Chapter 8
Regular Systems

In the previous chapter, we have described the construction of a regular groupoid starting with a symmetric presheaf of groups G over a biordered set E with respect to a regular equivalence relation δ on E and a regular family of matrices P for G. Here we describe the construction of regular semigroups from regular groupoids.

Throughout this chapter, S will denote a regular groupoid, E, the set of idempotents of S and δ, the equivalence relation on E induced by the distinct Rees groupoids of S. Since inverses of elements of S can be defined as in the case of regular semigroups, notations used for regular semigroups in (1.14) and (1.15) can be carried over to regular groupoids also. In particular, i will denote the inverse relation (cf. (1.14)) of S and i(x) the set of inverse of x S.

A partial transformation t of S is said to be idempotent preserving if and only if for all X S, (E(X))t E.

Definition 8.1. Let S be a regular groupiod. For eωre[eωle] let σr(e,e)[σl(e,e)] denote a mapping of Re[Le] into Re[Le] and let

σr = {σr(e,e)|eωre},σl = {σl(e,e)|eωle}. (8.1)

Then = (S,σr,σl) is a regular system if the following axioms (and their duals) are satisfied. In these statements products shown are products in S.

  1. For all e E,σr(e,e) is the identity map of Re and for all eωe,eσr(e,e) = e.
  2. Let e,e ωr(e) and e1re. Then
    σr(e,e) = σr(e,e)

    if either e = e or for some e1re,e = eτr(e1).

  3. Let x Re,x i(x;Le) and e,e ω(e). If ere, then xσl(e,e),xσl(e,e) are inverses of xσr(e,e).
  4. Let x Re and x i(x;Le). For e ω(e),xσl(e,e) is an inverse of xσr(e,e) and the mapping a(x,x) : ω(e) ω(xx) defined by
    ea(x,x) = (xσl(e,e)) (xσr(e,e))

    is an ω-isomorphism.

  5. Let x Re,x i(x;Le) and e ω(e). If xx = f and ea(x,x) = f, then
    xσr(e,e) = xσl(f,f).
  6. Let x He1,e,y Hf,f1,h S(e,f),x = i(x;e,e1) and g ωl(e) ωr(f). If h1 = (hτ(e))a(x,x) and g1 = (gτl(e))a(x,x), then
    [(xσl(e,h))(yσr(f,h))]σr(h 1,g1) = (xσl(e,g))(yσr(f,g)).

Remarks:

(8.1) a : (x,x) a(x,x) is a mapping of iS into T(E,δ) such that for all e E,a(e,e) is the identity map of ω(e).

Proof. Since by (R3) , xσl(e,e) is an inverse of xσr(e,e) (where x Re,x i(x;Le) and eωe), (xσl(e,e))(xσr(e,e)) is an idempotent in the Rees groupoid De of S and hence eδea(x,x). By (R4), a is a mapping of iS into E1 and the statement above implies that a(x,x) T(E,δ) for all (x,x) iS.

If e E, and e ω(e), by (R1) and (R1)eσr(e,e) = e = eσl(e,e). Thus ea(e,e) = e.

(8.2) For all eωre and e1re,e1σr(e,e) = eτr(e1).

Proof. By axioms (R2) and (R1),

  1. For every e E,a(e,e) and σr(e,e) are identity maps of ω(e) and Re respectively.
  2. If eωr(e) and x Re, then
    xσr(e,e) H e,ear(x,x)

    for every x i(x).

  3. Let eωr(e) and either e = e or for some e2re,e = eτr(e2). If ere1, then
    σr(e,e) = σr(e 1,e).
  4. Let e,e ωr(e) and ele. Then for every x Re,
    xσr(e,e) = e(xσr(e,e))

    where e(σr(e,e)) denotes the product of e and xσr(e,e) in S.

  5. Let eωre and x Re. If for some x i(x) such that xx = e f = ear(x,x) and if f = xx then
    xσr(e,e) = xσl(f,f) andfal(x,x) = eτr(e).
  6. Let x He1,e,y Hf,f1,h S(e,f),x = i(x,e,e1)
    e1σr(e,e) = e 1σr(e 1,eτr(e 1)) = eτr(e 1).

(8.3) Let x Re and eωre. If ar(x,x) = τr(x,x) a(x,x). Then

xσr(e,e) H e,ear(x,x),

for all x i(x). In particular ear(x,x)lear(x,x) for x,x i(x).

Proof. Suppose that x i(x). Let xx = e1. Then ere1 and hence, e1 = eτr(e1), and by axiom (R2), σr(e,e) = σr(e1,e1). Now by (R3), xσl(e1,e1) is an inverse of xσr(e1,e1) and hence xσr(e1,e1) is in the -class of (xσl(e1,e1))(xσr(e1,e1)) = e1a(x,x). Since by definitions xσr(e,e) = xσr(e,e1) Re, we have the required result.

It follows from the remarks above that He,ear(x,x) = He,ear(x,x) and hence ear(x,x) is -equivalent to ear(x,x)

(8.4) For all (x,x) ia,(a(x,x))1 = a(x,x).

Proof. Let eωe where e = xx. If f = xx and f = ea(x,x), then

fa(xx) = (xσl(f,f))(xσr(f,f)).

By (R5) and (R5), xσl(f,f) = xσr(e,e) and xσr(f,f) = xσl(ee). Hence

fa(x,x) = (xσr(e,e))(xσl(e,e)) = (xσr(e,e))f(xσl(e,e)) = (xσr(e,e))(xσl(e,e))(xσr(e,e))(xσl(e,e)) