a ∗ , a ∗ ) (a^*,a^*) ( a ∗ , a ∗ ) -inverse of a a a . Using Prop. 2.5 (ii) a † a^{\dagger} a † exists if and only if ( a y ) ∗ ∼ ( a y ) ∗ a ( y a ) ∗ ( y a ) ∗ (ay)^* \mathrel{\overset{(ay)^*a(ya)^*}{\sim}} (ya)^* ( a y ) ∗ ∼ ( a y ) ∗ a ( y a ) ∗ ( y a ) ∗ for some y ∈ I ( a ) y \in I(a) y ∈ I ( a ) , and a † = ( ( a y ) ∗ a ( y a ) ∗ ) ′ a^{\dagger}=((ay)^*a(ya)^*)' a † = (( a y ) ∗ a ( y a ) ∗ ) ′ .
Proposition 2.3 leaves open the question under what circumstances does when one-sided ( b , c ) (b,c) ( b , c ) -invertibility forces two-sided ( b , c ) (b,c) ( b , c ) -invertibility. Answering it requires a notion of stable range for skew corners, which we introduce in Section 3 and return to the question there. Here we examine in detail a special case of the ( b , c ) (b,c) ( b , c ) -inverse, Drazin’s Bott–Duffin inverse [10 ] . Recall that for idempotents e , f ∈ R e,f \in R e , f ∈ R , an element a ∈ R a \in R a ∈ R has a unique Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse if there exists a y ∈ R y \in R y ∈ R such that
(1) y = e y = y f y=ey=yf y = ey = y f
(2) y a e = e yae=e y a e = e and f a y = f fay=f f a y = f .
From the definition, it can be seen that for any a , b , c ∈ R a,b,c \in R a , b , c ∈ R , and any pair of idempotents e , f ∈ Idem ( R ) e,f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) such that b R = e R bR=eR b R = e R and R f = R c Rf=Rc R f = R c , the unique Drazin ( b , c ) (b,c) ( b , c ) inverse (if it exists) co-incides with the Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse. Therefore we can write for e , f ∈ R e,f \in R e , f ∈ R and a ∈ R a \in R a ∈ R : a a a has a Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse if and only if f ∼ f a e e f \mathrel{\overset{fae}{\sim}} e f ∼ f a e e .
In addition if ( 1 − e ) ∼ ( 1 − f ) (1-e)\sim(1-f) ( 1 − e ) ∼ ( 1 − f ) then the following holds.
Theorem 2.6. Let a ∈ R a \in R a ∈ R and idempotents e 2 = e e^2=e e 2 = e , f 2 = f f^2=f f 2 = f and let β \beta β be an invertible block in ( 1 − e ) R ( 1 − f ) (1-e)R(1-f) ( 1 − e ) R ( 1 − f ) with inverse β ′ \beta' β ′ . Then the following are equivalent.
(i) a a a has a Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse y y y .
(ii) f a e + β ′ fae+\beta' f a e + β ′ is invertible in R R R (equivalently f a + β ′ fa+\beta' f a + β ′ , or a e + β ′ ae+\beta' a e + β ′ ).
In this case, ( f a e + β ′ ) − 1 = e y f + β (fae+\beta')^{-1}=eyf+\beta ( f a e + β ′ ) − 1 = ey f + β .
Proof. (i) ⟹ \Longrightarrow ⟹ (ii): Since a a a has a Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse (say) y y y , we have f ∼ f a e e f \mathrel{\overset{fae}{\sim}} e f ∼ f a e e with ( f a e ) ′ = e y f (fae)'=eyf ( f a e ) ′ = ey f . Consider the associate pair ( f a e + β ′ , ) ( e y f + β ) (fae+\beta',)(eyf+\beta) ( f a e + β ′ , ) ( ey f + β ) . Then
( f a e + β ′ ) ( e y f + β ) = ( f a e ) ( e y f ) + ( f a e ) ( β ) + ( β ′ ) ( e y f ) + β ′ β = 1 (fae+\beta')(eyf+\beta)=(fae)(eyf)+(fae)(\beta)+(\beta')(eyf)+\beta'\beta=1 ( f a e + β ′ ) ( ey f + β ) = ( f a e ) ( ey f ) + ( f a e ) ( β ) + ( β ′ ) ( ey f ) + β ′ β = 1
Similarly, ( e y f + β ) ( f a e + β ′ ) = e + ( 1 − e ) = 1 (eyf+\beta)(fae+\beta')=e+(1-e)=1 ( ey f + β ) ( f a e + β ′ ) = e + ( 1 − e ) = 1 .
Also,
( f a e − β ′ ) ( e y f − β ) = ( f a e ) ( e y f ) − ( f a e ) ( β ) − ( β ′ ) ( e y f ) + β ′ β = 1 (fae-\beta')(eyf-\beta)=(fae)(eyf)-(fae)(\beta)-(\beta')(eyf)+\beta'\beta=1 ( f a e − β ′ ) ( ey f − β ) = ( f a e ) ( ey f ) − ( f a e ) ( β ) − ( β ′ ) ( ey f ) + β ′ β = 1
and similarly ( e y f − β ) ( f a e − β ′ ) = 1 (eyf-\beta)(fae-\beta')=1 ( ey f − β ) ( f a e − β ′ ) = 1 .
Thus f a e ± β ′ fae\pm\beta' f a e ± β ′ and e y f ± β eyf\pm\beta ey f ± β are invertible in R R R .
(ii) ⟹ \Longrightarrow ⟹ (i): Suppose ( f a e + β ′ ) ∈ U ( R ) (fae+\beta')\in U(R) ( f a e + β ′ ) ∈ U ( R ) . Set s = ( f a e + β ′ ) − 1 s=(fae+\beta')^{-1} s = ( f a e + β ′ ) − 1 and consider y = e ( s ) f y=e(s)f y = e ( s ) f . Then,
( f a e + β ′ ) ( s ) = 1 ⟹ f ( f a e + β ′ ) s = f ⟹ ( f a e ) s = f ⟹ ( f a e ) y = f (fae+\beta')(s)=1 \Longrightarrow f(fae+\beta')s=f \Longrightarrow(fae)s=f \Longrightarrow(fae)y=f ( f a e + β ′ ) ( s ) = 1 ⟹ f ( f a e + β ′ ) s = f ⟹ ( f a e ) s = f ⟹ ( f a e ) y = f
and
s ( f a e + β ′ ) = 1 ⟹ s ( f a e + β ′ ) e = e ⟹ s ( f a e ) e = e ⟹ y ( f a e ) = e s(fae+\beta')=1 \Longrightarrow s(fae+\beta')e=e \Longrightarrow s(fae)e=e \Longrightarrow y(fae)=e s ( f a e + β ′ ) = 1 ⟹ s ( f a e + β ′ ) e = e ⟹ s ( f a e ) e = e ⟹ y ( f a e ) = e
Thus, f a ~ a e f\widetilde{\phantom{a}}^{ae} f a a e and consequently, from above, we get a a a is Bott-Duffin ( e , f ) (e,f) ( e , f ) invertible.
Claim 1. f a e ± β ′ fae \pm\beta' f a e ± β ′ is invertible in R ⟺ f a ± β ′ R \Longleftrightarrow fa \pm\beta' R ⟺ f a ± β ′ is invertible in R R R .
Set η = y f a ( 1 − e ) ∈ e R ( 1 − e ) \eta=yfa(1-e)\in eR(1-e) η = y f a ( 1 − e ) ∈ e R ( 1 − e ) . Then η 2 = 0 ⟹ ( 1 + η ) ∈ U ( R ) \eta^2=0\Longrightarrow(1+\eta)\in U(R) η 2 = 0 ⟹ ( 1 + η ) ∈ U ( R ) . Then using β ′ ∈ ( 1 − f ) R ( 1 − e ) \beta'\in(1-f)R(1-e) β ′ ∈ ( 1 − f ) R ( 1 − e ) we get
( f a e + β ′ ) ( 1 + η ) = f a e + f a e η + β ′ + β ′ η = f a e + f a e ( y f a ( 1 − e ) ) + β ′ (fae+\beta')(1+\eta)=fae+fae\eta+\beta'+\beta'\eta=fae+fae(yfa(1-e))+\beta' ( f a e + β ′ ) ( 1 + η ) = f a e + f a eη + β ′ + β ′ η = f a e + f a e ( y f a ( 1 − e )) + β ′ = f a e + f a ( 1 − e ) + β ′ = f a + β ′ =fae+fa(1-e)+\beta'=fa+\beta' = f a e + f a ( 1 − e ) + β ′ = f a + β ′ Therefore, f a e + β ′ ∈ U ( R ) ⟺ f a + β ′ ∈ U ( R ) fae+\beta'\in U(R)\Longleftrightarrow fa+\beta'\in U(R) f a e + β ′ ∈ U ( R ) ⟺ f a + β ′ ∈ U ( R ) . Similarly, f a e − β ′ ∈ U ( R ) ⟺ f a − β ′ ∈ U ( R ) fae-\beta'\in U(R)\Longleftrightarrow fa-\beta'\in U(R) f a e − β ′ ∈ U ( R ) ⟺ f a − β ′ ∈ U ( R ) , using the same inverse ( 1 + η ) (1+\eta) ( 1 + η ) .
Claim 2. f a e ± β ′ fae\pm\beta' f a e ± β ′ is invertible in R ⟺ a e ± β ′ R\Longleftrightarrow ae\pm\beta' R ⟺ a e ± β ′ is invertible in R R R .
Taking γ = ( 1 − f ) a e y ∈ ( 1 − f ) R f \gamma=(1-f)aey\in(1-f)Rf γ = ( 1 − f ) a ey ∈ ( 1 − f ) R f and continuing similarly as above we get
( 1 + γ ) ( f a e ± β ′ ) = f a e + γ ( f a e ) ± β ′ ± γ β ′ = f a e + ( 1 − f ) a e y ( f a e ) ± β ′ (1+\gamma)(fae\pm\beta')=fae+\gamma(fae)\pm\beta'\pm\gamma\beta'=fae+(1-f)aey(fae)\pm\beta' ( 1 + γ ) ( f a e ± β ′ ) = f a e + γ ( f a e ) ± β ′ ± γ β ′ = f a e + ( 1 − f ) a ey ( f a e ) ± β ′ Hence our claim holds true.
The added consequence ( f a e ± β ′ ) − 1 = e y f + β (fae\pm\beta')^{-1}=eyf+\beta ( f a e ± β ′ ) − 1 = ey f + β is straightforward from the proof. □
As an immediate consequence, setting e = f e=f e = f generates and strengthens the following result from Drazin [10 ] .
Proposition 2.7. For any associative ring R R R with 1 1 1 and any a , e , y ∈ R a,e,y\in R a , e , y ∈ R with e 2 = e e^2=e e 2 = e , if a a a has a Bott-Duffin ( e , e ) (e,e) ( e , e ) inverse, then 1 − e + a e 1-e+ae 1 − e + a e is invertible, with ( 1 − e + a e ) − 1 = 1 − a y + y (1-e+ae)^{-1}=1-ay+y ( 1 − e + a e ) − 1 = 1 − a y + y and y = e ( 1 − e + a e ) − 1 y=e(1-e+ae)^{-1} y = e ( 1 − e + a e ) − 1 .
Dually, also 1 − e + e a 1-e+ea 1 − e + e a is invertible, with ( 1 − e + e a ) − 1 = 1 − y a + y (1-e+ea)^{-1}=1-ya+y ( 1 − e + e a ) − 1 = 1 − y a + y , and y = ( 1 − e + e a ) − 1 e y=(1-e+ea)^{-1}e y = ( 1 − e + e a ) − 1 e .
Proof. Taking β = β ′ = ( 1 − e ) ∈ U ( ( 1 − e ) R ( 1 − e ) ) \beta=\beta'=(1-e)\in U((1-e)R(1-e)) β = β ′ = ( 1 − e ) ∈ U (( 1 − e ) R ( 1 − e )) in Theorem 2.6 , we obtain e a ± ( 1 − e ) , a e ± ( 1 − e ) ∈ U ( R ) ea\pm(1-e), ae\pm(1-e)\in U(R) e a ± ( 1 − e ) , a e ± ( 1 − e ) ∈ U ( R ) . Additionally, if a a a has a Bott-Duffin ( e , e ) (e,e) ( e , e ) inverse, say y y y in the skew corner e R e eRe e R e , then we have y = e y = y e y=ey=ye y = ey = y e ; y a e = e = e a y yae=e=eay y a e = e = e a y . Then from Theorem 2.6 Claim 1, for η = y e a ( 1 − e ) = y a − e \eta=yea(1-e)=ya-e η = y e a ( 1 − e ) = y a − e
( e a ± ( 1 − e ) ) = ( e a e ± ( 1 − e ) ) ( 1 + η ) (ea\pm(1-e))=(eae\pm(1-e))(1+\eta) ( e a ± ( 1 − e )) = ( e a e ± ( 1 − e )) ( 1 + η )
⟹ ( e a ± ( 1 − e ) ) − 1 = ( 1 − η ) ( e a e ± ( 1 − e ) ) − 1 = ( 1 − η ) ( e y e ± ( 1 − e ) ) = e y e ± ( 1 − e ) ∓ η = y ± ( 1 − e ) ∓ ( y a − e ) = y ± ( 1 − y a ) \begin{aligned}
\Longrightarrow(ea\pm(1-e))^{-1}=(1-\eta)(eae\pm(1-e))^{-1} \\
&=(1-\eta)(eye\pm(1-e)) \\
&=eye\pm(1-e)\mp\eta\\
&=y\pm(1-e)\mp(ya-e)=y\pm(1-ya)
\end{aligned} ⟹ ( e a ± ( 1 − e ) ) − 1 = ( 1 − η ) ( e a e ± ( 1 − e ) ) − 1 = ( 1 − η ) ( ey e ± ( 1 − e )) = ey e ± ( 1 − e ) ∓ η = y ± ( 1 − e ) ∓ ( y a − e ) = y ± ( 1 − y a )
Similarly, using Claim 2, ( a e ± ( 1 − e ) ) − 1 = y ± ( 1 − a y ) (ae\pm(1-e))^{-1}=y\pm(1-ay) ( a e ± ( 1 − e ) ) − 1 = y ± ( 1 − a y ) . □
Applying left or right ( b , c ) (b,c) ( b , c ) invertibility on idempotents e , f e,f e , f where b R = e R bR=eR b R = e R and R c = R f Rc=Rf R c = R f allows us to define the concepts of right or left Bott-Duffin inverse of an element a a a .
Definition 2.8. Let e , f ∈ R e,f\in R e , f ∈ R be idempotents and let a ∈ R a\in R a ∈ R . An element y ∈ e R f y\in eRf y ∈ e R f is called
(i) a right Bott–Duffin ( e , f ) (e,f) ( e , f ) -inverse of a a a if f a y = f fay=f f a y = f so that f ≲ y e f\lesssim^{y}e f ≲ y e
(ii) a left Bott–Duffin ( e , f ) (e,f) ( e , f ) -inverse of a a a if y a e = e yae=e y a e = e so that e ≲ y f e\lesssim^{y}f e ≲ y f
If y 1 y_1 y 1 is a left, and y 2 y_2 y 2 a right Bott–Duffin ( e , f ) (e,f) ( e , f ) -inverse of a a a resp., then y 1 = y 1 f = y 1 ( f a y 2 ) = ( y 1 a e ) y 2 = e y 2 = y 2 y_1=y_1f=y_1(fay_2)=(y_1ae)y_2=ey_2=y_2 y 1 = y 1 f = y 1 ( f a y 2 ) = ( y 1 a e ) y 2 = e y 2 = y 2 and this common element is the Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse of a a a . Moreover, if y ∈ e R f y\in eRf y ∈ e R f is either a right or left Bott–Duffin ( e , f ) (e,f) ( e , f ) -inverse of a a a , then y y y is an inner inverse of f a e fae f a e and a a a is an inner inverse of y y y ; that is, ( f a e ) y ( f a e ) = f a e (fae)y(fae)=fae ( f a e ) y ( f a e ) = f a e and y a y = y yay=y y a y = y . This connects one-sided Bott–Duffin invertibility with von Neumann regularity as we see below.
Proposition 2.9. Let R R R be a ring and a ∈ R a\in R a ∈ R . Then the following are equivalent:
(i) a a a is von Neumann regular
(ii) There exists idempotents e , f ∈ R e,f\in R e , f ∈ R such that a ∈ f R e a\in fRe a ∈ f R e and a a a has a Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse.
(iii) There exist idempotents e , f ∈ R e,f\in R e , f ∈ R with a ∈ f R e a\in fRe a ∈ f R e such that a a a has a right Bott–Duffin inverse.
(iv) There exist idempotents e , f ∈ R e,f\in R e , f ∈ R with a ∈ f R e a\in fRe a ∈ f R e such that a a a has a left Bott–Duffin inverse.
Proof. (iii) ⟹ \Longrightarrow ⟹ (i): Suppose there exists idempotents e , f ∈ R e,f\in R e , f ∈ R such that a = f a e ∈ f R e a=fae\in fRe a = f a e ∈ f R e and a a a has a right Bott-Duffin ( e , f ) (e,f) ( e , f ) inverse y y y . Then by definition, y = e y = y f y=ey=yf y = ey = y f , f a y = f fay=f f a y = f . Then we have
f = f a y = ( f a e ) ( y ) = a y and f ( a e ) = f a e = a ⟹ a R = f R f=fay=(fae)(y)=ay \quad\text{and}\quad f(ae)=fae=a \Longrightarrow aR=fR f = f a y = ( f a e ) ( y ) = a y and f ( a e ) = f a e = a ⟹ a R = f R
Thus a a a is von Neumann regular.
(i) ⟹ \Longrightarrow ⟹ (ii) Conversely, let a a a be von Neumann regular. Then there exists x ∈ R x\in R x ∈ R such that a = a x a a=axa a = a x a . Take y = x a x y=xax y = x a x , e = x a e=xa e = x a and f = a x f=ax f = a x . Then a = ( a x ) ( a ) ( x a ) = f a e ∈ f R e a=(ax)(a)(xa)=fae\in fRe a = ( a x ) ( a ) ( x a ) = f a e ∈ f R e and x a x = x a ( x ) a x = e x f ∈ e R f xax=xa(x)ax=exf\in eRf x a x = x a ( x ) a x = e x f ∈ e R f and we have a x ∼ a x a ax\overset{a}{\sim}xa a x ∼ a x a . Thus a ∈ f R e a\in fRe a ∈ f R e and a a a has a Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse ( a ∼ ) = x a x (a^{\sim})=xax ( a ∼ ) = x a x .
(ii) ⟹ \Longrightarrow ⟹ (iii): Straightforward.
(iv) ⟹ \Longrightarrow ⟹ (i): Similarly as above. □
Remark 2.10. We see that for idempotents e , f ∈ R e,f\in R e , f ∈ R if an element a a a is anchored in the skew-corner f R e fRe f R e , i.e. a = f a e a=fae a = f a e , then every left (resp. right) Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse of a a a is Bott–Duffin invertible for some ( e , f ) (e,f) ( e , f ) .
In the above characterization, observe that if there exists idempotents e , f ∈ R e,f\in R e , f ∈ R such that a ∈ f R e a\in fRe a ∈ f R e and a a a has a Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse, say y ∈ e R f y\in eRf y ∈ e R f , then
a y ≡ 1 ( m o d R ( 1 − f ) ) and y a ≡ 1 ( m o d ( 1 − e ) R ) ay\equiv1 \pmod{R(1-f)} \quad\text{and}\quad ya\equiv1 \pmod{(1-e)R} a y ≡ 1 ( mod R ( 1 − f )) and y a ≡ 1 ( mod ( 1 − e ) R )
If every left unit lifts modulo every left ideal, which is a characterization of stable range one in [36 ] forces a a a to be unit-regular. Indeed, for u ∈ U ( R ) u\in U(R) u ∈ U ( R ) we have e u f = y euf=y e u f = y if and only if e ( u − y ) f = 0 e(u-y)f=0 e ( u − y ) f = 0 , and since ∀ y ∈ R \forall y\in R ∀ y ∈ R , y − e y f = e y ( 1 − f ) + ( 1 − e ) y y-eyf=ey(1-f)+(1-e)y y − ey f = ey ( 1 − f ) + ( 1 − e ) y , therefore we get the following characterization for unit-regularity in terms of Bott–Duffin ( e , f ) (e,f) ( e , f ) inverse.
Proposition 2.11. Let R R R be a ring and a ∈ R a\in R a ∈ R . Then a a a is unit-regular if and only if there exist idempotents e , f ∈ R e,f\in R e , f ∈ R and a unit u ∈ U ( R ) u\in U(R) u ∈ U ( R ) such that a ∈ f R e a\in fRe a ∈ f R e and e u f euf e u f is a Bott–Duffin ( e , f ) (e,f) ( e , f ) -inverse of $a.
The characterization of unit regularity obtained above is an element-wise statement: it detects, block by block, when a regular element admits a unit inner inverse. The classical theory of unit regularity, going back to Ehrlich and Handelman, is phrased at the level of idempotents, in terms of the equivalence e ∼ f e\sim f e ∼ f . We showed earlier that equivalence of idempotents is itself a statement about block inverses: e ∼ x f e\overset{x}{\sim}f e ∼ x f says precisely when the skew corner e R f eRf e R f contains an invertible block x x x . Using Block generalized inverses, we provide a short proof to Ehrlich [12 ] and Handelman's [14 ] classical result on equivalence of idempotents.
Theorem 2.12 ([12 , 14 ] ). Every von Neumann regular element in a ring R R R is unit-regular if and only if for every two idempotents e , f ∈ R e,f\in R e , f ∈ R , e ∼ f ⟹ ( 1 − e ) ∼ ( 1 − f ) e\sim f\Longrightarrow(1-e)\sim(1-f) e ∼ f ⟹ ( 1 − e ) ∼ ( 1 − f ) .
Proof. ⟸ \Longleftarrow ⟸ : Let a ∈ R a\in R a ∈ R be regular and set x ∈ Ref ( a ) x\in\operatorname{Ref}(a) x ∈ Ref ( a ) . Then choosing f = a x f=ax f = a x and e = x a e=xa e = x a , from Prop. 2.9 we obtain x x x to be the Bott–Duffin ( e , f ) (e,f) ( e , f ) -inverse of a a a and f ∼ f a e e f\overset{fae}{\sim}e f ∼ f a e e . Now f ∼ e ⟹ ( 1 − f ) ∼ ( 1 − e ) f\sim e\Longrightarrow(1-f)\sim(1-e) f ∼ e ⟹ ( 1 − f ) ∼ ( 1 − e ) . Then there exists invertible block β ′ \beta' β ′ in ( 1 − f ) R ( 1 − e ) (1-f)R(1-e) ( 1 − f ) R ( 1 − e ) with inverse β ∈ ( 1 − e ) R ( 1 − f ) \beta\in(1-e)R(1-f) β ∈ ( 1 − e ) R ( 1 − f ) , and subsequently by Theorem 2.6 , f a e + β ′ = a + β ′ ∈ U ( R ) fae+\beta'=a+\beta'\in U(R) f a e + β ′ = a + β ′ ∈ U ( R ) with inverse x + β x+\beta x + β . Set v = ( x + β ) v=(x+\beta) v = ( x + β ) . Then
a v a = a ( x + β ) a = a x a + a β a = a + ( f a e ) β a = a ava=a(x+\beta)a=axa+a\beta a=a+(fae)\beta a=a a v a = a ( x + β ) a = a x a + a β a = a + ( f a e ) β a = a
⟹ \Longrightarrow ⟹ : Suppose f ∼ b e f \mathrel{\sim_b} e f ∼ b e for idempotents e , f ∈ R e,f\in R e , f ∈ R with b ′ = x b'=x b ′ = x . We prove that ( 1 − f ) ∼ ( 1 − e ) (1-f)\sim(1-e) ( 1 − f ) ∼ ( 1 − e ) . Then clearly b x = f bx=f b x = f , x b = e xb=e x b = e , and b x b = b bxb=b b x b = b and therefore by assumption, there exists u ∈ U ( R ) u\in U(R) u ∈ U ( R ) such that b = b u b b=bub b = b u b . Label f 1 = b u f_1=bu f 1 = b u and e 1 = u b e_1=ub e 1 = u b . Then both e 1 , f 1 ∈ Idem ( R ) e_1,f_1\in\operatorname{Idem}(R) e 1 , f 1 ∈ Idem ( R ) . Consider the elements γ = ( 1 − e 1 ) u ( 1 − f 1 ) \gamma=(1-e_1)u(1-f_1) γ = ( 1 − e 1 ) u ( 1 − f 1 ) and γ ′ = ( 1 − f 1 ) u − 1 ( 1 − e 1 ) \gamma'=(1-f_1)u^{-1}(1-e_1) γ ′ = ( 1 − f 1 ) u − 1 ( 1 − e 1 ) . Then it can be seen that
( 1 − e ) ∼ ( 1 − e 1 ) ( 1 − e 1 ) ∼ ( 1 − f 1 ) ∼ 1 − f 1 ( 1 − f ) (1-e)\mathrel{\sim^{(1-e_1)}}(1-e_1)\mathrel{\sim}(1-f_1)\mathrel{\sim^{1-f_1}}(1-f) ( 1 − e ) ∼ ( 1 − e 1 ) ( 1 − e 1 ) ∼ ( 1 − f 1 ) ∼ 1 − f 1 ( 1 − f )
Now, in the proof of Proposition 2.9 on a characterization of von Neumann regularity, we observed when a subequivalence ≲ \lesssim ≲ improves to an equivalence ∼ \sim ∼ . It is therefore natural to ask what becomes of Theorem 2.12 when the hypothesis e ∼ f e\sim f e ∼ f is weakened to e ≲ f e\lesssim f e ≲ f .
Observe first that the conclusion must reverse direction: That is, if for idempotents e , f ∈ R e,f\in R e , f ∈ R we have e ≤ f e\leq f e ≤ f then automatically, 1 − f ≤ 1 − e 1-f\leq1-e 1 − f ≤ 1 − e . Therefore our complement condition compatible with ≲ \lesssim ≲ is 1 − f ≲ 1 − e 1-f\lesssim1-e 1 − f ≲ 1 − e . Observe that such a ring R R R where for arbitrary e , f ∈ R e,f\in R e , f ∈ R , e ≲ f ⟹ ( 1 − f ) ≲ ( 1 − e ) e\lesssim f\Longrightarrow(1-f)\lesssim(1-e) e ≲ f ⟹ ( 1 − f ) ≲ ( 1 − e ) , is always directly finite. This is because whenever p q = 1 pq=1 pq = 1 for some p , q ∈ R p,q\in R p , q ∈ R then 1 ∼ q p 1\mathrel{\sim}qp 1 ∼ q p which gives ( 1 − q p ) ≲ 0 (1-qp)\lesssim0 ( 1 − q p ) ≲ 0 forcing q p = 1 qp=1 q p = 1 .
Surprisingly, the weakened hypothesis loses none of its strength as we see below.
Theorem 2.13. Every von Neumann regular element in a ring R R R is unit-regular if and only if for arbitrary idempotents e , f ∈ R e,f\in R e , f ∈ R , e ≲ f ⟹ ( 1 − f ) ≲ ( 1 − e ) e\lesssim f\Longrightarrow(1-f)\lesssim(1-e) e ≲ f ⟹ ( 1 − f ) ≲ ( 1 − e ) .
Proof. ⟹ \Longrightarrow ⟹ : Let every von Neumann regular element in R R R be unit-regular. Choose arbitrary idempotents e , f ∈ Idem ( R ) e,f\in\operatorname{Idem}(R) e , f ∈ Idem ( R ) such that e ≲ f e\lesssim f e ≲ f . Then by definition, there exists an idempotent p ∈ Idem ( R ) p\in\operatorname{Idem}(R) p ∈ Idem ( R ) such that e ∼ p e\sim p e ∼ p and p ≤ f p\leq f p ≤ f . Now by Theorem 2.12 , p ∼ e ⟹ ( 1 − p ) ∼ ( 1 − e ) p\sim e\Longrightarrow(1-p)\sim(1-e) p ∼ e ⟹ ( 1 − p ) ∼ ( 1 − e ) . Also, since p ≤ f p\leq f p ≤ f then by definition, p = f p = p f p=fp=pf p = f p = p f . Therefore, ( 1 − p ) ( 1 − f ) = 1 − p − f + p f = ( 1 − f ) (1-p)(1-f)=1-p-f+pf=(1-f) ( 1 − p ) ( 1 − f ) = 1 − p − f + p f = ( 1 − f ) and ( 1 − f ) ( 1 − p ) = 1 − p − f + f p = ( 1 − f ) ⟹ ( 1 − f ) ≲ ( 1 − p ) (1-f)(1-p)=1-p-f+fp=(1-f)\Longrightarrow(1-f)\lesssim(1-p) ( 1 − f ) ( 1 − p ) = 1 − p − f + f p = ( 1 − f ) ⟹ ( 1 − f ) ≲ ( 1 − p ) . Then ( 1 − f ) ≲ ( 1 − p ) (1-f)\lesssim(1-p) ( 1 − f ) ≲ ( 1 − p ) . Thus by Lemma 1.4 (1), ( 1 − f ) ≲ ( 1 − p ) ∼ ( 1 − e ) ⟹ ( 1 − f ) ≲ ( 1 − e ) (1-f)\lesssim(1-p)\sim(1-e)\Longrightarrow(1-f)\lesssim(1-e) ( 1 − f ) ≲ ( 1 − p ) ∼ ( 1 − e ) ⟹ ( 1 − f ) ≲ ( 1 − e ) .
Conversely, let a ∈ R a\in R a ∈ R be regular and set x ∈ Ref ( a ) x\in\operatorname{Ref}(a) x ∈ Ref ( a ) . Then from Prop. 2.9 (iii) choosing f = a x f=ax f = a x and e = x a e=xa e = x a we obtain f a e ≲ e fae\lesssim e f a e ≲ e where a = f a e a=fae a = f a e and x ∈ e R f x\in eRf x ∈ e R f .
Then by hypothesis, ( 1 − e ) ≲ r ( 1 − f ) (1-e)\mathrel{\lesssim^r}(1-f) ( 1 − e ) ≲ r ( 1 − f ) where r ∈ ( 1 − e ) R ( 1 − f ) r\in(1-e)R(1-f) r ∈ ( 1 − e ) R ( 1 − f ) with r ′ ∈ ( 1 − f ) R ( 1 − e ) r'\in(1-f)R(1-e) r ′ ∈ ( 1 − f ) R ( 1 − e ) such that r r ′ = 1 − e rr'=1-e r r ′ = 1 − e . Then ( x + r ) (x+r) ( x + r ) is right-invertible in R R R since ( x + r ) ( a + r ′ ) = e + ( 1 − e ) = 1 (x+r)(a+r')=e+(1-e)=1 ( x + r ) ( a + r ′ ) = e + ( 1 − e ) = 1 and therefore using direct finiteness as shown earlier, x + r ∈ U ( R ) x+r\in U(R) x + r ∈ U ( R ) . Label u = x + r u=x+r u = x + r . Then a u a = a ( x + r ) a = a x a = a aua=a(x+r)a=axa=a a u a = a ( x + r ) a = a x a = a .
In both global characterizations of unit-regularity above, the implication required is for every pair of idempotents of R R R . However it is of interest to know when such an equivalence holds true locally for fixed arbitrary idempotents e , f e,f e , f . The relations e ∼ f e\sim f e ∼ f and ( 1 − e ) ∼ ( 1 − f ) (1-e)\sim(1-f) ( 1 − e ) ∼ ( 1 − f ) say precisely that the complementary skew corners f R e fRe f R e and ( 1 − f ) R ( 1 − e ) (1-f)R(1-e) ( 1 − f ) R ( 1 − e ) contain invertible blocks. In the next couple of results we approach the question through (one-sided) block inverses in their respective skew specific corners.
We begin with what the two relations give when both hold. The Lemma below allows us to manufactures units in situations where a direct computation becomes very technical. The mechanism is the ring-theoretic counterpart of block inversion in matrices. Compressing x ∈ R x\in R x ∈ R on the left by idempotents p , 1 − p p,1-p p , 1 − p and on the right by idempotents e , 1 − e e,1-e e , 1 − e produces the two-sided Peirce decomposition
x = p x e + p x ( 1 − e ) + ( 1 − p ) x e + ( 1 − p ) x ( 1 − e ) , x = pxe + px(1-e) + (1-p)xe + (1-p)x(1-e), x = p x e + p x ( 1 − e ) + ( 1 − p ) x e + ( 1 − p ) x ( 1 − e ) ,
which we can represent in matrix form
x ≡ ( α η γ β ) x \equiv\begin{pmatrix} \alpha& \eta\\ \gamma& \beta\end{pmatrix} x ≡ ( α γ η β )
The two diagonal entries lie in different skew corners, so invertible is read as an invertible block in the corresponding skew corner. This is the degenerate case of Banachiewicz’s [4 ] full inversion formula for a partitioned matrix where either of the off-diagonal elements B , C B,C B , C are zero. In a matrix M = ( A C B D ) M = \begin{pmatrix} A & C \\ B & D \end{pmatrix} M = ( A B C D ) , if A A A is invertible and either of B B B or C C C is zero, then M M M is invertible if and only if the Schur complement S = D − B A − 1 C = D S = D - BA^{-1}C = D S = D − B A − 1 C = D is, in which case
M − 1 = ( A − 1 − A − 1 C S − 1 − S − 1 B A − 1 D − 1 ) . M^{-1} =
\begin{pmatrix}
A^{-1} & -A^{-1}CS^{-1} \\
-S^{-1}BA^{-1} & D^{-1}
\end{pmatrix}. M − 1 = ( A − 1 − S − 1 B A − 1 − A − 1 C S − 1 D − 1 ) .
where either − A − 1 C S − 1 = 0 -A^{-1}CS^{-1} = 0 − A − 1 C S − 1 = 0 or − S − 1 B A − 1 = 0 -S^{-1}BA^{-1} = 0 − S − 1 B A − 1 = 0 , and where A A A and D D D are invertible matrices of different sizes. Thus, in Peirce form, if α \alpha α is an invertible block of p R e pRe pR e with inverse α ′ ∈ e R p \alpha' \in eRp α ′ ∈ e R p , and the Schur complement S = β ∈ ( 1 − p ) R ( 1 − e ) S = \beta\in(1-p)R(1-e) S = β ∈ ( 1 − p ) R ( 1 − e ) is an invertible block with inverse β ′ ∈ ( 1 − e ) R ( 1 − p ) \beta' \in(1-e)R(1-p) β ′ ∈ ( 1 − e ) R ( 1 − p ) , then x ∈ U ( R ) x \in U(R) x ∈ U ( R ) whenever γ = 0 \gamma= 0 γ = 0 or η = 0 \eta= 0 η = 0 , and
x − 1 = α ′ − α ′ η β ′ − β ′ γ α ′ + β ′ . x^{-1} = \alpha' - \alpha'\eta\beta' - \beta'\gamma\alpha' + \beta'. x − 1 = α ′ − α ′ η β ′ − β ′ γ α ′ + β ′ .
We shall refer to the following lemma as the Block Lemma throughout this article.
Lemma 2.14. Let p , e p,e p , e be arbitrary idempotents and p ∼ α e p \mathrel{\overset{\alpha}{\sim}} e p ∼ α e and ( 1 − p ) ∼ β ( 1 − e ) (1-p) \mathrel{\overset{\beta}{\sim}} (1-e) ( 1 − p ) ∼ β ( 1 − e ) . Then the following results hold true.
(i) α + γ + β ∈ U ( R ) \alpha+ \gamma+ \beta\in U(R) α + γ + β ∈ U ( R ) for every γ ∈ ( 1 − p ) R e \gamma\in(1-p)Re γ ∈ ( 1 − p ) R e .
(ii) α + η + β ∈ U ( R ) \alpha+ \eta+ \beta\in U(R) α + η + β ∈ U ( R ) for every η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) .
Proof. (i) Let ( α , α ′ ) (\alpha,\alpha') ( α , α ′ ) be an invertible block in ( p R e , e R p ) (pRe,eRp) ( pR e , e R p ) and ( β , β ′ ) (\beta,\beta') ( β , β ′ ) be an invertible block in ( 1 − p ) R ( 1 − e ) , ( 1 − e ) R ( 1 − p ) (1-p)R(1-e),(1-e)R(1-p) ( 1 − p ) R ( 1 − e ) , ( 1 − e ) R ( 1 − p ) . Then for any γ ∈ ( 1 − p ) R e \gamma\in(1-p)Re γ ∈ ( 1 − p ) R e the cross products vanish
α β ′ = 0 = β α ′ ; α ′ β = 0 = β ′ α ; α ′ γ = 0 = γ β ′ \alpha\beta' = 0 = \beta\alpha';\quad\alpha'\beta= 0 = \beta'\alpha;\quad\alpha'\gamma= 0 = \gamma\beta' α β ′ = 0 = β α ′ ; α ′ β = 0 = β ′ α ; α ′ γ = 0 = γ β ′
Set d = ( α + γ + β ) d = (\alpha+ \gamma+ \beta) d = ( α + γ + β ) . Then, since γ = ( 1 − p ) γ \gamma= (1-p)\gamma γ = ( 1 − p ) γ we have
d ( α ′ + β ′ − β ′ γ α ′ ) = α α ′ + γ α ′ + β β ′ − α β ′ γ α ′ ⏟ 0 − γ β ′ γ α ′ ⏟ 0 − β β ′ γ α ′ = p + γ α ′ + ( 1 − p ) − ( 1 − p ) γ α ′ = p + ( 1 − p ) + [ γ α ′ − γ α ′ ] = 1 \begin{aligned}
d(\alpha' + \beta' - \beta'\gamma\alpha') = \alpha\alpha' + \gamma\alpha' + \beta\beta' - \underbrace{\alpha\beta'\gamma\alpha'}_{0} - \underbrace{\gamma\beta'\gamma\alpha'}_{0} - \beta\beta'\gamma\alpha' \\
&= p + \gamma\alpha' + (1-p) - (1-p)\gamma\alpha' \\
&= p + (1-p) + [\gamma\alpha' - \gamma\alpha'] = 1
\end{aligned} d ( α ′ + β ′ − β ′ γ α ′ ) = α α ′ + γ α ′ + β β ′ − 0 α β ′ γ α ′ − 0 γ β ′ γ α ′ − β β ′ γ α ′ = p + γ α ′ + ( 1 − p ) − ( 1 − p ) γ α ′ = p + ( 1 − p ) + [ γ α ′ − γ α ′ ] = 1
On the other hand, since γ = γ e \gamma= \gamma e γ = γ e we also get
( α ′ + β ′ − β ′ γ α ′ ) d = α ′ α + β ′ γ + β ′ β − β ′ γ ( α ′ α ) − β ′ γ α ′ γ ⏟ 0 − β ′ γ α ′ β ⏟ 0 = e + β ′ γ + ( 1 − e ) − β ′ γ ( e ) = 1 \begin{aligned}
(\alpha' + \beta' - \beta'\gamma\alpha')d = \alpha'\alpha+ \beta'\gamma+ \beta'\beta- \beta'\gamma(\alpha'\alpha) - \underbrace{\beta'\gamma\alpha'\gamma}_{0} - \underbrace{\beta'\gamma\alpha'\beta}_{0} \\
&= e + \beta'\gamma+ (1-e) - \beta'\gamma(e) = 1
\end{aligned} ( α ′ + β ′ − β ′ γ α ′ ) d = α ′ α + β ′ γ + β ′ β − β ′ γ ( α ′ α ) − 0 β ′ γ α ′ γ − 0 β ′ γ α ′ β = e + β ′ γ + ( 1 − e ) − β ′ γ ( e ) = 1
Thus, d = α + γ + β ∈ U ( R ) d = \alpha+ \gamma+ \beta\in U(R) d = α + γ + β ∈ U ( R ) for every γ ∈ ( 1 − p ) R e \gamma\in(1-p)Re γ ∈ ( 1 − p ) R e .
(ii) The proof of (ii) follows similarly as above. □
As an immediate consequence for γ = 0 \gamma= 0 γ = 0 , if p ∼ α e p \mathrel{\overset{\alpha}{\sim}} e p ∼ α e and ( 1 − p ) ∼ β ( 1 − e ) (1-p) \mathrel{\overset{\beta}{\sim}} (1-e) ( 1 − p ) ∼ β ( 1 − e ) then α + β ∈ U \alpha+ \beta\in U α + β ∈ U with ( α + β ) − 1 = α ′ + β ′ (\alpha+ \beta)^{-1} = \alpha' + \beta' ( α + β ) − 1 = α ′ + β ′ . In the diagonal case where p = e p = e p = e , we have α ∈ p R p \alpha\in pRp α ∈ pR p and β ∈ ( 1 − p ) R ( 1 − p ) \beta\in(1-p)R(1-p) β ∈ ( 1 − p ) R ( 1 − p ) , then α + β ∈ U \alpha+ \beta\in U α + β ∈ U if and only if α ∈ U ( p R p ) \alpha\in U(pRp) α ∈ U ( pR p ) and β ∈ U ( ( 1 − p ) R ( 1 − p ) ) \beta\in U((1-p)R(1-p)) β ∈ U (( 1 − p ) R ( 1 − p )) .
Note that Blocks Lemma 2.14 above is a generalization of Lemma 6.1 in [21 ] Khurana and Lam.
Lemma 2.15 ([21 ] Khurana and Lam). Let e , f ∈ idem ( R ) e,f \in\operatorname{idem}(R) e , f ∈ idem ( R ) be such that e + f = 1 e+f=1 e + f = 1 , and let x ∈ U ( e R e ) x \in U(eRe) x ∈ U ( e R e ) , y ∈ U ( f R f ) y \in U(fRf) y ∈ U ( f R f ) . Then x + p + y ∈ U ( R ) x+p+y \in U(R) x + p + y ∈ U ( R ) for every p ∈ f R e p \in fRe p ∈ f R e .
It is natural to ask to what extent the converse holds: if such an x x x is a unit, must the diagonal blocks α \alpha α and β \beta β themselves admit block inverses? Observe that for some distinct p , e ∈ Idem ( R ) p,e \in\operatorname{Idem}(R) p , e ∈ Idem ( R ) , whenever x = α + η + β ∈ U ( R ) x=\alpha+\eta+\beta\in U(R) x = α + η + β ∈ U ( R ) where α ∈ p R e \alpha\in pRe α ∈ pR e , η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) , and β ∈ ( 1 − p ) R ( 1 − e ) \beta\in(1-p)R(1-e) β ∈ ( 1 − p ) R ( 1 − e ) , then α ∈ p R e \alpha\in pRe α ∈ pR e is automatically left block-invertible and β \beta β is right block invertible in their respective skew-corners since ( e x − 1 p ) α = e (ex^{-1}p)\alpha=e ( e x − 1 p ) α = e and β ( ( 1 − e ) x − 1 ( 1 − p ) ) = 1 − p \beta((1-e)x^{-1}(1-p))=1-p β (( 1 − e ) x − 1 ( 1 − p )) = 1 − p .
In general the blocks α \alpha α or β \beta β are not both-sided invertible in their respective corners, and the unit condition alone does not force the invertibility of either of those blocks. What it does force is an association between the two: α \alpha α now admits a right block inverse precisely when β \beta β is left-block invertible, and vice versa. In other words p ∼ α e ⟺ ( 1 − p ) ∼ β ( 1 − e ) p \mathrel{\mathop{\sim}\limits^{\alpha}} e \Longleftrightarrow(1-p) \mathrel{\mathop{\sim}\limits^{\beta}} (1-e) p ∼ α e ⟺ ( 1 − p ) ∼ β ( 1 − e ) . The following lemma makes this relationship precise.
Proposition 2.16. Let α ∈ p R e \alpha\in pRe α ∈ pR e and β ∈ ( 1 − p ) R ( 1 − e ) \beta\in(1-p)R(1-e) β ∈ ( 1 − p ) R ( 1 − e ) . Then the following hold.
If for some η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) , we get α + η + β ∈ U ( R ) \alpha+\eta+\beta\in U(R) α + η + β ∈ U ( R ) then
p ≲ α e ⟺ ( 1 − e ) ≲ β ( 1 − p ) p \mathrel{\mathop{\lesssim}\limits^{\alpha}} e \Longleftrightarrow(1-e) \mathrel{\mathop{\lesssim}\limits^{\beta}} (1-p) p ≲ α e ⟺ ( 1 − e ) ≲ β ( 1 − p ) If for some γ ∈ ( 1 − p ) R e \gamma\in(1-p)Re γ ∈ ( 1 − p ) R e , we get α + γ + β ∈ U ( R ) \alpha+\gamma+\beta\in U(R) α + γ + β ∈ U ( R ) then
e ≲ α p ⟺ ( 1 − p ) ≲ β ( 1 − e ) e \mathrel{\mathop{\lesssim}\limits^{\alpha}} p \Longleftrightarrow(1-p) \mathrel{\mathop{\lesssim}\limits^{\beta}} (1-e) e ≲ α p ⟺ ( 1 − p ) ≲ β ( 1 − e )
Proof. (1) Label x = α + η + β ∈ U ( R ) x=\alpha+\eta+\beta\in U(R) x = α + η + β ∈ U ( R ) for some η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) .
( ⟹ ) (\Longrightarrow) ( ⟹ ) : Let p ≲ α e p \mathrel{\mathop{\lesssim}\limits^{\alpha}} e p ≲ α e with α ′ ∈ e R p \alpha' \in eRp α ′ ∈ e R p . Then α α ′ = p \alpha\alpha'=p α α ′ = p . We prove that β ′ = ( 1 − e ) x − 1 ( 1 − p ) \beta'=(1-e)x^{-1}(1-p) β ′ = ( 1 − e ) x − 1 ( 1 − p ) satisfies β β ′ = 1 − p \beta\beta'=1-p β β ′ = 1 − p and β ′ β = 1 − e \beta'\beta=1-e β ′ β = 1 − e .
x α ′ = ( α + η + β ) α ′ = α α ′ + η α ′ + β α ′ = p + ( η ( 1 − e ) ) ( e α ′ ) + ( β ( 1 − e ) ) ( e α ′ ) = p ⟹ α ′ = x − 1 p \begin{aligned}
x\alpha'&=(\alpha+\eta+\beta)\alpha'\\
&=\alpha\alpha'+\eta\alpha'+\beta\alpha'\\
&=p+(\eta(1-e))(e\alpha')+(\beta(1-e))(e\alpha')=p\\
&\Longrightarrow\alpha'=x^{-1}p
\end{aligned} x α ′ = ( α + η + β ) α ′ = α α ′ + η α ′ + β α ′ = p + ( η ( 1 − e )) ( e α ′ ) + ( β ( 1 − e )) ( e α ′ ) = p ⟹ α ′ = x − 1 p
Now, e α ′ = α ′ ⟹ ( 1 − e ) α ′ = 0 ⟹ ( 1 − e ) x − 1 p = 0 e\alpha'=\alpha' \Longrightarrow(1-e)\alpha'=0 \Longrightarrow(1-e)x^{-1}p=0 e α ′ = α ′ ⟹ ( 1 − e ) α ′ = 0 ⟹ ( 1 − e ) x − 1 p = 0 .
Since η = p η \eta=p\eta η = p η and β = ( 1 − p ) β \beta=(1-p)\beta β = ( 1 − p ) β we get x ( 1 − e ) = ( α + η + β ) ( 1 − e ) = α ( 1 − e ) + η ( 1 − e ) + β ( 1 − e ) = η + β x(1-e)=(\alpha+\eta+\beta)(1-e)=\alpha(1-e)+\eta(1-e)+\beta(1-e)=\eta+\beta x ( 1 − e ) = ( α + η + β ) ( 1 − e ) = α ( 1 − e ) + η ( 1 − e ) + β ( 1 − e ) = η + β
⟹ ( 1 − e ) x − 1 ( x ( 1 − e ) ) = ( 1 − e ) x − 1 ( η + β ) = ( 1 − e ) x − 1 ( p η + ( 1 − p ) β ) = ( 1 − e ) x − 1 p η + ( 1 − e ) x − 1 ( 1 − p ) β = 0 + β ′ β = β ′ β \begin{aligned}
&\Longrightarrow\quad(1-e)x^{-1}(x(1-e))=(1-e)x^{-1}(\eta+\beta)\\
&=(1-e)x^{-1}(p\eta+(1-p)\beta)\\
&=(1-e)x^{-1}p\eta+(1-e)x^{-1}(1-p)\beta\\
&=0+\beta'\beta=\beta'\beta
\end{aligned} ⟹ ( 1 − e ) x − 1 ( x ( 1 − e )) = ( 1 − e ) x − 1 ( η + β ) = ( 1 − e ) x − 1 ( p η + ( 1 − p ) β ) = ( 1 − e ) x − 1 p η + ( 1 − e ) x − 1 ( 1 − p ) β = 0 + β ′ β = β ′ β
Thus, we have ( 1 − e ) ≲ β ( 1 − p ) (1-e) \mathrel{\mathop{\lesssim}\limits^{\beta}} (1-p) ( 1 − e ) ≲ β ( 1 − p ) with β ′ β = 1 − e \beta'\beta=1-e β ′ β = 1 − e .
In fact, since β \beta β is both left and right invertible, β \beta β is block invertible with inverse β ′ \beta' β ′ .
Remark 2.17. In the proof above, we simply assumed α \alpha α is right invertible whence α α ′ = p \alpha\alpha'=p α α ′ = p . Then as a consequence we not only obtain β \beta β being an invertible block in ( 1 − p ) R ( 1 − e ) (1-p)R(1-e) ( 1 − p ) R ( 1 − e ) but α \alpha α being also left invertible. Indeed, e = ( e M − 1 ) ( M e ) = ( e M − 1 ) ( α + η + β ) e = ( e M − 1 ) α e=(eM^{-1})(Me)=(eM^{-1})(\alpha+\eta+\beta)e=(eM^{-1})\alpha e = ( e M − 1 ) ( M e ) = ( e M − 1 ) ( α + η + β ) e = ( e M − 1 ) α . Also M α ′ = p ⟹ M α ′ α = p α = α ⟹ ( α ′ α ) = M − 1 α M\alpha'=p \Longrightarrow M\alpha'\alpha=p\alpha=\alpha\Longrightarrow(\alpha'\alpha)=M^{-1}\alpha M α ′ = p ⟹ M α ′ α = p α = α ⟹ ( α ′ α ) = M − 1 α . Therefore e = e M − 1 α = e ( α ′ α ) ⟹ α ′ α = e e=eM^{-1}\alpha=e(\alpha'\alpha) \Longrightarrow\alpha'\alpha=e e = e M − 1 α = e ( α ′ α ) ⟹ α ′ α = e .
( ⟸ ) (\Longleftarrow) ( ⟸ ) : Suppose ( 1 − e ) ≲ β ( 1 − p ) (1-e) \lesssim^{\beta} (1-p) ( 1 − e ) ≲ β ( 1 − p ) with β ′ β = 1 − e \beta'\beta=1-e β ′ β = 1 − e for β ′ ∈ ( 1 − e ) R ( 1 − p ) \beta' \in(1-e)R(1-p) β ′ ∈ ( 1 − e ) R ( 1 − p ) . We prove that α ′ = e M − 1 p \alpha'=eM^{-1}p α ′ = e M − 1 p satisfies α α ′ = p \alpha\alpha'=p α α ′ = p .
Since α = p α \alpha=p\alpha α = p α , η = p η \eta=p\eta η = p η and β ′ = β ′ ( 1 − p ) \beta'=\beta'(1-p) β ′ = β ′ ( 1 − p ) we have
β ′ M = β ′ ( α + η + β ) = β ′ α + β ′ η + β ′ β = ( β ′ ( 1 − p ) ) ( p α ) + ( β ′ ( 1 − p ) p η ) + ( 1 − e ) = ( 1 − e ) \begin{aligned}
\beta'M &= \beta'(\alpha+\eta+\beta) \\
&= \beta'\alpha+\beta'\eta+\beta'\beta\\
&= (\beta'(1-p))(p\alpha)+(\beta'(1-p)p\eta)+(1-e)=(1-e)
\end{aligned} β ′ M = β ′ ( α + η + β ) = β ′ α + β ′ η + β ′ β = ( β ′ ( 1 − p )) ( p α ) + ( β ′ ( 1 − p ) p η ) + ( 1 − e ) = ( 1 − e )
⟹ β ′ = ( 1 − e ) M − 1 \Longrightarrow\quad\beta'=(1-e)M^{-1} ⟹ β ′ = ( 1 − e ) M − 1
Also, β ′ = β ′ ( 1 − p ) ⟹ β ′ p = 0 ⟹ ( 1 − e ) M − 1 p = 0 \beta'=\beta'(1-p)\Longrightarrow\beta'p=0\Longrightarrow(1-e)M^{-1}p=0 β ′ = β ′ ( 1 − p ) ⟹ β ′ p = 0 ⟹ ( 1 − e ) M − 1 p = 0 .
Now p M = ( p ) ( α + η + β ) = α + η pM=(p)(\alpha+\eta+\beta)=\alpha+\eta pM = ( p ) ( α + η + β ) = α + η . Since η = η ( 1 − e ) \eta=\eta(1-e) η = η ( 1 − e ) and α = α e \alpha=\alpha e α = α e we have
⟹ ( p M ) M − 1 p = ( α + η ) M − 1 p = ( α ( e ) + η ( 1 − e ) ) M − 1 p = α e M − 1 p ⏟ α ′ + η ( 1 − e ) M − 1 p ⏟ 0 = α α ′ \begin{aligned}
&\Longrightarrow\quad(pM)M^{-1}p=(\alpha+\eta)M^{-1}p \\
&=(\alpha(e)+\eta(1-e))M^{-1}p \\
&=\underbrace{\alpha eM^{-1}p}_{\alpha'}+\underbrace{\eta(1-e)M^{-1}p}_{0}=\alpha\alpha'
\end{aligned} ⟹ ( pM ) M − 1 p = ( α + η ) M − 1 p = ( α ( e ) + η ( 1 − e )) M − 1 p = α ′ α e M − 1 p + 0 η ( 1 − e ) M − 1 p = α α ′
We therefore have α α ′ = p \alpha\alpha'=p α α ′ = p .
In fact, since α \alpha α is both left and right invertible, α \alpha α is block invertible with inverse α ′ \alpha' α ′ .
(2) The proof for (2) holds similarly as above for γ ∈ ( 1 − p ) R e \gamma\in(1-p)Re γ ∈ ( 1 − p ) R e by taking α 1 = β \alpha_1=\beta α 1 = β , β 1 = α \beta_1=\alpha β 1 = α , p 1 = 1 − p p_1=1-p p 1 = 1 − p and e 1 = 1 − e e_1=1-e e 1 = 1 − e and applying (1). □
Block Invertibility and Stable Range
We now look at defining the notion of stable range on skew-corners such as e R f eRf e R f for distinct idempotents e , f e,f e , f , since they are intimately related to the concept of invertible blocks.
Definition 3.1. Let e , f ∈ Idem ( R ) e,f\in\operatorname{Idem}(R) e , f ∈ Idem ( R ) . The skew-corner e R f eRf e R f has right stable range one, written sr r ( e R f ) = 1 \operatorname{sr}_{r}(eRf)=1 sr r ( e R f ) = 1 , if whenever a ∈ e R f a\in eRf a ∈ e R f and b ∈ e R e b\in eRe b ∈ e R e satisfy a R + b R = e R aR+bR=eR a R + b R = e R , there exists y ∈ e R f y\in eRf y ∈ e R f such that a + b y a+by a + b y is right block invertible in e R f eRf e R f , i.e. there exists z ∈ f R e z\in fRe z ∈ f R e such that ( a + b y ) z = e (a+by)z=e ( a + b y ) z = e .
Similarly, the skew-corner e R f eRf e R f has left stable rank one, written sr l ( e R f ) = 1 \operatorname{sr}_{l}(eRf)=1 sr l ( e R f ) = 1 , if whenever a ∈ e R f a\in eRf a ∈ e R f and b ∈ f R f b\in fRf b ∈ f R f satisfy R a + R b = R f Ra+Rb=Rf R a + R b = R f , there is y ∈ e R f y\in eRf y ∈ e R f such that a + y b a+yb a + y b is left block invertible in e R f eRf e R f , i.e. there exists z ∈ f R e z\in fRe z ∈ f R e such that z ( a + y b ) = f z(a+yb)=f z ( a + y b ) = f .
When e = f e=f e = f we obtain the usual notion of stable range on the corner ring e R e eRe e R e .
However, we have the following association.
Proposition 3.3. Let e , f ∈ Idem ( R ) e, f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) . Then e R f eRf e R f has right (resp. left) stable range one if and only if its associate skew corner f R e fRe f R e has left (resp. right) stable range one.
Proof. It suffices to prove that sr l ( f R e ) = 1 \operatorname{sr}_{l}(fRe)=1 sr l ( f R e ) = 1 when sr r ( e R f ) = 1 \operatorname{sr}_{r}(eRf)=1 sr r ( e R f ) = 1 .
Suppose x a + b = e xa+b=e x a + b = e for some a ∈ f R e a \in fRe a ∈ f R e , x ∈ e R f x \in eRf x ∈ e R f and b ∈ e R e b \in eRe b ∈ e R e . Then there exists y ∈ e R f y \in eRf y ∈ e R f such that x + b y x+by x + b y is right block invertible in e R f eRf e R f . Choose z ∈ f R e z \in fRe z ∈ f R e such that ( x + b y ) z = e (x+by)z=e ( x + b y ) z = e .
We need a y ′ ∈ f R e y^{\prime} \in fRe y ′ ∈ f R e such that a + y ′ b a+y^{\prime}b a + y ′ b is left block invertible in f R e fRe f R e .
Define y ′ = ( f − a y ) z y^{\prime}=(f-ay)z y ′ = ( f − a y ) z and s = x + y ( f − a x ) s=x+y(f-ax) s = x + y ( f − a x ) . Then s a = x a + y a − y a x a = x a + y a b sa=xa+ya-yaxa=xa+yab s a = x a + y a − y a x a = x a + y ab and
s ( f − a y ) = ( x + y ( f − a x ) ) ( f − a y ) = x − x a y + y − y a x − y a y + y a x a y = ( x − x a y + y ) − y a ( x − x a y + y ) = ( e − y a ) ( x − x a y + y ) \begin{aligned}
s(f-ay)&=(x+y(f-ax))(f-ay) \\
&=x-xay+y-yax-yay+yaxay \\
&=(x-xay+y)-ya(x-xay+y) \\
&=(e-ya)(x-xay+y)
\end{aligned} s ( f − a y ) = ( x + y ( f − a x )) ( f − a y ) = x − x a y + y − y a x − y a y + y a x a y = ( x − x a y + y ) − y a ( x − x a y + y ) = ( e − y a ) ( x − x a y + y ) where ( x − x a y + y ) = x + ( e − x a ) y = x + b y (x-xay+y)=x+(e-xa)y=x+by ( x − x a y + y ) = x + ( e − x a ) y = x + b y .
Then s ( a + y ′ b ) = s a + s y ′ b = s a + ( e − y a ) ( x + b y ) z ⏟ e b = s a + ( e − y a ) b = x a + b = e s(a+y^{\prime}b)=sa+sy^{\prime}b=sa+(e-ya)(x+by)\underbrace{z}_{e}b=sa+(e-ya)b=xa+b=e s ( a + y ′ b ) = s a + s y ′ b = s a + ( e − y a ) ( x + b y ) e z b = s a + ( e − y a ) b = x a + b = e .
Thus a + y ′ b a+y^{\prime}b a + y ′ b is left block invertible in f R e fRe f R e .
Using e = f e=f e = f , we recover the following result of Vaserstein.
Corollary 3.4 (Vaserstein [38 ] ). A ring R R R has left stable range one if and only if it has right stable range one.
Right (resp. Left) stable range one of a skew-corner ring e R f eRf e R f being a stronger form of e ≲ f e \lesssim f e ≲ f (resp. f ≲ e f \lesssim e f ≲ e ), provides the right conditions under which a subequivalent pair of idempotents gets promoted to equivalency as we see below.
Theorem 3.5. Let a ∈ R a \in R a ∈ R and b , c ∈ R b,c \in R b , c ∈ R be von Neumann regular with b R = e R bR=eR b R = e R and R c = R f Rc=Rf R c = R f for some e , f ∈ Idem ( R ) e,f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) . Then the following holds.
If sr l ( e R f ) = 1 \operatorname{sr}_{l}(eRf)=1 sr l ( e R f ) = 1 then a a a is left ( b , c ) (b,c) ( b , c ) invertible if and and only if a a a is ( b , c ) (b,c) ( b , c ) invertible.
If sr r ( e R f ) = 1 \operatorname{sr}_{r}(eRf)=1 sr r ( e R f ) = 1 then a a a is right ( b , c ) (b,c) ( b , c ) invertible if and and only if a a a is ( b , c ) (b,c) ( b , c ) invertible.
If e ∼ f e \sim f e ∼ f and sr l ( e R e ) = 1 \operatorname{sr}_{l}(eRe)=1 sr l ( e R e ) = 1 then every one-sided left ( b , c ) (b,c) ( b , c ) inverse is two-sided.
Proof. (i) Suppose sr l ( e R f ) = 1 \operatorname{sr}_{l}(eRf)=1 sr l ( e R f ) = 1 . Under this scenario, we observe that every right invertible block in e R f eRf e R f is invertible.
Let a ∈ e R f a \in eRf a ∈ e R f be right invertible in e R f eRf e R f i.e. there exists x ∈ f R e x \in fRe x ∈ f R e such that a x = e ax=e a x = e . We can write x a + b = f xa+b=f x a + b = f where b = ( f − x a ) ∈ f R f b=(f-xa) \in fRf b = ( f − x a ) ∈ f R f and x a ∈ f R f xa \in fRf x a ∈ f R f . By definition, there exists y ∈ e R f y \in eRf y ∈ e R f such that a + y b a+yb a + y b is left invertible in e R f eRf e R f i.e. there exists z ∈ f R e z \in fRe z ∈ f R e such that z ( a + y b ) = f z(a+yb)=f z ( a + y b ) = f . Post-multiplying by x x x gives z a x = x zax=x z a x = x or z = x z=x z = x . Thus x x x is an invertible block in f R e fRe f R e and consequently, x a = e xa=e x a = e . Thus a a a is an invertible block in e R f eRf e R f .
Now if an element a a a is left ( b , c ) (b,c) ( b , c ) invertible then by Proposition 2.3 (ii) we have e ≲ f a e f e \lesssim^{fae} f e ≲ f a e f , and therefore there exists w ∈ e R f w \in eRf w ∈ e R f such that w ( f a e ) = e w(fae)=e w ( f a e ) = e . Then w w w being right invertible in e R f eRf e R f is invertible and therefore, f ≲ e e f \lesssim^{e} e f ≲ e e which by Proposition 2.3 (i) proves that a a a is ( b , c ) (b,c) ( b , c ) invertible.
(ii) If sr r ( e R f ) = 1 \operatorname{sr}_{r}(eRf)=1 sr r ( e R f ) = 1 then we can similarly show that every left invertible block in e R f eRf e R f is invertible. And the proof follows similarly as above.
(iii) If e ∼ f e \sim f e ∼ f and sr l ( e R e ) = 1 \operatorname{sr}_{l}(eRe)=1 sr l ( e R e ) = 1 then by [2 Lemma 2(b)] , sr r ( e R f ) = 1 \operatorname{sr}_{r}(eRf)=1 sr r ( e R f ) = 1 and sr l ( e R f ) = 1 \operatorname{sr}_{l}(eRf)=1 sr l ( e R f ) = 1 . Then using the arguments above we can see that every left and right invertible block in e R f eRf e R f is invertible. Thus every one-sided left ( b , c ) (b,c) ( b , c ) inverse is two-sided.
Remark 3.6. From ( i i i ) (iii) ( iii ) and from [2 Lemma 2(b)] , we can see that sr r ( e R f ) = 1 \operatorname{sr}_{r}(eRf)=1 sr r ( e R f ) = 1 and sr l ( e R f ) = 1 \operatorname{sr}_{l}(eRf)=1 sr l ( e R f ) = 1 if and only if e ∼ f e\sim f e ∼ f and sr ( e R e ) = 1 \operatorname{sr}(eRe)=1 sr ( e R e ) = 1 .
In the proof above we showed that sr l ( e R f ) = 1 \operatorname{sr}_{l}(eRf)=1 sr l ( e R f ) = 1 (resp. sr r ( e R f ) = 1 \operatorname{sr}_{r}(eRf)=1 sr r ( e R f ) = 1 ) then every right (resp. left) invertible block in e R f eRf e R f is invertible. Now recall that if e ∼ f e\sim f e ∼ f , every right (resp. left) invertible block of e R f eRf e R f is exactly x x x times a right (resp. left) block invertible element of f R f fRf f R f . The same works for e R e eRe e R e but by post-multiplication by x x x . Hence we also have a weaker characterization of the form:
If e ∼ f e\sim f e ∼ f and e R e eRe e R e is directly finite then every left and right invertible block in e R f eRf e R f is invertible .
The converse does not necessary hold unless e R f eRf e R f contains atleast one left or right invertible block i.e. If e ≾ f e\precsim f e ≾ f or f ≾ e f\precsim e f ≾ e , then every left and every right invertible block in e R f eRf e R f is invertible if and only if e ∼ f e\sim f e ∼ f and e R e eRe e R e is directly finite.
This inference gives a partial answer to a question [10 Sec. 4, Pg 71] :
If a ring R R R has stable range one, is it ( b , b ) (b,b) ( b , b ) -finite for all b ∈ R b\in R b ∈ R ?
In [10 ] , for a fixed b ∈ R b\in R b ∈ R , Drazin defines a ring R R R to be ( b , b ) (b,b) ( b , b ) -finite if for all a ∈ R a\in R a ∈ R , a a a is left ( b , b ) (b,b) ( b , b ) -invertible if and only if a a a is right ( b , b ) (b,b) ( b , b ) -invertible.
For von Neumann regular elements b ∈ R b\in R b ∈ R the result is true for directly finite rings.
Proposition 3.7. R R R is directly finite if and only if R R R is ( b , b ) (b,b) ( b , b ) -finite for every von Neumann regular element b ∈ R b\in R b ∈ R .
Proof. ( ⟹ ) (\Longrightarrow) ( ⟹ ) : Suppose R R R is directly finite. Let b ∈ R b\in R b ∈ R be von Neumann regular and b = b x b b=bxb b = b x b for some x ∈ R x\in R x ∈ R . Set e = b x e=bx e = b x and f = x b f=xb f = x b . Then e , f ∈ Idem ( R ) e,f\in\operatorname{Idem}(R) e , f ∈ Idem ( R ) such that b R = e R bR=eR b R = e R and R b = R f Rb=Rf R b = R f and e ∼ f e\sim f e ∼ f . Then from Prop. 2.3 (ii), a a a is left ( b , b ) (b,b) ( b , b ) -invertible if and only if e ≾ f a e f e\precsim fae f e ≾ f a e f , that is, the block f a e ∈ f R e fae\in fRe f a e ∈ f R e has a left inverse in e R f eRf e R f . Likewise, a a a is right ( b , b ) (b,b) ( b , b ) -invertible if and only if f f a e ≾ e f^{fae}\precsim e f f a e ≾ e , that is, f a e fae f a e has a right inverse in e R f eRf e R f . Since R R R is directly finite, so is its corner ring e R e eRe e R e . As e ∼ f e\sim f e ∼ f , Remark 3.6 shows that any block in f R e fRe f R e with a one-sided inverse in e R f eRf e R f is invertible. Hence, if a a a is left or right ( b , b ) (b,b) ( b , b ) -invertible, then f a e fae f a e is invertible, that is, f ∼ e f\sim e f ∼ e . By Prop. 2.3 (i), a a a is then ( b , b ) (b,b) ( b , b ) -invertible, and in particular both left and right ( b , b ) (b,b) ( b , b ) -invertible. Thus R R R is ( b , b ) (b,b) ( b , b ) -finite.
Conversely, if suppose a y = 1 ay=1 a y = 1 for some y ∈ R y\in R y ∈ R . Then a a a has a right ( 1 , 1 ) (1,1) ( 1 , 1 ) inverse y y y . Then R R R being ( 1 , 1 ) (1,1) ( 1 , 1 ) finite forces y a = 1 ya=1 y a = 1 . Thus R R R is directly finite. □ \square □
In the case when b b b is not assumed to be regular we obtain an equivalent characterization of left and right ( b , b ) (b,b) ( b , b ) invertibility from Remark 2.2 in Section 2.
If R R R is an exchange ring
( i ) (i) ( i ) a a a is right ( b , b ) (b,b) ( b , b ) -invertible if and only if b ∈ R y b\in Ry b ∈ R y for some y ∈ T ( a b ) y\in\mathrm{T}(ab) y ∈ T ( ab ) , and in this case ( b y ) (by) ( b y ) is a right ( b , b ) (b,b) ( b , b ) -inverse of a a a .
( i i ) (ii) ( ii ) a a a is left ( b , b ) (b,b) ( b , b ) -invertible if and only if b ∈ y R b\in yR b ∈ y R for some y ∈ T ( b a ) y\in\mathrm{T}(ba) y ∈ T ( ba ) , in which case ( y b ) (yb) ( y b ) is a left ( b , b ) (b,b) ( b , b ) -inverse of a a a .
Proposition 3.8. Let R R R be an exchange ring. Then R R R is directly finite if and only if R R R is ( b , b ) (b,b) ( b , b ) -finite for all b ∈ R b\in R b ∈ R .
Proof. ( ⟹ ) (\Longrightarrow) ( ⟹ ) : Suppose R R R is directly finite and fix b ∈ R b\in R b ∈ R . For any a ∈ R a\in R a ∈ R , let a a a be right ( b , b ) (b,b) ( b , b ) invertible. Then from above, b ∈ R y b\in Ry b ∈ R y for some y ∈ T ( a b ) y\in\mathrm{T}(ab) y ∈ T ( ab ) , and in this case ( b y ) (by) ( b y ) is a right ( b , b ) (b,b) ( b , b ) -inverse of a a a .
Set w = b y w=by w = b y . Then w a w w = w waww=w w a w w = w and b a w w = b a b y = b baww=baby=b ba w w = bab y = b . Label e = w a e=wa e = w a and f = a w f=aw f = a w . Clearly, e , f ∈ Idem ( R ) e,f\in\operatorname{Idem}(R) e , f ∈ Idem ( R )
and f ∼ f a e e f \overset{fae}{\sim} e f ∼ f a e e . We prove that b R = e R bR = eR b R = e R and R b = R f Rb = Rf R b = R f .
Clearly, e = b y a ⟹ e ∈ b R e = bya \implies e \in bR e = b y a ⟹ e ∈ b R and b = b f ⟹ b ∈ R f b = bf \implies b \in Rf b = b f ⟹ b ∈ R f .
Now e = w a = b y a ⟹ e = ( e b f ) ( f y a e ) e = wa = bya \implies e = (ebf)(fyae) e = w a = b y a ⟹ e = ( e b f ) ( f y a e ) . Label x = e b f x = ebf x = e b f . Then x x x is a right-invertible block in e R f eRf e R f . Since e R e eRe e R e is directly finite and e ∼ f e \sim f e ∼ f , by Remark 3.6, x x x is invertible and subsequently, ( f y a e ) ( e b f ) = f ⟹ f y a e b a w = f ⟹ f ∈ R b (fyae)(ebf) = f \implies fyae\,baw = f \implies f \in Rb ( f y a e ) ( e b f ) = f ⟹ f y a e ba w = f ⟹ f ∈ R b . Therefore R b = R f Rb = Rf R b = R f .
Also, b = b f = b ( f y a e ) ( e b f ) = b y a ( e b f ) ∈ e R b = bf = b(fyae)(ebf) = bya(ebf) \in eR b = b f = b ( f y a e ) ( e b f ) = b y a ( e b f ) ∈ e R . Therefore e R = b R eR = bR e R = b R .
Hence, by Prop. 2.3 (i), a a a is ( b , b ) (b,b) ( b , b ) -invertible with inverse w w w .
The proof works similarly if a a a is assumed to be left ( b , b ) (b,b) ( b , b ) -invertible.
Since a a a is arbitrary, R R R is ( b , b ) (b,b) ( b , b ) -finite.
The converse is straightforward from Proposition 3.7 . □
Remark 3.9. Without the hypothesis of exchange on R R R , if for some a ∈ R a \in R a ∈ R we choose b ∈ R b \in R b ∈ R such that a b 2 , b 2 a ab^2,b^2a a b 2 , b 2 a are von Neumann regular and R R R is directly finite, then the same technique can be applied alongwith the equality in [37 Pg 12] to prove that a a a is left ( b , b ) (b,b) ( b , b ) -invertible if and only if a a a is right ( b , b ) (b,b) ( b , b ) -invertible.
When e = f e = f e = f , the skew corner e R f eRf e R f reduces to the corner ring e R e eRe e R e . We now record several cases in which an element of e R e eRe e R e has stable range one. By sr S ( a ) \operatorname{sr}_{S}(a) sr S ( a ) in a ring S S S we mean the left stable range, i.e whenever S a + S b = S Sa + Sb = S S a + S b = S for any b ∈ S b \in S b ∈ S , there exists s ∈ S s \in S s ∈ S such that a + s b ∈ U ( S ) a + sb \in U(S) a + s b ∈ U ( S ) . This is equivalent to the right stable range of the element [21 ] . We note a simple well-known characterization of stable range one condition on elements first proved by Menal and Moncasi in [30 ] as pointed out by P. Ara. It turns out to be quite useful as it is not based on a condition and thus limits the number of independent variables in the pot. Suprisingly, a variation of it also holds for regular and unit-regular elements [24 ] .
Lemma 3.10. Let R R R be a ring with identity 1 R 1_R 1 R . Then the following are equivalent
(i) sr R ( a ) = 1 \operatorname{sr}_{R}(a) = 1 sr R ( a ) = 1
(ii) For each s ∈ R s \in R s ∈ R , there exists r ∈ R r \in R r ∈ R such that a + r ( 1 R − s a ) ∈ U ( R ) a + r(1_R - sa) \in U(R) a + r ( 1 R − s a ) ∈ U ( R )
(iii) For each s ∈ R s \in R s ∈ R , there exists r ∈ R r \in R r ∈ R such that a + ( 1 R − a s ) r ∈ U ( R ) a + (1_R - as)r \in U(R) a + ( 1 R − a s ) r ∈ U ( R )
Proof. (i) ⟹ \implies ⟹ (ii): Let a a a have stable range one. Choose arbitrary s ∈ R s \in R s ∈ R . Then s a + ( 1 − s a ) = 1 sa + (1 - sa) = 1 s a + ( 1 − s a ) = 1 . Since sr R ( a ) = 1 \operatorname{sr}_{R}(a) = 1 sr R ( a ) = 1 , we have a + r ( 1 − s a ) ∈ U ( R ) a + r(1 - sa) \in U(R) a + r ( 1 − s a ) ∈ U ( R ) for some r ∈ R r \in R r ∈ R .
(ii) ⟹ \implies ⟹ (i): Conversely, for each s ∈ R s \in R s ∈ R let a + r ( 1 − s a ) ∈ U ( R ) a + r(1 - sa) \in U(R) a + r ( 1 − s a ) ∈ U ( R ) for some r ∈ R r \in R r ∈ R . Suppose x a + b = 1 xa + b = 1 x a + b = 1 . Then b = ( 1 − x a ) b = (1 - xa) b = ( 1 − x a ) . Take s = x s = x s = x . Then a + r x ( 1 − s a ) = a + r x ( 1 − x a ) = a + r x b ∈ U ( R ) a + r_x(1 - sa) = a + r_x(1 - xa) = a + r_xb \in U(R) a + r x ( 1 − s a ) = a + r x ( 1 − x a ) = a + r x b ∈ U ( R ) for some r x ∈ R r_x \in R r x ∈ R . Thus a a a has left stable range one.
(i) ⟺ \Longleftrightarrow ⟺ (iii): Similar, using the left-right symmetry of stable range one elements. □
Khurana and Lam developed the Suspension Theorem in [21 ] that allows transfer of stable range one from the Peirce corner to the ambient ring. We extend this notion from a single Peirce decomposition to transporting stable range across two-sided Peirce decompositions for distinct idempotents. The use of distinct idempotents in the preceding theorem is motivated by the rectangular-corner techniques of Ara and Goodearl [2 ] . In their study of stable rank of full corners, Ara and Goodearl introduced a stable rank condition for a skew corner p R q pRq pR q , and showed that this condition is preserved upon replacing either endpoint idempotent by an equivalent one ([2, Lemma 2(b)]). In keeping with the terminology used by Khurana and Lam, we shall refer to the following result as the General Suspension Theorem .
Theorem 3.11 (General Suspension Theorem). Let p , e ∈ idem ( R ) p,e \in\operatorname{idem}(R) p , e ∈ idem ( R ) such that
p ∼ θ e and ( 1 − p ) ∼ β ( 1 − e ) p \overset{\theta}{\sim} e \quad\text{and} \quad(1-p) \overset{\beta}{\sim} (1-e) p ∼ θ e and ( 1 − p ) ∼ β ( 1 − e ) Fix arbitrary a ∈ e R e , c ∈ p R p a \in eRe,c \in pRp a ∈ e R e , c ∈ pR p . Then the following conditions hold.
(1) For every η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) and η ~ ∈ e R ( 1 − p ) \widetilde{\eta} \in eR(1-p) η ∈ e R ( 1 − p ) , the following hold.
(i) sr p R p ( c ) = 1 ⟺ sr R ( c θ + η + β ) = 1 \operatorname{sr}_{pRp}(c)=1 \Longleftrightarrow\operatorname{sr}_{R}(c\theta+\eta+\beta)=1 sr pR p ( c ) = 1 ⟺ sr R ( c θ + η + β ) = 1
(ii) sr e R e ( a ) = 1 ⟺ sr R ( a θ ′ + η ~ + β ′ ) = 1 \operatorname{sr}_{eRe}(a)=1 \Longleftrightarrow\operatorname{sr}_{R}(a\theta^{\prime}+\widetilde{\eta}+\beta^{\prime})=1 sr e R e ( a ) = 1 ⟺ sr R ( a θ ′ + η + β ′ ) = 1
(2) For every η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) and η ~ ∈ e R ( 1 − p ) \widetilde{\eta} \in eR(1-p) η ∈ e R ( 1 − p ) , the following hold.
(i) sr p R p ( c ) = 1 ⟺ sr R ( θ ′ c + η ~ + β ′ ) = 1 \operatorname{sr}_{pRp}(c)=1 \Longleftrightarrow\operatorname{sr}_{R}(\theta^{\prime}c+\widetilde{\eta}+\beta^{\prime})=1 sr pR p ( c ) = 1 ⟺ sr R ( θ ′ c + η + β ′ ) = 1
(ii) sr e R e ( a ) = 1 ⟺ sr R ( θ a + η + β ) = 1 \operatorname{sr}_{eRe}(a)=1 \Longleftrightarrow\operatorname{sr}_{R}(\theta a+\eta+\beta)=1 sr e R e ( a ) = 1 ⟺ sr R ( θ a + η + β ) = 1
In each of the cases, necessarily sr R ( a ) = 1 \operatorname{sr}_{R}(a)=1 sr R ( a ) = 1 .
Proof. By definition, θ ′ ∈ e R p \theta^{\prime} \in eRp θ ′ ∈ e R p and β ′ ∈ ( 1 − e ) R ( 1 − p ) \beta^{\prime} \in(1-e)R(1-p) β ′ ∈ ( 1 − e ) R ( 1 − p ) .
(1)(i) For arbitrary c ∈ p R p c \in pRp c ∈ pR p , we prove that for every η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e )
sr p R p ( c ) = 1 ⟺ sr R ( c θ + η + β ) = 1 \operatorname{sr}_{pRp}(c)=1 \Longleftrightarrow\operatorname{sr}_{R}(c\theta+\eta+\beta)=1 sr pR p ( c ) = 1 ⟺ sr R ( c θ + η + β ) = 1
⟹ \Longrightarrow ⟹ : Let sr p R p ( c ) = 1 \operatorname{sr}_{pRp}(c)=1 sr pR p ( c ) = 1 and label M = c θ + η + β M=c\theta+\eta+\beta M = c θ + η + β for arbitrary η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) . Fix an arbitrary s ∈ R s \in R s ∈ R . We prove M + r ( 1 − s M ) ∈ U ( R ) M+r(1-sM)\in U(R) M + r ( 1 − s M ) ∈ U ( R ) for some r ∈ R r\in R r ∈ R .
Consider the element s ~ = θ s p ∈ p R p \widetilde{s}=\theta sp\in pRp s = θ s p ∈ pR p . Then in the ring p R p pRp pR p we have s ~ c + ( p − s ~ c ) = p \widetilde{s}c+(p-\widetilde{s}c)=p s c + ( p − s c ) = p . Hence by sr p R p ( c ) = 1 \operatorname{sr}_{pRp}(c)=1 sr pR p ( c ) = 1 , there exists r 0 ∈ p R p r_{0}\in pRp r 0 ∈ pR p such that
c + r 0 ( p − s ~ c ) ∈ U ( p R p ) c+r_{0}(p-\widetilde{s}c)\in U(pRp) c + r 0 ( p − s c ) ∈ U ( pR p )
Label r = r 0 θ r=r_{0}\theta r = r 0 θ and consider the element M + r ( 1 − s M ) M+r(1-sM) M + r ( 1 − s M ) . Since c = p c c=pc c = p c , η = p η \eta=p\eta η = p η and β = ( 1 − p ) β \beta=(1-p)\beta β = ( 1 − p ) β we can write
M + r ( 1 − s M ) = M + r − r s ( c θ + η + β ) = M + r − ( r 0 θ s ( p c ) θ + r 0 θ s ( p η ) + r 0 θ s ( ( 1 − p ) β ) ) = M + r − ( r 0 s ~ c θ + r 0 s ~ η + r 0 θ s ( ( 1 − p ) β ) ) = ( c θ + η + β ) + ( r 0 θ ) − ( r 0 s ~ c θ + r 0 s ~ η + r 0 θ s ( ( 1 − p ) β ) ) = ( c + r 0 − r 0 s ~ c ) θ + [ η − r 0 s ~ η − r 0 θ s ( ( 1 − p ) β ) ] + β \begin{aligned}
M+r(1-sM)
&=M+r-rs(c\theta+\eta+\beta)\\
&=M+r-\left(r_{0}\theta s(pc)\theta+r_{0}\theta s(p\eta)+r_{0}\theta s((1-p)\beta)\right)\\
&=M+r-\left(r_{0}\widetilde{s}c\theta+r_{0}\widetilde{s}\eta+r_{0}\theta s((1-p)\beta)\right)\\
&=(c\theta+\eta+\beta)+(r_{0}\theta)-\left(r_{0}\widetilde{s}c\theta+r_{0}\widetilde{s}\eta+r_{0}\theta s((1-p)\beta)\right)\\
&=(c+r_{0}-r_{0}\widetilde{s}c)\theta+
[\eta-r_{0}\widetilde{s}\eta-r_{0}\theta s((1-p)\beta)]+\beta
\end{aligned} M + r ( 1 − s M ) = M + r − r s ( c θ + η + β ) = M + r − ( r 0 θ s ( p c ) θ + r 0 θ s ( p η ) + r 0 θ s (( 1 − p ) β ) ) = M + r − ( r 0 s c θ + r 0 s η + r 0 θ s (( 1 − p ) β ) ) = ( c θ + η + β ) + ( r 0 θ ) − ( r 0 s c θ + r 0 s η + r 0 θ s (( 1 − p ) β ) ) = ( c + r 0 − r 0 s c ) θ + [ η − r 0 s η − r 0 θ s (( 1 − p ) β )] + β
Label η 1 = η − r 0 s ~ η − r 0 θ s ( ( 1 − p ) β ) \eta_{1}=\eta-r_{0}\widetilde{s}\eta-r_{0}\theta s((1-p)\beta) η 1 = η − r 0 s η − r 0 θ s (( 1 − p ) β ) . Then η 1 ∈ p R ( 1 − e ) \eta_{1}\in pR(1-e) η 1 ∈ pR ( 1 − e ) .
Also label c 1 = c + r 0 − r 0 s ~ c c_{1}=c+r_{0}-r_{0}\widetilde{s}c c 1 = c + r 0 − r 0 s c . Then note that since r 0 = r 0 p ∈ p R p r_{0}=r_{0}p\in pRp r 0 = r 0 p ∈ pR p we have c 1 = c + r 0 − r 0 s ~ c = c + r 0 ( p − s ~ c ) ∈ U ( p R p ) c_{1}=c+r_{0}-r_{0}\widetilde{s}c=c+r_{0}(p-\widetilde{s}c)\in U(pRp) c 1 = c + r 0 − r 0 s c = c + r 0 ( p − s c ) ∈ U ( pR p ) .
Now c 1 c_{1} c 1 being invertible in p R p pRp pR p and ( θ , θ ′ ) (\theta,\theta^{\prime}) ( θ , θ ′ ) being an invertible block in p R e ⟹ ( c 1 θ , θ ′ c 1 − 1 ) pRe\Longrightarrow(c_{1}\theta,\theta^{\prime}c_{1}^{-1}) pR e ⟹ ( c 1 θ , θ ′ c 1 − 1 ) is an invertible block in p R e pRe pR e since
( θ ′ c 1 − 1 ) ( c 1 θ ) = θ ′ θ = e and ( c 1 θ ) ( θ ′ c 1 − 1 ) = c 1 ( θ θ ′ ) c 1 − 1 = p (\theta^{\prime}c_{1}^{-1})(c_{1}\theta)=\theta^{\prime}\theta=e
\quad\text{and}\quad
(c_{1}\theta)(\theta^{\prime}c_{1}^{-1})=c_{1}(\theta\theta^{\prime})c_{1}^{-1}=p ( θ ′ c 1 − 1 ) ( c 1 θ ) = θ ′ θ = e and ( c 1 θ ) ( θ ′ c 1 − 1 ) = c 1 ( θ θ ′ ) c 1 − 1 = p
Then we have M + r ( 1 − s M ) = c 1 θ + η 1 + β M+r(1-sM)=c_{1}\theta+\eta_{1}+\beta M + r ( 1 − s M ) = c 1 θ + η 1 + β where c 1 θ c_{1}\theta c 1 θ is an invertible block in p R e pRe pR e and β \beta β is an invertible block in ( 1 − p ) R ( 1 − e ) (1-p)R(1-e) ( 1 − p ) R ( 1 − e ) and η 1 ∈ p R ( 1 − e ) \eta_{1}\in pR(1-e) η 1 ∈ pR ( 1 − e ) . Then by Lemma 2.14 (ii)
M + r ( 1 − s M ) = c 1 θ + η 1 + β ∈ U ( R ) M+r(1-sM)=c_{1}\theta+\eta_{1}+\beta\in U(R) M + r ( 1 − s M ) = c 1 θ + η 1 + β ∈ U ( R )
Since s s s is arbitrary, by Lemma 3.10 we have sr R ( M ) = sr R ( c θ + η + β ) = 1 \operatorname{sr}_{R}(M)=\operatorname{sr}_{R}(c\theta+\eta+\beta)=1 sr R ( M ) = sr R ( c θ + η + β ) = 1 for any η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) .
Conversely, let sr R ( c θ + η + β ) = 1 \operatorname{sr}_{R}(c\theta+\eta+\beta)=1 sr R ( c θ + η + β ) = 1 for every η ∈ p R ( 1 − e ) \eta\in pR(1-e) η ∈ pR ( 1 − e ) . Since ( θ ′ + β ′ ) (\theta^{\prime}+\beta^{\prime}) ( θ ′ + β ′ ) is invertible using Block Lemma, we can write
( c θ + η + β ) ( θ ′ + β ′ ) = c + η β ′ + ( 1 − p ) = ( 1 + η β ′ ) ( c + 1 − p ) (c\theta+\eta+\beta)(\theta^{\prime}+\beta^{\prime})=c+\eta\beta^{\prime}+(1-p)=(1+\eta\beta^{\prime})(c+1-p) ( c θ + η + β ) ( θ ′ + β ′ ) = c + η β ′ + ( 1 − p ) = ( 1 + η β ′ ) ( c + 1 − p )
and since, multiplication by units does not change the stable range of an element we have sr ( c + ( 1 − p ) ) = 1 \operatorname{sr}(c+(1-p))=1 sr ( c + ( 1 − p )) = 1 . Then by Lemma 3.10, for each s ∈ R s\in R s ∈ R there exists r ∈ R r\in R r ∈ R such that
( c + 1 − p ) + r ( 1 − s ( c + 1 − p ) ) ∈ U ( R ) (1) (c+1-p)+r(1-s(c+1-p))\in U(R) \tag*{(1)} ( c + 1 − p ) + r ( 1 − s ( c + 1 − p )) ∈ U ( R ) (1)
Fix arbitrary s ∈ p R p s \in pRp s ∈ pR p . We prove that there exists r ∈ p R p r \in pRp r ∈ pR p such that c + r ( p − s c ) ∈ U ( p R p ) c + r(p - sc) \in U(pRp) c + r ( p − sc ) ∈ U ( pR p ) .
Label s ′ = s + ( 1 − p ) s' = s + (1 - p) s ′ = s + ( 1 − p ) . Then there exists r 1 ∈ R r_1 \in R r 1 ∈ R such that
V = ( c + 1 − p ) + r 1 ( 1 − s ′ ( c + 1 − p ) ) ∈ U ( R ) V = (c + 1 - p) + r_1(1 - s'(c + 1 - p)) \in U(R) V = ( c + 1 − p ) + r 1 ( 1 − s ′ ( c + 1 − p )) ∈ U ( R )
Observe that s ′ ( c + ( 1 − p ) ) = s c + ( 1 − p ) ⟹ 1 − s ′ ( c + ( 1 − p ) ) = p − s c ∈ p R p s'(c + (1 - p)) = sc + (1 - p) \Longrightarrow1 - s'(c + (1 - p)) = p - sc \in pRp s ′ ( c + ( 1 − p )) = sc + ( 1 − p ) ⟹ 1 − s ′ ( c + ( 1 − p )) = p − sc ∈ pR p .
Then,
V = c + ( 1 − p ) + r 1 ( p − s c ) = c + ( 1 − p ) + ( p r 1 + r 1 − p r 1 ) ( p − s c ) = c + ( p r 1 p ) ( p − s c ) + ( 1 − p ) r 1 ( p − s c ) + ( 1 − p ) ∈ U ( R ) . \begin{aligned}
V &= c + (1 - p) + r_1(p - sc) \\
&= c + (1 - p) + (pr_1 + r_1 - pr_1)(p - sc) \\
&= c + (pr_1p)(p - sc) + (1 - p)r_1(p - sc) + (1 - p) \in U(R).
\end{aligned} V = c + ( 1 − p ) + r 1 ( p − sc ) = c + ( 1 − p ) + ( p r 1 + r 1 − p r 1 ) ( p − sc ) = c + ( p r 1 p ) ( p − sc ) + ( 1 − p ) r 1 ( p − sc ) + ( 1 − p ) ∈ U ( R ) .
Since ( 1 − p ) (1 - p) ( 1 − p ) is invertible in ( 1 − p ) R ( 1 − p ) (1 - p)R(1 - p) ( 1 − p ) R ( 1 − p ) , therefore by the converse of Prop 2.16, c + ( p r 1 p ) ( p − s c ) ∈ U ( p R p ) c + (pr_1p)(p - sc) \in U(pRp) c + ( p r 1 p ) ( p − sc ) ∈ U ( pR p ) .
Hence by Lemma 3.10 we have sr p R p ( c ) = 1 \operatorname{sr}_{pRp}(c) = 1 sr pR p ( c ) = 1 .
(1)(ii) The proof for the second part comes directly from (i). Label p 1 = e p_1 = e p 1 = e and e 1 = p e_1 = p e 1 = p , θ 1 = θ ′ \theta_1 = \theta' θ 1 = θ ′ and θ 1 ′ = θ \theta_1' = \theta θ 1 ′ = θ , β 1 = β ′ \beta_1 = \beta' β 1 = β ′ and β 1 ′ = β \beta_1' = \beta β 1 ′ = β . Then ( θ 1 , θ 1 ′ ) (\theta_1, \theta_1') ( θ 1 , θ 1 ′ ) is an invertible block in p 1 R e 1 p_1Re_1 p 1 R e 1 and ( β 1 , β 1 ′ ) (\beta_1, \beta_1') ( β 1 , β 1 ′ ) is an invertible block in ( 1 − p 1 ) R ( 1 − e 1 ) (1-p_1)R(1-e_1) ( 1 − p 1 ) R ( 1 − e 1 ) . Choose c 1 = a ∈ e R e = p 1 R p 1 c_1 = a \in eRe = p_1Rp_1 c 1 = a ∈ e R e = p 1 R p 1 . Then for any η 1 = η ~ ∈ p 1 R ( 1 − e 1 ) \eta_1 = \widetilde{\eta} \in p_1R(1-e_1) η 1 = η ∈ p 1 R ( 1 − e 1 ) by (1)(i) we get
sr p 1 R p 1 ( c 1 ) = 1 ⟺ sr R ( c 1 θ 1 + η 1 + β 1 ) = 1 \operatorname{sr}_{p_1Rp_1}(c_1) = 1 \Longleftrightarrow\operatorname{sr}_{R}(c_1\theta_1 + \eta_1 + \beta_1) = 1 sr p 1 R p 1 ( c 1 ) = 1 ⟺ sr R ( c 1 θ 1 + η 1 + β 1 ) = 1
In other words, for η ~ ∈ e R ( 1 − p ) \widetilde{\eta} \in eR(1-p) η ∈ e R ( 1 − p ) we get
sr e R e ( a ) = 1 ⟺ sr R ( a θ ′ + η ~ + β ′ ) = 1 \operatorname{sr}_{eRe}(a) = 1 \Longleftrightarrow\operatorname{sr}_{R}(a\theta' + \widetilde{\eta} + \beta') = 1 sr e R e ( a ) = 1 ⟺ sr R ( a θ ′ + η + β ′ ) = 1
(2)(i) and (ii): The proof for (2)(i) can be obtained by taking any element c ∈ p R p c \in pRp c ∈ pR p and considering the element a = θ ′ c θ ∈ e R e a = \theta'c\theta\in eRe a = θ ′ c θ ∈ e R e and following (1)(ii).
Additionally in each of these subcases, it is straightforward to see that sr R ( a ) = 1 \operatorname{sr}_{R}(a) = 1 sr R ( a ) = 1 (or, sr R ( c ) = 1 \operatorname{sr}_{R}(c) = 1 sr R ( c ) = 1 ).
As a consequence we recover the Suspension Theorem in [21 ] .
Theorem 3.12 ([21 ] , Theorem 6.2). Let e , f ∈ idem ( R ) e, f \in\operatorname{idem}(R) e , f ∈ idem ( R ) be such that e + f = 1 e + f = 1 e + f = 1 , and let a ∈ e R e a \in eRe a ∈ e R e , γ ∈ f R e \gamma\in fRe γ ∈ f R e . Then sr e R e ( a ) = 1 ⟺ sr R ( a + γ + f ) = 1 \operatorname{sr}_{eRe}(a) = 1 \Longleftrightarrow\operatorname{sr}_{R}(a + \gamma+ f) = 1 sr e R e ( a ) = 1 ⟺ sr R ( a + γ + f ) = 1 .
Proof. We can write a + γ + f = ( a + f ) ( 1 + γ ) a+\gamma+f = (a+f)(1+\gamma) a + γ + f = ( a + f ) ( 1 + γ ) where ( 1 + γ ) ∈ U ( R ) (1+\gamma) \in U(R) ( 1 + γ ) ∈ U ( R ) . Then using idempotents e = p e = p e = p , θ = e \theta= e θ = e , η = 0 \eta= 0 η = 0 and the identity block β = f \beta= f β = f , Theorem 3.11 (2)(ii) above gives us the equivalence.
Theorem 3.13 ([21 ] , Theorem 6.6). Let e , f ∈ idem ( R ) e, f \in\operatorname{idem}(R) e , f ∈ idem ( R ) with e + f = 1 e + f = 1 e + f = 1 .
(1) If a ∈ e R e a \in eRe a ∈ e R e , γ ∈ f R e \gamma\in fRe γ ∈ f R e and b ∈ f R f b \in fRf b ∈ f R f are such that sr e R e ( a ) = 1 \operatorname{sr}_{eRe}(a) = 1 sr e R e ( a ) = 1 and sr f R f ( b ) = 1 \operatorname{sr}_{fRf}(b) = 1 sr f R f ( b ) = 1 , then
sr R ( a + γ + b ) = 1 \operatorname{sr}_{R}(a + \gamma+ b) = 1 sr R ( a + γ + b ) = 1 (2) If a ∈ e R e a \in eRe a ∈ e R e , γ ∈ f R e \gamma\in fRe γ ∈ f R e and u ∈ U ( f R f ) u \in U(fRf) u ∈ U ( f R f ) , then
sr e R e ( a ) = 1 ⟺ sr R ( a + γ + u ) = 1. \operatorname{sr}_{eRe}(a) = 1 \Longleftrightarrow\operatorname{sr}_{R}(a + \gamma+ u) = 1. sr e R e ( a ) = 1 ⟺ sr R ( a + γ + u ) = 1. Proof. (1) Since sr e R e ( a ) = 1 \operatorname{sr}_{eRe}(a) = 1 sr e R e ( a ) = 1 , Theorem 3.12 gives sr R ( a + γ + f ) = 1 \operatorname{sr}_{R}(a+\gamma+f) = 1 sr R ( a + γ + f ) = 1 . Also using Theorem 3.11 with η = 0 \eta= 0 η = 0 gives sr R ( e + b ) = 1 \operatorname{sr}_{R}(e + b) = 1 sr R ( e + b ) = 1 . Now using the expression ( a + γ + f ) ( e + b ) = a e + γ e + f b = a + γ + b (a + \gamma+ f)(e + b) = ae + \gamma e + fb = a + \gamma+ b ( a + γ + f ) ( e + b ) = a e + γ e + f b = a + γ + b , and that products of stable range one elements are also stable range one [9 Lemma 17] , the proof follows.
(2) Straightforward.
Using the Suspension Theorem, we provide an analogue to Ara and Goodearl’s result [2 Lemma 2] on stable range of corners through subequivalence.
Corollary 3.14. Let e , f ∈ Idem ( R ) e, f \in\operatorname{Idem}(R) e , f ∈ Idem ( R ) such that e ≲ f e \lesssim f e ≲ f . If sr ( f R f ) = 1 \operatorname{sr}(fRf)=1 sr ( f R f ) = 1 , then sr ( e R e ) = 1 \operatorname{sr}(eRe)=1 sr ( e R e ) = 1 .
Proof. There exists elements r ∈ e R f r \in eRf r ∈ e R f and s ∈ f R e s \in fRe s ∈ f R e such that
Label S = f R f S=fRf S = f R f with identity f f f and take g = ( s ) ( r ) ∈ S g=(s)(r)\in S g = ( s ) ( r ) ∈ S . Then clearly, g 2 = g g^2=g g 2 = g and g f = f g = g gf=fg=g g f = f g = g . By the definition of g g g , we also have r = r ( g ) ∈ R g r=r(g)\in Rg r = r ( g ) ∈ R g and s = ( g ) s ∈ g R s=(g)s\in gR s = ( g ) s ∈ g R . Now consider the corner ring g S g gSg g S g in the ambient ring S S S . Define h = f − g h=f-g h = f − g . Then h ∈ Idem ( S ) h\in\operatorname{Idem}(S) h ∈ Idem ( S ) and h g = g h = 0 hg=gh=0 h g = g h = 0 .
We prove sr ( e R e ) = 1 \operatorname{sr}(eRe)=1 sr ( e R e ) = 1 . Suppose x a + b = e xa+b=e x a + b = e for arbitrary a , b , x ∈ e R e a,b,x\in eRe a , b , x ∈ e R e .
Consider the following elements in g S g gSg g S g .
a ′ = ( s ) a ( r ) b ′ = ( s ) b ( r ) x ′ = ( s ) x ( r ) a'=(s)a(r) \qquad b'=(s)b(r) \qquad x'=(s)x(r) a ′ = ( s ) a ( r ) b ′ = ( s ) b ( r ) x ′ = ( s ) x ( r )
( ⋆ ) (\star) ( ⋆ ) Then, x ′ a ′ + b ′ = ( s ) ( x a + b ) ( r ) = s e r = s r = g x'a'+b'=(s)(xa+b)(r)=ser=sr=g x ′ a ′ + b ′ = ( s ) ( x a + b ) ( r ) = ser = sr = g
Consider the element a ′ + h ∈ S a'+h\in S a ′ + h ∈ S . Then sr ( S ) = 1 ⟹ sr S ( a ′ + h ) = 1 \operatorname{sr}(S)=1\Longrightarrow\operatorname{sr}_{S}(a'+h)=1 sr ( S ) = 1 ⟹ sr S ( a ′ + h ) = 1 . Since g g g is the identity in g S g gSg g S g and f − g f-g f − g is the identity in ( f − g ) S ( f − g ) (f-g)S(f-g) ( f − g ) S ( f − g ) , by Theorem 3.11, we have
sr g S g ( a ′ ) = 1 ⟺ sr S ( a ′ + h ) = 1 \operatorname{sr}_{gSg}(a')=1 \Longleftrightarrow\operatorname{sr}_{S}(a'+h)=1 sr g S g ( a ′ ) = 1 ⟺ sr S ( a ′ + h ) = 1
Thus sr g S g ( a ′ ) = 1 ⟹ \operatorname{sr}_{gSg}(a')=1\Longrightarrow sr g S g ( a ′ ) = 1 ⟹ e.q. ( ⋆ ) (\star) ( ⋆ ) is reducible. Thus there exists y ∈ g S g y\in gSg y ∈ g S g such that a ′ + y b ′ ∈ U ( g S g ) a'+yb'\in U(gSg) a ′ + y b ′ ∈ U ( g S g ) . Label u g = a ′ + y b ′ ∈ U ( g S g ) u_g=a'+yb'\in U(gSg) u g = a ′ + y b ′ ∈ U ( g S g ) .
Then, since y , b ′ ∈ g S g y,b'\in gSg y , b ′ ∈ g S g and g = s r g=sr g = sr we have
r ( a ′ + y b ′ ) s = r a ′ s + ( r y ) ( g ) ( b ′ s ) = ( r a ′ s ) + ( r y s ) ( r b ′ s ) = a + ( r y s ) b r(a'+yb')s=ra's+(ry)(g)(b's)=(ra's)+(rys)(rb's)=a+(rys)b r ( a ′ + y b ′ ) s = r a ′ s + ( r y ) ( g ) ( b ′ s ) = ( r a ′ s ) + ( r y s ) ( r b ′ s ) = a + ( r y s ) b
where r ( a ′ + y b ′ ) s = r u g s ∈ U ( e R e ) r(a'+yb')s=ru_gs\in U(eRe) r ( a ′ + y b ′ ) s = r u g s ∈ U ( e R e ) since
r ( u g ) s r ( u g − 1 ) s = r ( u g ) g ( u g − 1 ) s = r g s = e = r ( u g − 1 ) g ( u g ) s = r ( u g − 1 ) s r ( u g ) s r(u_g)s\,r(u_g^{-1})s=r(u_g)g(u_g^{-1})s=rgs=e=r(u_g^{-1})g(u_g)s=r(u_g^{-1})sr(u_g)s r ( u g ) s r ( u g − 1 ) s = r ( u g ) g ( u g − 1 ) s = r g s = e = r ( u g − 1 ) g ( u g ) s = r ( u g − 1 ) sr ( u g ) s
Thus a + ( r y s ) b ∈ U ( e R e ) ⟹ sr ( e R e ) = 1 a+(rys)b\in U(eRe)\Longrightarrow\operatorname{sr}(eRe)=1 a + ( r y s ) b ∈ U ( e R e ) ⟹ sr ( e R e ) = 1 .
:::
Remark 3.15. A rather neat consequence is rediscovered through the above corollary. Suppose e e e is a full idempotent in a ring R R R such that e R e eRe e R e has stable range one. Then e e e being full, by Ara-Goodearl [2 p. 3] , for some n ∈ N n\in\mathbb{N} n ∈ N we have
E 11 ≲ e ^ = diag ( e , … , e ) ∈ M n ( R ) . E_{11}\lesssim\hat{e}=\operatorname{diag}(e,\ldots,e)\in M_n(R). E 11 ≲ e ^ = diag ( e , … , e ) ∈ M n ( R ) . Now e ^ M n ( R ) e ^ = M n ( e R e ) \hat{e}M_n(R)\hat{e}=M_n(eRe) e ^ M n ( R ) e ^ = M n ( e R e ) has stable range one, since e R e eRe e R e does, due to Vaserstein. Then, applying the above corollary, we have sr ( R ) = 1 \operatorname{sr}(R)=1 sr ( R ) = 1 .
Block Invertibility and Special-Clean elements
Abrams and Rangaswamy [1 ] coined the term special clean for rings whose elements decompose in the following way:
An element a ∈ R a\in R a ∈ R that can be written in the form a = e + u a=e+u a = e + u with e 2 = e e^2=e e 2 = e , u ∈ U ( R ) u\in U(R) u ∈ U ( R ) and a R ∩ e R = { 0 } aR\cap eR=\{0\} a R ∩ e R = { 0 } .
They used it to compare regularity conditions for Leavitt path algebras. Camillo and Khurana [8 ] proved that on the entire ambient ring, unit regularity is equivalent to special clean. However, locally a unit regular element is not necessary special-clean (or even clean), and there are numerous such examples, one of which was provided in [20 ] where Khurana and Lam showed the element
( 1 + x y x 2 0 0 ) \begin{pmatrix}
1+xy & x^2\\
0 & 0
\end{pmatrix} ( 1 + x y 0 x 2 0 )
is unit-regular but not clean over K = k [ x , y ] K = k[x,y] K = k [ x , y ] for any commutative domain k k k . Many authors investigated additional conditions under which unit-regular elements are special clean [23 , 22 , 29 , 40 ] . Nielsen and Ster gave an example in [33 ] where they showed that even when a , 1 − a a,1-a a , 1 − a is unit regular, even then a a a need not be clean. They also used regularity of the element a a a say a = a r a a=ara a = a r a and unit regularity of the specific corner element e a e eae e a e where e = r a e=ra e = r a , to obtain a stronger condition than special clean (see below).
Theorem 4.1. [33 Theorem 3.14] Let a a a be an element of a ring R R R . The following are equivalent:
The exists r ∈ R r\in R r ∈ R with a r a = a ara=a a r a = a , and writing e : = r a ∈ idem ( R ) e:=ra\in\operatorname{idem}(R) e := r a ∈ idem ( R ) , the element e a e eae e a e is unit-regular in e R e eRe e R e .
There exist e ∈ idem ( R ) e\in\operatorname{idem}(R) e ∈ idem ( R ) and u ∈ U ( R ) u\in U(R) u ∈ U ( R ) such that a = e + u a=e+u a = e + u , a R ∩ e R = ( 0 ) aR\cap eR=(0) a R ∩ e R = ( 0 ) , and a 2 R ∩ a e R = ( 0 ) a^{2}R\cap aeR=(0) a 2 R ∩ a e R = ( 0 ) .
Mary and Patricio identified where stable range can be applied to evoke special-cleanness of a a a . In particular, they obtained that when a v − 1 a = a av^{-1}a=a a v − 1 a = a , and the Pierce decomposition of v v v with respect to f = a v − 1 f=av^{-1} f = a v − 1 is given by v ≡ ( v 1 v 2 v 3 v 4 ) v\equiv\begin{pmatrix}v_{1}&v_{2}\\v_{3}&v_{4}\end{pmatrix} v ≡ ( v 1 v 3 v 2 v 4 ) and if there exists a γ ∈ f ˉ R f \gamma\in\bar{f}Rf γ ∈ f ˉ R f (resp. σ ∈ f R f ˉ \sigma\in fR\bar{f} σ ∈ f R f ˉ ) such that v 4 + γ v 2 ∈ U ( f ˉ R f ˉ ) v_{4}+\gamma v_{2}\in U(\bar{f}R\bar{f}) v 4 + γ v 2 ∈ U ( f ˉ R f ˉ ) (resp. v 4 + v 3 σ ∈ U ( f ˉ R f ˉ ) v_{4}+v_{3}\sigma\in U(\bar{f}R\bar{f}) v 4 + v 3 σ ∈ U ( f ˉ R f ˉ ) ) then a a a is special clean. Subsequently, they proved the following:
Corollary 4.2. [29 Cor 3.4] Let a v − 1 a = a av^{-1}a=a a v − 1 a = a and f = a v − 1 f=av^{-1} f = a v − 1 . If sr ( v 4 ) = 1 \operatorname{sr}(v_{4})=1 sr ( v 4 ) = 1 in f ˉ R f ˉ \bar{f}R\bar{f} f ˉ R f ˉ (in particular if v 4 v_{4} v 4 is unit-regular in f ˉ R f ˉ \bar{f}R\bar{f} f ˉ R f ˉ ), then a a a is special clean.
We give an element-wise characterization under which an unit-regular element is special-clean. Recall Bass’s stable range condition as follows: A ring R R R has stable range 1 if every left unimodular pair ( a , b ) (a,b) ( a , b ) is reducible, that is whenever r a + s b = 1 ⇒ ( a + x b ) ra+sb=1\Rightarrow(a+xb) r a + s b = 1 ⇒ ( a + x b ) is invertible for some x ∈ R x\in R x ∈ R for arbitrary a , b ∈ R a,b\in R a , b ∈ R . On a specific element a a a , sr ( a ) = 1 \operatorname{sr}(a)=1 sr ( a ) = 1 is defined as every left (right) unimodular pair of the form ( a , y ) (a,y) ( a , y ) is left (right) reducible for arbitrary y ∈ R y\in R y ∈ R . We observed that the missing link is simply reducibility for a specific unimodular pair ( u 4 , − ) (u_{4},-) ( u 4 , − ) where a = a u a a=aua a = a u a and the Pierce decomposition of u u u is
u = ( a u ) u ( a u ) ⏟ u 1 + ( a u ) u ( 1 − a u ) ⏟ u 2 + ( 1 − a u ) u ( a u ) ⏟ u 3 + ( 1 − a u ) u ( 1 − a u ) ⏟ u 4 u=\underbrace{(au)u(au)}_{u_{1}}+\underbrace{(au)u(1-au)}_{u_{2}}+\underbrace{(1-au)u(au)}_{u_{3}}+\underbrace{(1-au)u(1-au)}_{u_{4}} u = u 1 ( a u ) u ( a u ) + u 2 ( a u ) u ( 1 − a u ) + u 3 ( 1 − a u ) u ( a u ) + u 4 ( 1 − a u ) u ( 1 − a u )
We show below that a necessary and sufficient condition for a a a to be special clean requires that the specific pair ( u 4 , ψ ) (u_{4},\psi) ( u 4 , ψ ) where u 4 = ( 1 − a u ) u ( 1 − a u ) u_{4}=(1-au)u(1-au) u 4 = ( 1 − a u ) u ( 1 − a u ) , ψ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) \psi=(1-au)(u^{-1}au)u(1-au) ψ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) which is left unimodular in the corner ring ( 1 − a u ) R ( 1 − a u ) (1-au)R(1-au) ( 1 − a u ) R ( 1 − a u ) is left reducible: i.e. α + z ψ ∈ U ( S ) \alpha+z\psi\in U(S) α + z ψ ∈ U ( S ) for some z ∈ S z\in S z ∈ S .
We first consider the following lemma for any element a = a u a a=aua a = a u a :
Lemma 4.3. Let a ∈ R a\in R a ∈ R such that a = a u a a=aua a = a u a with u ≡ ( u 1 u 2 u 3 u 4 ) u\equiv\begin{pmatrix}u_{1}&u_{2}\\u_{3}&u_{4}\end{pmatrix} u ≡ ( u 1 u 3 u 2 u 4 ) as above. Then for any fixed σ ∈ ( 1 − a u ) R ( a u ) \sigma\in(1-au)R(au) σ ∈ ( 1 − a u ) R ( a u ) the following are equivalent:
(i) u ( 1 − a u ) + ( a u ) + σ ∈ U ( R ) u(1-au)+(au)+\sigma\in U(R) u ( 1 − a u ) + ( a u ) + σ ∈ U ( R )
(ii) u 4 − σ u 2 ∈ U ( ( 1 − a u ) R ( 1 − a u ) ) u_{4}-\sigma u_{2}\in U((1-au)R(1-au)) u 4 − σ u 2 ∈ U (( 1 − a u ) R ( 1 − a u )) .
Proof. Observe that u 4 − σ u 2 = ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) u_{4}-\sigma u_{2}=(1-au)u(1-au)-\sigma u(1-au) u 4 − σ u 2 = ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) .
( i ) ⟺ ( i i ) (i)\Longleftrightarrow(ii) ( i ) ⟺ ( ii ) : Let σ ∈ ( 1 − a u ) R ( a u ) \sigma\in(1-au)R(au) σ ∈ ( 1 − a u ) R ( a u ) be such that
u ( 1 − a u ) + ( a u ) + σ ∈ U ( R ) u(1-au)+(au)+\sigma\in U(R) u ( 1 − a u ) + ( a u ) + σ ∈ U ( R )
Write p = a u p=au p = a u . Label α = ( 1 − p ) u ( 1 − p ) \alpha=(1-p)u(1-p) α = ( 1 − p ) u ( 1 − p ) and γ = p u ( 1 − p ) ∈ p R ( 1 − p ) \gamma=pu(1-p)\in pR(1-p) γ = p u ( 1 − p ) ∈ pR ( 1 − p ) . Label V = u ( 1 − a u ) + ( a u ) + σ = ( α + γ ) + p + σ V=u(1-au)+(au)+\sigma=(\alpha+\gamma)+p+\sigma V = u ( 1 − a u ) + ( a u ) + σ = ( α + γ ) + p + σ . Now σ ∈ ( 1 − p ) R p ⟹ ( 1 − p ) σ p = σ \sigma\in(1-p)Rp\Longrightarrow(1-p)\sigma p=\sigma σ ∈ ( 1 − p ) R p ⟹ ( 1 − p ) σ p = σ . Similarly, for γ \gamma γ , we have p γ ( 1 − p ) = γ p\gamma(1-p)=\gamma p γ ( 1 − p ) = γ . Then clearly, σ 2 = 0 = γ 2 ⟹ ( 1 + σ ) , ( 1 + γ ) ∈ U ( R ) \sigma^{2}=0=\gamma^{2}\Longrightarrow(1+\sigma),(1+\gamma)\in U(R) σ 2 = 0 = γ 2 ⟹ ( 1 + σ ) , ( 1 + γ ) ∈ U ( R ) . Then, expanding the product ( 1 + σ ) ( α − σ γ + p ) ( 1 + γ ) (1+\sigma)(\alpha-\sigma\gamma+p)(1+\gamma) ( 1 + σ ) ( α − σ γ + p ) ( 1 + γ ) we get
( 1 + σ ) [ ( α − σ γ + p ) ( 1 + γ ) ] = ( 1 + σ ) [ α − σ γ + p + ( α − σ γ ) γ + p γ ] = α − σ γ + p + γ + σ ( α − σ γ ) + σ p + σ γ = α + σ + γ + p \begin{aligned}
(1+\sigma)\left[(\alpha-\sigma\gamma+p)(1+\gamma)\right]=(1+\sigma)\left[\alpha-\sigma\gamma+p+(\alpha-\sigma\gamma)\gamma+p\gamma\right] \\
&=\alpha-\sigma\gamma+p+\gamma+\sigma(\alpha-\sigma\gamma)+\sigma p+\sigma\gamma\\
&=\alpha+\sigma+\gamma+p
\end{aligned} ( 1 + σ ) [ ( α − σ γ + p ) ( 1 + γ ) ] = ( 1 + σ ) [ α − σ γ + p + ( α − σ γ ) γ + p γ ] = α − σ γ + p + γ + σ ( α − σ γ ) + σ p + σ γ = α + σ + γ + p
Then 1 + σ 1+\sigma 1 + σ being invertible shows that
V ∈ U ( R ) ⟺ α − σ γ ⏟ ( 1 − p ) R ( 1 − p ) + p ∈ U ( R ) V\in U(R)\Longleftrightarrow\underbrace{\alpha-\sigma\gamma}_{(1-p)R(1-p)}+p\in U(R) V ∈ U ( R ) ⟺ ( 1 − p ) R ( 1 − p ) α − σ γ + p ∈ U ( R )
Since p ∈ U ( p R p ) p\in U(pRp) p ∈ U ( pR p ) and α − σ γ + p ∈ U ( R ) \alpha-\sigma\gamma+p\in U(R) α − σ γ + p ∈ U ( R ) , then by Proposition 2.16 , α − σ γ \alpha-\sigma\gamma α − σ γ must have a block inverse in ( 1 − p ) R ( 1 − p ) (1-p)R(1-p) ( 1 − p ) R ( 1 − p ) . in other words,
u 4 − σ u 2 = α − σ γ ∈ U ( ( 1 − p ) R ( 1 − p ) ) u_4-\sigma u_2=\alpha-\sigma\gamma\in U((1-p)R(1-p)) u 4 − σ u 2 = α − σ γ ∈ U (( 1 − p ) R ( 1 − p ))
The above line is actually reversible using Lemma 2.14 and Proposition 2.16 that is,
( u 4 − σ u 2 ) + p ∈ U ( R ) ⟺ ( u 4 − σ u 2 ) ∈ U ( ( 1 − p ) R ( 1 − p ) ) (u_4-\sigma u_2)+p\in U(R)\Longleftrightarrow(u_4-\sigma u_2)\in U((1-p)R(1-p)) ( u 4 − σ u 2 ) + p ∈ U ( R ) ⟺ ( u 4 − σ u 2 ) ∈ U (( 1 − p ) R ( 1 − p ))
Thus the converse also holds. This proves (i) ⟺ \Longleftrightarrow ⟺ (ii), with the same pair u u u and σ \sigma σ on both sides.
For our construction of clean elements, we follow the strategy given in (Corollary 27,[37]) and ([27] Corollary 2.3.), namely for an element a ∈ R a\in R a ∈ R , a a a is clean if and only if there exists outer inverses x , y x,y x , y of a a a and ( 1 − a ) (1-a) ( 1 − a ) resp. such that a x + ( 1 − a ) y = 1 ax+(1-a)y=1 a x + ( 1 − a ) y = 1 and ( x − y ) (x-y) ( x − y ) is invertible.
Theorem 4.4. Let a ∈ R a\in R a ∈ R . Then the following are equivalent:
(i) a = f + v a=f+v a = f + v for f 2 = f f^2=f f 2 = f and unit v ∈ U ( R ) v\in U(R) v ∈ U ( R ) such that a R ∩ f R = { 0 } aR\cap fR=\{0\} a R ∩ f R = { 0 }
(ii) There exists unit u ∈ U ( R ) u\in U(R) u ∈ U ( R ) and σ ∈ ( 1 − a u ) R ( a u ) \sigma\in(1-au)R(au) σ ∈ ( 1 − a u ) R ( a u ) such that a = a u a a=aua a = a u a and
( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U ( ( 1 − a u ) R ( 1 − a u ) ) (1-au)u(1-au)-\sigma u(1-au)\in U((1-au)R(1-au)) ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U (( 1 − a u ) R ( 1 − a u )) (iii) there exists u ∈ U ( R ) u\in U(R) u ∈ U ( R ) with a = a u a a=aua a = a u a such that the pair ( α , ψ ) (\alpha,\psi) ( α , ψ ) , where
α = ( 1 − a u ) u ( 1 − a u ) , ψ = μ γ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) , \alpha=(1-au)u(1-au),\qquad\psi=\mu\gamma=(1-au)(u^{-1}au)u(1-au), α = ( 1 − a u ) u ( 1 − a u ) , ψ = μ γ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) , which is left unimodular in S = ( 1 − a u ) R ( 1 − a u ) S=(1-au)R(1-au) S = ( 1 − a u ) R ( 1 − a u ) , is left reducible: α + z ψ ∈ U ( S ) \alpha+z\psi\in U(S) α + z ψ ∈ U ( S ) for some z ∈ S z\in S z ∈ S .
Proof. (ii) ⟹ \Longrightarrow ⟹ (i): Let there be a unit u ∈ U ( R ) u\in U(R) u ∈ U ( R ) and σ ∈ ( 1 − a u ) R ( a u ) \sigma\in(1-au)R(au) σ ∈ ( 1 − a u ) R ( a u ) such that a = a u a a=aua a = a u a and
u ( 1 − a u ) + ( a u ) + σ ∈ U ( R ) u(1-au)+(au)+\sigma\in U(R) u ( 1 − a u ) + ( a u ) + σ ∈ U ( R )
We will find outer inverses x , y x,y x , y of a a a and ( 1 − a ) (1-a) ( 1 − a ) resp. with a x + ( 1 − a ) y = 1 ax+(1-a)y=1 a x + ( 1 − a ) y = 1 , such that ( x − y ) (x-y) ( x − y ) is invertible and a R ∩ ( x − y ) ( 1 − a x ) R = { 0 } aR\cap(x-y)(1-ax)R=\{0\} a R ∩ ( x − y ) ( 1 − a x ) R = { 0 } .
Define p = a u p=au p = a u , q = u a q=ua q = u a , e = p + a ( 1 − p ) e=p+a(1-p) e = p + a ( 1 − p ) , and y = 1 − p y=1-p y = 1 − p
Then clearly p 2 = p p^2=p p 2 = p , q 2 = q q^2=q q 2 = q , p a = a pa=a p a = a ;
Claim 1. e 2 = e e^2=e e 2 = e ; e a = a ea=a e a = a ; p e = e pe=e p e = e ;
{ ( 1 − p ) ( 1 − e ) = 1 − p ( 1 − e ) ( 1 − p ) = 1 − e \begin{cases}
(1-p)(1-e)=1-p\\
(1-e)(1-p)=1-e
\end{cases} { ( 1 − p ) ( 1 − e ) = 1 − p ( 1 − e ) ( 1 − p ) = 1 − e
Proof. Using p a = a pa=a p a = a and ( 1 − p ) a = 0 (1-p)a=0 ( 1 − p ) a = 0 we get
e 2 = ( p + a ( 1 − p ) ) ( p + a ( 1 − p ) ) = p 2 + p a ( 1 − p ) + a ( 1 − p ) p + a ( 1 − p ) a ( 1 − p ) = p + a ( 1 − p ) = p + ( e − p ) = e \begin{aligned}
e^2 &= (p+a(1-p))(p+a(1-p))\\
&=p^2+pa(1-p)+a(1-p)p+a(1-p)a(1-p)\\
&=p+a(1-p)=p+(e-p)=e
\end{aligned} e 2 = ( p + a ( 1 − p )) ( p + a ( 1 − p )) = p 2 + p a ( 1 − p ) + a ( 1 − p ) p + a ( 1 − p ) a ( 1 − p ) = p + a ( 1 − p ) = p + ( e − p ) = e
Using p a = a pa=a p a = a we also get
e a = ( p + a ( 1 − p ) ) a = p a + a ( 1 − p ) a = a ea=(p+a(1-p))a=pa+a(1-p)a=a e a = ( p + a ( 1 − p )) a = p a + a ( 1 − p ) a = a
Also, using p a = a pa=a p a = a we get p e = p + a ( 1 − p ) = e pe=p+a(1-p)=e p e = p + a ( 1 − p ) = e and therefore,
( 1 − p ) e = 0 ⟺ ( 1 − p ) ( 1 − e ) = 1 − p (1-p)e=0\Longleftrightarrow(1-p)(1-e)=1-p ( 1 − p ) e = 0 ⟺ ( 1 − p ) ( 1 − e ) = 1 − p
Finally we have e p = p ⟺ ( 1 − e ) p = 0 ⟺ ( 1 − e ) ( 1 − p ) = 1 − e ep=p\Longleftrightarrow(1-e)p=0\Longleftrightarrow(1-e)(1-p)=1-e e p = p ⟺ ( 1 − e ) p = 0 ⟺ ( 1 − e ) ( 1 − p ) = 1 − e .
Now label y = ( 1 − p ) y=(1-p) y = ( 1 − p ) . Then clearly, y a = ( 1 − p ) a = 0 ya=(1-p)a=0 y a = ( 1 − p ) a = 0 . Therefore, y ( 1 − a ) y = y 2 = y ⟹ y y(1-a)y=y^2=y\Longrightarrow y y ( 1 − a ) y = y 2 = y ⟹ y is an outer inverse of ( 1 − a ) (1-a) ( 1 − a ) . Finally, using e = p + a ( 1 − p ) e=p+a(1-p) e = p + a ( 1 − p ) gives ( 1 − e ) = ( 1 − p ) − a ( 1 − p ) = ( 1 − a ) ( 1 − p ) = ( 1 − a ) y (1-e)=(1-p)-a(1-p)=(1-a)(1-p)=(1-a)y ( 1 − e ) = ( 1 − p ) − a ( 1 − p ) = ( 1 − a ) ( 1 − p ) = ( 1 − a ) y . Thus
e + ( 1 − a ) y = 1 e+(1-a)y=1 e + ( 1 − a ) y = 1
We still need to obtain the outer inverse x x x of a a a such that a x = e ax=e a x = e which would imply that a x + ( 1 − a ) y = 1 ax+(1-a)y=1 a x + ( 1 − a ) y = 1 .
Define
α = ( 1 − p ) u ( 1 − p ) , γ = p u ( 1 − p ) , λ = ( 1 − p ) u − 1 ( 1 − p ) , μ = ( 1 − p ) u − 1 p \alpha=(1-p)u(1-p),\qquad\gamma=pu(1-p),\qquad\lambda=(1-p)u^{-1}(1-p),\qquad\mu=(1-p)u^{-1}p α = ( 1 − p ) u ( 1 − p ) , γ = p u ( 1 − p ) , λ = ( 1 − p ) u − 1 ( 1 − p ) , μ = ( 1 − p ) u − 1 p
Now by assumption, there exists σ ∈ ( 1 − p ) R p \sigma\in(1-p)Rp σ ∈ ( 1 − p ) R p such that
( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U ( ( 1 − a u ) R ( 1 − a u ) ) (1-au)u(1-au)-\sigma u(1-au)\in U((1-au)R(1-au)) ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U (( 1 − a u ) R ( 1 − a u ))
Then for the same σ \sigma σ and u u u by Lemma 4.3 we have an equivalent statement
u ( 1 − a u ) + ( a u ) + σ = α + γ + p + σ ∈ U ( R ) u(1-au)+(au)+\sigma=\alpha+\gamma+p+\sigma\in U(R) u ( 1 − a u ) + ( a u ) + σ = α + γ + p + σ ∈ U ( R )
Label V = α + γ + p + σ V=\alpha+\gamma+p+\sigma V = α + γ + p + σ . Post-multiplying V V V with p p p and ( 1 − p ) (1-p) ( 1 − p ) respectively gives
( V ) ( p ) = α p + γ p + σ p + p = σ + p (V)(p)=\alpha p+\gamma p+\sigma p+p=\sigma+p ( V ) ( p ) = α p + γ p + σ p + p = σ + p
and
( V ) ( 1 − p ) = α + γ = u ( 1 − a u ) = ( 1 − u a ) u ⋯ ( B ⋆ ) (V)(1-p)=\alpha+\gamma=u(1-au)=(1-ua)u\qquad\cdots\qquad(B^\star) ( V ) ( 1 − p ) = α + γ = u ( 1 − a u ) = ( 1 − u a ) u ⋯ ( B ⋆ )
Now label V p = σ + p = g 1 Vp=\sigma+p=g_1 V p = σ + p = g 1 and π = V p V − 1 \pi=VpV^{-1} π = V p V − 1 . Then g 1 , π ∈ Idem ( R ) g_1,\pi\in\operatorname{Idem}(R) g 1 , π ∈ Idem ( R ) as seen below.
g 1 2 = ( σ + p ) 2 ⋯ where σ ∈ ( 1 − p ) R p = σ 2 ⏟ 0 + σ p ⏟ σ + p σ ⏟ 0 + p = σ + p = g 1 \begin{aligned}
g_1^2&=(\sigma+p)^2\qquad\cdots\qquad\text{where }\sigma\in(1-p)Rp\\
&=\underbrace{\sigma^2}_{0}+\underbrace{\sigma p}_{\sigma}+\underbrace{p\sigma}_{0}+p=\sigma+p=g_1
\end{aligned} g 1 2 = ( σ + p ) 2 ⋯ where σ ∈ ( 1 − p ) R p = 0 σ 2 + σ σ p + 0 p σ + p = σ + p = g 1
And,
π 2 = ( V p V − 1 ) ( V p V − 1 ) = V p V − 1 = π . \pi^2=(VpV^{-1})(VpV^{-1})=VpV^{-1}=\pi. π 2 = ( V p V − 1 ) ( V p V − 1 ) = V p V − 1 = π .
We also have π g 1 = ( V p V − 1 ) ( V p ) = V p = g 1 \pi g_1=(VpV^{-1})(Vp)=Vp=g_1 π g 1 = ( V p V − 1 ) ( V p ) = V p = g 1 and
g 1 π = ( V p ) ( V p V − 1 ) = ( V p ) ( V p ) V − 1 = ( V p ) V − 1 = π g_1\pi=(Vp)(VpV^{-1})=(Vp)(Vp)V^{-1}=(Vp)V^{-1}=\pi g 1 π = ( V p ) ( V p V − 1 ) = ( V p ) ( V p ) V − 1 = ( V p ) V − 1 = π
From ( B ⋆ ) (B^\star) ( B ⋆ ) , ( 1 − π ) = V ( 1 − p ) V − 1 = ( 1 − u a ) u V − 1 ⟹ ( 1 − u a ) ( 1 − π ) = 1 − π ⟹ a π = a (1-\pi)=V(1-p)V^{-1}=(1-ua)uV^{-1}\Longrightarrow(1-ua)(1-\pi)=1-\pi\Longrightarrow a\pi=a ( 1 − π ) = V ( 1 − p ) V − 1 = ( 1 − u a ) u V − 1 ⟹ ( 1 − u a ) ( 1 − π ) = 1 − π ⟹ aπ = a .
Furthermore, from ( B ⋆ ) (B^\star) ( B ⋆ ) we also get
( 1 − π ) ( 1 − u a ) = V ( 1 − p ) V − 1 V ( 1 − p ) u − 1 = V ( 1 − p ) u − 1 = ( 1 − u a ) (1-\pi)(1-ua)=V(1-p)V^{-1}V(1-p)u^{-1}=V(1-p)u^{-1}=(1-ua) ( 1 − π ) ( 1 − u a ) = V ( 1 − p ) V − 1 V ( 1 − p ) u − 1 = V ( 1 − p ) u − 1 = ( 1 − u a )
⟹ π ( u a ) = π ⋯ ( C ⋆ ) \Longrightarrow\pi(ua)=\pi\qquad\cdots\qquad(C^\star) ⟹ π ( u a ) = π ⋯ ( C ⋆ )
Now choose x = π u e ∈ R e x=\pi ue\in Re x = π u e ∈ R e . Then a x = p e = e ax=pe=e a x = p e = e . Also x ( a x ) = x e = x x(ax)=xe=x x ( a x ) = x e = x and a x a = e a = a axa=ea=a a x a = e a = a from Claim 1. Hence x ∈ R e f ( a ) x\in Ref(a) x ∈ R e f ( a ) .
Set d = x − y d=x-y d = x − y . Then since ( 1 − p ) ( 1 − e ) = ( 1 − p ) (1-p)(1-e)=(1-p) ( 1 − p ) ( 1 − e ) = ( 1 − p ) we can write
d = p x ⏟ ∈ p R e + ( 1 − p ) x ⏟ ∈ ( 1 − p ) R e + ( − ( 1 − p ) ) ⏟ ∈ ( 1 − p ) R ( 1 − e ) d=\underbrace{px}_{\in pRe}+\underbrace{(1-p)x}_{\in(1-p)Re}+\underbrace{(-(1-p))}_{\in(1-p)R(1-e)} d = ∈ pR e p x + ∈ ( 1 − p ) R e ( 1 − p ) x + ∈ ( 1 − p ) R ( 1 − e ) ( − ( 1 − p ))
Set β = − y = − ( 1 − p ) \beta=-y=-(1-p) β = − y = − ( 1 − p ) and β ′ = − ( 1 − e ) ∈ ( 1 − e ) R ( 1 − p ) \beta' =-(1-e)\in(1-e)R(1-p) β ′ = − ( 1 − e ) ∈ ( 1 − e ) R ( 1 − p ) . Then using Claim 1, we get β ∈ ( 1 − p ) R ( 1 − e ) \beta\in(1-p)R(1-e) β ∈ ( 1 − p ) R ( 1 − e ) and β ′ ∈ ( 1 − e ) R ( 1 − p ) \beta'\in(1-e)R(1-p) β ′ ∈ ( 1 − e ) R ( 1 − p ) such that
β β ′ = ( 1 − p ) ( 1 − e ) = ( 1 − p ) and β ′ β = ( 1 − e ) ( 1 − p ) = ( 1 − e ) \beta\beta'=(1-p)(1-e)=(1-p)\quad\text{and}\quad\beta'\beta=(1-e)(1-p)=(1-e) β β ′ = ( 1 − p ) ( 1 − e ) = ( 1 − p ) and β ′ β = ( 1 − e ) ( 1 − p ) = ( 1 − e )
Also, with e a = a ea=a e a = a , and g 1 p = g 1 g_1p=g_1 g 1 p = g 1 , for elements a g 1 ∈ e R p ag_1\in eRp a g 1 ∈ e R p , and p x ∈ p R e px\in pRe p x ∈ pR e we obtain
( p x ) ( a g 1 ) = p ( x a ) g 1 = p ( π u e a ) g 1 = p ( π u a ) g 1 = p ( π ) g 1 = p g 1 = p (px)(ag_1)=p(xa)g_1=p(\pi uea)g_1=p(\pi ua)g_1=p(\pi)g_1=pg_1=p ( p x ) ( a g 1 ) = p ( x a ) g 1 = p ( π u e a ) g 1 = p ( π u a ) g 1 = p ( π ) g 1 = p g 1 = p
and,
( a g 1 ) ( p x ) = a ( g 1 p ) x = a g 1 x = a g 1 ( π u e ) = a ( g 1 π ) u e = a π u e = a x = e (ag_1)(px)=a(g_1p)x=ag_1x=ag_1(\pi ue)=a(g_1\pi)ue=a\pi ue=ax=e ( a g 1 ) ( p x ) = a ( g 1 p ) x = a g 1 x = a g 1 ( π u e ) = a ( g 1 π ) u e = aπ u e = a x = e
Thus, p x px p x and β \beta β are invertible blocks in p R e pRe pR e and ( 1 − p ) R ( 1 − e ) (1-p)R(1-e) ( 1 − p ) R ( 1 − e ) with inverses a g 1 ag_1 a g 1 and β ′ \beta' β ′ . Therefore, by our Block Lemma 2.14, d d d is invertible in R R R .
In fact, Lemma 2.14 gives us its inverse to be the sum of the block inverse of p x px p x in e R p eRp e R p , block inverse of β \beta β , and ( β ) ( ( 1 − p ) x ) ( a g 1 ) − 1 (\beta)((1-p)x)(ag_1)^{-1} ( β ) (( 1 − p ) x ) ( a g 1 ) − 1
d − 1 = a g 1 + β ′ − β ′ ( 1 − p ) x ( a g 1 ) d^{-1}=ag_1+\beta'-\beta'(1-p)x(ag_1) d − 1 = a g 1 + β ′ − β ′ ( 1 − p ) x ( a g 1 )
Thus, x − y x-y x − y being invertible proves that a a a is clean. For the sake of completition we show it below.
a d − d ( 1 − e ) = a ( x − y ) − ( x − y ) ( 1 − e ) = a x − a y − x ( 1 − e ) + y ( 1 − e ) = a x − a y + y ad-d(1-e)=a(x-y)-(x-y)(1-e)=ax-ay-x(1-e)+y(1-e)=ax-ay+y a d − d ( 1 − e ) = a ( x − y ) − ( x − y ) ( 1 − e ) = a x − a y − x ( 1 − e ) + y ( 1 − e ) = a x − a y + y
⟹ a d − d ( 1 − e ) = e + ( 1 − e ) = 1 ⟹ a = d − 1 + d ( 1 − e ) d − 1 \Longrightarrow\ ad-d(1-e)=e+(1-e)=1\Longrightarrow a=d^{-1}+d(1-e)d^{-1} ⟹ a d − d ( 1 − e ) = e + ( 1 − e ) = 1 ⟹ a = d − 1 + d ( 1 − e ) d − 1
We observe that a a a is actually special-clean.
Set f = d ( 1 − e ) d − 1 f=d(1-e)d^{-1} f = d ( 1 − e ) d − 1 . Then a − d − 1 = f a-d^{-1}=f a − d − 1 = f .
Suppose ∃ r ∈ a R ∩ f R \exists r\in aR\cap fR ∃ r ∈ a R ∩ f R . Then we can write r = a r 1 = f s 1 r=ar_1=fs_1 r = a r 1 = f s 1 for r 1 , s 1 ∈ R r_1,s_1\in R r 1 , s 1 ∈ R . Now
a d a = a x a − a ( y a ) = a − 0 = a ada=axa-a(ya)=a-0=a a d a = a x a − a ( y a ) = a − 0 = a
Also f = a − d − 1 ⟹ a d f = a d ( a − d − 1 ) = a d a − a d d − 1 = a − a = 0 f=a-d^{-1}\Longrightarrow adf=ad(a-d^{-1})=ada-add^{-1}=a-a=0 f = a − d − 1 ⟹ a df = a d ( a − d − 1 ) = a d a − a d d − 1 = a − a = 0
Then a d r = a d a r 1 = a r 1 = r adr=adar_1=ar_1=r a d r = a d a r 1 = a r 1 = r while a d r = a d f s 1 = 0 adr=adfs_1=0 a d r = a df s 1 = 0 . Hence r = 0 r=0 r = 0 .
Thus a R ∩ f R = { 0 } aR\cap fR=\{0\} a R ∩ f R = { 0 } and a a a is therefore, special-clean.
( i ) ⟹ ( i i ) (i)\Longrightarrow(ii) ( i ) ⟹ ( ii ) : Suppose there exists v ∈ U ( R ) v\in U(R) v ∈ U ( R ) and f 2 = f f^2=f f 2 = f such that a = f + v a=f+v a = f + v where a R ∩ f R = { 0 } aR\cap fR=\{0\} a R ∩ f R = { 0 } .
Take u = v − 1 u=v^{-1} u = v − 1 and p = a u p=au p = a u .
Then a u = f u + 1 ⟹ ( a u − 1 ) = f u ∈ f R ⟹ ( a u − 1 ) a ∈ f R au=fu+1\Longrightarrow(au-1)=fu\in fR\Longrightarrow(au-1)a\in fR a u = f u + 1 ⟹ ( a u − 1 ) = f u ∈ f R ⟹ ( a u − 1 ) a ∈ f R
Together with ( a u − 1 ) a = a ( u a − 1 ) ∈ a R (au-1)a=a(ua-1)\in aR ( a u − 1 ) a = a ( u a − 1 ) ∈ a R , we must have ( a u − 1 ) a = 0 (au-1)a=0 ( a u − 1 ) a = 0 since a R ∩ f R = { 0 } aR\cap fR=\{0\} a R ∩ f R = { 0 } . Thus a = a u a a=aua a = a u a .
Then clearly p 2 = p p^2=p p 2 = p and p ( f ) = a u ( a − v ) = a u a − a = 0 p(f)=au(a-v)=aua-a=0 p ( f ) = a u ( a − v ) = a u a − a = 0 . Consequently, ( 1 − p ) f = f (1-p)f=f ( 1 − p ) f = f and p ( 1 − f ) = p p(1-f)=p p ( 1 − f ) = p .
Also from above ( a u − 1 ) = f u ⟹ ( 1 − f ) ( p − 1 ) = 0 ⟹ ( 1 − f ) p = 1 − f (au-1)=fu\Longrightarrow(1-f)(p-1)=0\Longrightarrow(1-f)p=1-f ( a u − 1 ) = f u ⟹ ( 1 − f ) ( p − 1 ) = 0 ⟹ ( 1 − f ) p = 1 − f , and f ( 1 − p ) = ( 1 − p ) f(1-p)=(1-p) f ( 1 − p ) = ( 1 − p ) .
We are now ready to construct our σ ∈ ( 1 − p ) R p \sigma\in(1-p)Rp σ ∈ ( 1 − p ) R p .
Set σ = ( 1 − f ) − p \sigma=(1-f)-p σ = ( 1 − f ) − p . Then using ( 1 − f ) p = 1 − f (1-f)p=1-f ( 1 − f ) p = 1 − f gives
σ = − f p = − ( 1 − p ) f p ∈ ( 1 − p ) R p \sigma=-fp=-(1-p)fp\in(1-p)Rp σ = − f p = − ( 1 − p ) f p ∈ ( 1 − p ) R p
Label V = u ( 1 − a u ) + ( a u ) + σ = u ( 1 − p ) + p + σ = u ( 1 − p ) + p + ( 1 − f − p ) = u ( 1 − p ) + ( 1 − f ) V=u(1-au)+(au)+\sigma=u(1-p)+p+\sigma=u(1-p)+p+(1-f-p)=u(1-p)+(1-f) V = u ( 1 − a u ) + ( a u ) + σ = u ( 1 − p ) + p + σ = u ( 1 − p ) + p + ( 1 − f − p ) = u ( 1 − p ) + ( 1 − f ) .
We prove that V ∈ U ( R ) V\in U(R) V ∈ U ( R ) using Block Lemma. Using f u = − ( 1 − p ) fu=-(1-p) f u = − ( 1 − p ) , f ( 1 − p ) = 1 − p f(1-p)=1-p f ( 1 − p ) = 1 − p and ( 1 − f ) p = 1 − f (1-f)p=1-f ( 1 − f ) p = 1 − f , we rewrite V V V as
V = u ( 1 − p ) + ( 1 − f ) = ( 1 − f ) u ( 1 − p ) + f u ( 1 − p ) + ( 1 − f ) p = ( 1 − f ) u ( 1 − p ) − ( 1 − p ) + ( 1 − f ) p = ( 1 − f ) u ( 1 − p ) − f ( 1 − p ) + ( 1 − f ) p \begin{aligned}
V &= u(1-p)+(1-f)=(1-f)u(1-p)+fu(1-p)+(1-f)p\\
&=(1-f)u(1-p)-(1-p)+(1-f)p\\
&=(1-f)u(1-p)-f(1-p)+(1-f)p
\end{aligned} V = u ( 1 − p ) + ( 1 − f ) = ( 1 − f ) u ( 1 − p ) + f u ( 1 − p ) + ( 1 − f ) p = ( 1 − f ) u ( 1 − p ) − ( 1 − p ) + ( 1 − f ) p = ( 1 − f ) u ( 1 − p ) − f ( 1 − p ) + ( 1 − f ) p
Thus we can write V = ( 1 − f ) p + ( 1 − f ) u ( 1 − p ) − f ( 1 − p ) V=(1-f)p+(1-f)u(1-p)-f(1-p) V = ( 1 − f ) p + ( 1 − f ) u ( 1 − p ) − f ( 1 − p ) . To use the Block lemma, we show the extreme components are inverses in ( 1 − f ) R p (1-f)Rp ( 1 − f ) R p and f R ( 1 − p ) fR(1-p) f R ( 1 − p ) .
Using p ( 1 − f ) = p p(1-f)=p p ( 1 − f ) = p and ( 1 − f ) p = 1 − f (1-f)p=1-f ( 1 − f ) p = 1 − f we get
( 1 − f ) p ⋅ p ( 1 − f ) = ( 1 − f ) and p ( 1 − f ) ( 1 − f ) p = p (1-f)p\cdot p(1-f)=(1-f)\quad\text{and}\quad p(1-f)(1-f)p=p ( 1 − f ) p ⋅ p ( 1 − f ) = ( 1 − f ) and p ( 1 − f ) ( 1 − f ) p = p
and using f ( 1 − p ) = ( 1 − p ) f(1-p)=(1-p) f ( 1 − p ) = ( 1 − p ) and ( 1 − p ) f = f (1-p)f=f ( 1 − p ) f = f we get
f ( 1 − p ) ( 1 − p ) f = f and ( 1 − p ) f ⋅ f ( 1 − p ) = 1 − p f(1-p)(1-p)f=f\quad\text{and}\quad(1-p)f\cdot f(1-p)=1-p f ( 1 − p ) ( 1 − p ) f = f and ( 1 − p ) f ⋅ f ( 1 − p ) = 1 − p
Thus ( 1 − f ) p (1-f)p ( 1 − f ) p is invertible in ( 1 − f ) R p (1-f)Rp ( 1 − f ) R p and f ( 1 − p ) f(1-p) f ( 1 − p ) is invertible in f R ( 1 − p ) fR(1-p) f R ( 1 − p ) . Then by Block Lemma 2.14 (ii), V V V is invertible in R R R .
Hence by Lemma 4.3, there exists a σ ∈ ( 1 − a u ) R ( a u ) \sigma\in(1-au)R(au) σ ∈ ( 1 − a u ) R ( a u ) such that
( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U ( ( 1 − a u ) R ( 1 − a u ) ) (1-au)u(1-au)-\sigma u(1-au)\in U((1-au)R(1-au)) ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U (( 1 − a u ) R ( 1 − a u ))
( i i i ) ⟹ ( i i ) (iii)\Longrightarrow(ii) ( iii ) ⟹ ( ii ) : Label S = ( 1 − a u ) R ( 1 − a u ) S=(1-au)R(1-au) S = ( 1 − a u ) R ( 1 − a u ) . Let u ∈ U ( R ) u\in U(R) u ∈ U ( R ) such that a = a u a a=aua a = a u a and ( α , ψ ) (\alpha,\psi) ( α , ψ ) is unimodular in S S S where α = ( 1 − a u ) u ( 1 − a u ) \alpha=(1-au)u(1-au) α = ( 1 − a u ) u ( 1 − a u ) and ψ = ( 1 − a u ) ( u − 1 ( a u ) u ) ( 1 − a u ) \psi=(1-au)(u^{-1}(au)u)(1-au) ψ = ( 1 − a u ) ( u − 1 ( a u ) u ) ( 1 − a u ) . Since ( α , ψ ) (\alpha,\psi) ( α , ψ ) is left reducible in S S S , there exists z ∈ S z\in S z ∈ S such that
α + z ψ = α + z ( μ γ ) ∈ U ( S ) \alpha+z\psi=\alpha+z(\mu\gamma)\in U(S) α + z ψ = α + z ( μ γ ) ∈ U ( S )
Label σ = − z ( ( 1 − a u ) u − 1 ( a u ) ) ∈ ( 1 − a u ) R ( a u ) \sigma=-z((1-au)u^{-1}(au))\in(1-au)R(au) σ = − z (( 1 − a u ) u − 1 ( a u )) ∈ ( 1 − a u ) R ( a u ) . Then σ u ( 1 − a u ) = − z ψ \sigma u(1-au)=-z\psi σ u ( 1 − a u ) = − z ψ and
( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) = α + z ψ ∈ U ( S ) . (1-au)u(1-au)-\sigma u(1-au)=\alpha+z\psi\in U(S). ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) = α + z ψ ∈ U ( S ) .
( i ) ⟹ ( i i i ) (i)\Longrightarrow(iii) ( i ) ⟹ ( iii ) : If a a a is special clean, then there exists f 2 = f f^2=f f 2 = f and v ∈ U ( R ) v\in U(R) v ∈ U ( R ) such that
a = f + v and a R ∩ f R = { 0 } a=f+v\quad\text{and}\quad aR\cap fR=\{0\} a = f + v and a R ∩ f R = { 0 }
Take u = v − 1 u=v^{-1} u = v − 1 and p = a u p=au p = a u . Consider the pair ( α , ψ ) (\alpha,\psi) ( α , ψ ) , where α = ( 1 − a u ) u ( 1 − a u ) \alpha=(1-au)u(1-au) α = ( 1 − a u ) u ( 1 − a u ) and ψ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) \psi=(1-au)(u^{-1}au)u(1-au) ψ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) . It can easily be seen that this pair is left unimodular in S = ( 1 − a u ) R ( 1 − a u ) S=(1-au)R(1-au) S = ( 1 − a u ) R ( 1 − a u ) . Then from the proof in ( i ) ⟺ ( i i ) (i)\Longleftrightarrow(ii) ( i ) ⟺ ( ii ) , for σ = 1 − f − p = − f p ∈ ( 1 − p ) R p \sigma=1-f-p=-fp\in(1-p)Rp σ = 1 − f − p = − f p ∈ ( 1 − p ) R p we have
α − σ u ( 1 − a u ) ∈ U ( ( 1 − p ) R ( 1 − p ) ) ⋯ ( ⋆ ⋆ ) \alpha-\sigma u(1-au)\in U((1-p)R(1-p))\quad\cdots\quad(\star\star) α − σ u ( 1 − a u ) ∈ U (( 1 − p ) R ( 1 − p )) ⋯ ( ⋆ ⋆ )
Now ( 1 − p ) u − 1 p = ( 1 − p ) v p = ( 1 − p ) ( a − f ) p = ( 1 − p ) a p − ( 1 − p ) f p (1-p)u^{-1}p=(1-p)vp=(1-p)(a-f)p=(1-p)ap-(1-p)fp ( 1 − p ) u − 1 p = ( 1 − p ) v p = ( 1 − p ) ( a − f ) p = ( 1 − p ) a p − ( 1 − p ) f p , where
( 1 − p ) f p = ( 1 − a u ) ( a − u − 1 ) ( p ) = ( 1 − a u ) ( − u − 1 ) ( p ) = ( a − u − 1 ) ( p ) = f p (1-p)fp=(1-au)(a-u^{-1})(p)=(1-au)(-u^{-1})(p)=(a-u^{-1})(p)=fp ( 1 − p ) f p = ( 1 − a u ) ( a − u − 1 ) ( p ) = ( 1 − a u ) ( − u − 1 ) ( p ) = ( a − u − 1 ) ( p ) = f p
Therefore, ( 1 − p ) u − 1 p = − f p = 1 − f − p = σ (1-p)u^{-1}p=-fp=1-f-p=\sigma ( 1 − p ) u − 1 p = − f p = 1 − f − p = σ . Then ψ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) = σ u ( 1 − a u ) \psi=(1-au)(u^{-1}au)u(1-au)=\sigma u(1-au) ψ = ( 1 − a u ) ( u − 1 a u ) u ( 1 − a u ) = σ u ( 1 − a u ) . Then from ( ⋆ ⋆ ) (\star\star) ( ⋆ ⋆ ) above,
α + ( − 1 S ) ψ ∈ U ( S ) \alpha+(-1_S)\psi\in U(S) α + ( − 1 S ) ψ ∈ U ( S )
Hence the pair ( α , ψ ) (\alpha,\psi) ( α , ψ ) is left reduced in S S S .
Remark 4.5. Applying this theorem on the entire ring enhances a known characterization of unit-regular rings. It is quite well known that unit-regularity is encoded by a canonical internal decomposition R = a R ⊕ ( a − u ) R R=aR\oplus(a-u)R R = a R ⊕ ( a − u ) R and can be found in several papers such as [7 ] , [25 ] , [6 ] . We can further enhance and construct what the element a − u a-u a − u looks like:
A ring R R R is unit regular if and only if for every element a a a there exists unit u ∈ U ( R ) u\in U(R) u ∈ U ( R ) and f ∈ Idem ( R ) f\in\operatorname{Idem}(R) f ∈ Idem ( R ) such that a = a u a a=aua a = a u a and R = a R ⊕ f R R=aR\oplus fR R = a R ⊕ f R where ( 1 − a u ) − f ∈ ( 1 − a u ) R ( a u ) (1-au)-f\in(1-au)R(au) ( 1 − a u ) − f ∈ ( 1 − a u ) R ( a u ) .
As a consequence of the above theorem, we recover and strengthen Mary and Patricio’s [29 ] result to show that it is an equivalent characterization than a consequence. The proof shows that the y v 2 + v 4 yv_{2}+v_{4} y v 2 + v 4 defined below is the inverse of the element ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) (1-au)u(1-au)-\sigma u(1-au) ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) in (ii) of the above theorem.
Lemma 4.6. [29 Lemma 3.1] Let a = a u a a=aua a = a u a and f = a u f=au f = a u . Consider the Peirce decomposition of u − 1 u^{-1} u − 1 with respect to ( a u ) (au) ( a u ) as
u − 1 = ( a u ) u − 1 ( a u ) + ( a u ) u − 1 ( 1 − a u ) + ( 1 − a u ) u − 1 ( a u ) + ( 1 − a u ) u − 1 ( 1 − a u ) u^{-1}=(au)u^{-1}(au)+(au)u^{-1}(1-au)+(1-au)u^{-1}(au)+(1-au)u^{-1}(1-au) u − 1 = ( a u ) u − 1 ( a u ) + ( a u ) u − 1 ( 1 − a u ) + ( 1 − a u ) u − 1 ( a u ) + ( 1 − a u ) u − 1 ( 1 − a u )
If y v 2 + v 4 yv_{2}+v_{4} y v 2 + v 4 (resp. v 3 x + v 4 v_{3}x+v_{4} v 3 x + v 4 ) is a unit in f ˉ R f ˉ \bar{f}R\bar{f} f ˉ R f ˉ for some y ∈ f ˉ R f y\in\bar{f}Rf y ∈ f ˉ R f (resp. x ∈ f R f ˉ x\in fR\bar{f} x ∈ f R f ˉ ), then a a a is special clean.
Proof. We prove that it is a characterization.
Label f ˉ = ( 1 − a u ) \bar{f}=(1-au) f ˉ = ( 1 − a u ) . Then that any y ∈ f ˉ R f y\in\bar{f}Rf y ∈ f ˉ R f
y v 2 + v 4 = y ( u − 1 ) ( 1 − a u ) + ( 1 − a u ) u − 1 ( 1 − a u ) = ( y + f ˉ ) u − 1 f ˉ yv_{2}+v_{4}=y(u^{-1})(1-au)+(1-au)u^{-1}(1-au)=(y+\bar{f})u^{-1}\bar{f} y v 2 + v 4 = y ( u − 1 ) ( 1 − a u ) + ( 1 − a u ) u − 1 ( 1 − a u ) = ( y + f ˉ ) u − 1 f ˉ
Suppose V = ( f ˉ + y ) u − 1 f ˉ V=(\bar{f}+y)u^{-1}\bar{f} V = ( f ˉ + y ) u − 1 f ˉ is a unit in f ˉ R f ˉ \bar{f}R\bar{f} f ˉ R f ˉ for some y ∈ f ˉ R f y\in\bar{f}Rf y ∈ f ˉ R f . Then
V − 1 V = V − 1 ( f ˉ + y ) u − 1 f ˉ = f ˉ V^{-1}V=V^{-1}(\bar{f}+y)u^{-1}\bar{f}=\bar{f} V − 1 V = V − 1 ( f ˉ + y ) u − 1 f ˉ = f ˉ
Label σ 0 = V − 1 ( f ˉ + y ) u − 1 ∈ f ˉ R \sigma_{0}=V^{-1}(\bar{f}+y)u^{-1}\in\bar{f}R σ 0 = V − 1 ( f ˉ + y ) u − 1 ∈ f ˉ R . Then σ 0 f ˉ = V − 1 V = f ˉ ⟹ σ 0 ( f ) = σ 0 − f ˉ \sigma_{0}\bar{f}=V^{-1}V=\bar{f}\Longrightarrow\sigma_{0}(f)=\sigma_{0}-\bar{f} σ 0 f ˉ = V − 1 V = f ˉ ⟹ σ 0 ( f ) = σ 0 − f ˉ .
Now label σ = − σ 0 ( f ) = − σ 0 + f ˉ ∈ f ˉ R f \sigma=-\sigma_{0}(f)=-\sigma_{0}+\bar{f}\in\bar{f}Rf σ = − σ 0 ( f ) = − σ 0 + f ˉ ∈ f ˉ R f . Then we obtain
( 1 − a u ) u ( 1 − a u ) − σ ( u ) ( 1 − a u ) = f ˉ u f ˉ − σ u f ˉ = f ˉ u f ˉ + ( σ 0 − f ˉ ) u f ˉ = σ 0 u f ˉ (1-au)u(1-au)-\sigma(u)(1-au)=\bar{f}u\bar{f}-\sigma u\bar{f}=\bar{f}u\bar{f}+(\sigma_{0}-\bar{f})u\bar{f}=\sigma_{0}u\bar{f} ( 1 − a u ) u ( 1 − a u ) − σ ( u ) ( 1 − a u ) = f ˉ u f ˉ − σ u f ˉ = f ˉ u f ˉ + ( σ 0 − f ˉ ) u f ˉ = σ 0 u f ˉ
Now using σ 0 = V − 1 ( f ˉ + y ) u − 1 \sigma_{0}=V^{-1}(\bar{f}+y)u^{-1} σ 0 = V − 1 ( f ˉ + y ) u − 1 and y f ˉ = 0 y\bar{f}=0 y f ˉ = 0 we obtain
σ 0 u f ˉ = V − 1 ( f ˉ + y ) u − 1 u f ˉ = V − 1 f ˉ = V − 1 \sigma_{0}u\bar{f}=V^{-1}(\bar{f}+y)u^{-1}u\bar{f}=V^{-1}\bar{f}=V^{-1} σ 0 u f ˉ = V − 1 ( f ˉ + y ) u − 1 u f ˉ = V − 1 f ˉ = V − 1
Thus ( 1 − a u ) u ( 1 − a u ) − σ ( u ) ( 1 − a u ) (1-au)u(1-au)-\sigma(u)(1-au) ( 1 − a u ) u ( 1 − a u ) − σ ( u ) ( 1 − a u ) is the inverse of y v 2 + v 4 yv_{2}+v_{4} y v 2 + v 4 for some y ∈ f ˉ R f ˉ y\in\bar{f}R\bar{f} y ∈ f ˉ R f ˉ . Then by condition (ii) of the above Theorem 4.4 , a = a u a a=aua a = a u a is special clean.
Conversely, suppose a = e + v a=e+v a = e + v be special clean and put u = v − 1 u=v^{-1} u = v − 1 , so that a u a = a aua=a a u a = a and 1 − a u = − e v − 1 1-au=-ev^{-1} 1 − a u = − e v − 1 . Then ( 1 − a u ) ∈ Idem ( R ) (1-au)\in\operatorname{Idem}(R) ( 1 − a u ) ∈ Idem ( R ) gives ( 1 − a u ) v = − ( 1 − a u ) e (1-au)v=-(1-au)e ( 1 − a u ) v = − ( 1 − a u ) e , while e ( 1 − a u ) = 1 − a u e(1-au)=1-au e ( 1 − a u ) = 1 − a u , so
v 4 = ( 1 − a u ) u − 1 ( 1 − a u ) = − ( 1 − a u ) e ( 1 − a u ) = − ( 1 − a u ) , v_{4}=(1-au)u^{-1}(1-au)=-(1-au)e(1-au)=-(1-au), v 4 = ( 1 − a u ) u − 1 ( 1 − a u ) = − ( 1 − a u ) e ( 1 − a u ) = − ( 1 − a u ) ,
which is a unit of ( 1 − a u ) R ( 1 − a u ) (1-au)R(1-au) ( 1 − a u ) R ( 1 − a u ) . Hence the hypothesis holds with y = 0 y=0 y = 0 .
Remark 4.7. It is not too difficult to see that in the opposite ring R o p R^{\mathrm{op}} R op , for the element u − 1 a u u^{-1}au u − 1 a u where ( u − 1 a u ) − 1 ∘ u ∘ ( u − 1 a u ) = u − 1 a u (u^{-1}au)^{-1}\circ u\circ(u^{-1}au)=u^{-1}au ( u − 1 a u ) − 1 ∘ u ∘ ( u − 1 a u ) = u − 1 a u , the Peirce decomposition of u − 1 u^{-1} u − 1 with respect to the idempotent ( u − 1 a u ) ∘ u = a u (u^{-1}au)\circ u=au ( u − 1 a u ) ∘ u = a u transforms as
v 4 o p = v 4 , v 2 o p = v 3 , x = y o p ∈ ( 1 − a u ) R o p ( a u ) = ( a u ) R ( 1 − a u ) v_{4}^{\mathrm{op}}=v_{4},\quad v_{2}^{\mathrm{op}}=v_{3},\quad x=y^{\mathrm{op}}\in(1-au)R^{\mathrm{op}}(au)=(au)R(1-au) v 4 op = v 4 , v 2 op = v 3 , x = y op ∈ ( 1 − a u ) R op ( a u ) = ( a u ) R ( 1 − a u )
so the term y v 2 o p v 2 o p + v 4 o p y v_{2}^{op}v_{2}^{op}+v_{4}^{op} y v 2 o p v 2 o p + v 4 o p ∈ U ( S ) U(S) U ( S ) becomes v 3 x + v 4 v_{3}x+v_{4} v 3 x + v 4 ∈ U ( S ) U(S) U ( S ) . Thus u − 1 a u u^{-1}au u − 1 a u is special clean ⟺ \Longleftrightarrow ⟺ a a a is special clean. (using symmetricity of left ideal and right ideal intersection in a special-clean decomposition as proved in [23 ] Theorem 2.3).
Corollary 4.8. Let a ∈ R a \in R a ∈ R with a = a u a a=aua a = a u a for some u ∈ U ( R ) u \in U(R) u ∈ U ( R ) . Label the idempotent p = a u p=au p = a u and the corner ring S = ( 1 − p ) R ( 1 − p ) S=(1-p)R(1-p) S = ( 1 − p ) R ( 1 − p ) . If
sr S ( ( 1 − p ) u ( 1 − p ) ) = 1 \operatorname{sr}_{S}\bigl((1-p)u(1-p)\bigr)=1 sr S ( ( 1 − p ) u ( 1 − p ) ) = 1 then a a a is special clean.
Proof. Label λ = ( 1 − p ) u − 1 ( 1 − p ) \lambda=(1-p)u^{-1}(1-p) λ = ( 1 − p ) u − 1 ( 1 − p ) , μ = ( 1 − p ) u − 1 p \mu=(1-p)u^{-1}p μ = ( 1 − p ) u − 1 p , α = ( 1 − p ) u ( 1 − p ) \alpha=(1-p)u(1-p) α = ( 1 − p ) u ( 1 − p ) and γ = p u ( 1 − p ) \gamma=pu(1-p) γ = p u ( 1 − p ) . Label ψ = μ γ \psi=\mu\gamma ψ = μ γ . Then clearly λ , α , ψ ∈ ( 1 − p ) R ( 1 − p ) \lambda,\alpha,\psi\in(1-p)R(1-p) λ , α , ψ ∈ ( 1 − p ) R ( 1 − p ) and
λ α + ψ = ( 1 − p ) u − 1 ( 1 − p ) u ( 1 − p ) + ( 1 − p ) u − 1 p u ( 1 − p ) \lambda\alpha+\psi=(1-p)u^{-1}(1-p)u(1-p)+(1-p)u^{-1}pu(1-p) λ α + ψ = ( 1 − p ) u − 1 ( 1 − p ) u ( 1 − p ) + ( 1 − p ) u − 1 p u ( 1 − p ) = ( 1 − p ) u − 1 ( ( 1 − p ) + p ) u ( 1 − p ) = ( 1 − p ) = 1 S =(1-p)u^{-1}\bigl((1-p)+p\bigr)u(1-p)=(1-p)=1_{S} = ( 1 − p ) u − 1 ( ( 1 − p ) + p ) u ( 1 − p ) = ( 1 − p ) = 1 S Since sr S ( α ) = 1 \operatorname{sr}_{S}(\alpha)=1 sr S ( α ) = 1 there exists z ∈ S z \in S z ∈ S such that
α + z ψ = α + z μ γ ∈ U ( S ) . \alpha+z\psi=\alpha+z\mu\gamma\in U(S). α + z ψ = α + z μ γ ∈ U ( S ) . Label σ = − z μ \sigma=-z\mu σ = − z μ . Then σ ∈ ( 1 − p ) R p \sigma\in(1-p)Rp σ ∈ ( 1 − p ) R p . Then
α − σ γ = ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U ( ( 1 − a u ) R ( 1 − a u ) ) \alpha-\sigma\gamma=(1-au)u(1-au)-\sigma u(1-au)\in U\bigl((1-au)R(1-au)\bigr) α − σ γ = ( 1 − a u ) u ( 1 − a u ) − σ u ( 1 − a u ) ∈ U ( ( 1 − a u ) R ( 1 − a u ) ) Then by above Theorem 4.4 (ii), a a a is special-clean.
We can recover the following result of Camillo and Khurana on the entire ring R R R .
Theorem 4.9. [8 ] A ring R R R is unit-regular if and only if every element a a a of R R R can be written as e + u e+u e + u such that a R ∩ e R = { 0 } aR\cap eR=\{0\} a R ∩ e R = { 0 } , where e e e is an idempotent and u u u a unit in R R R .
Proof. Since R R R is unit regular, R R R has stable range 1 and consequently, for any idempotent e ∈ R e \in R e ∈ R , e R e eRe e R e also has stable range 1. For any element a ∈ R a \in R a ∈ R , there exists u ∈ U ( R ) u \in U(R) u ∈ U ( R ) such that a = a u a a=aua a = a u a . Set e = ( 1 − a u ) e=(1-au) e = ( 1 − a u ) . Then sr e R e ( e u e ) = 1 \operatorname{sr}_{eRe}(eue)=1 sr e R e ( e u e ) = 1 and subsequently, by Corollary 4.8, a a a is special clean. The converse is straightforward.
Department of Mathematics, University of Wisconsin Eau Claire, WI-54701, USA
Email address: siddiquf@uwec.edu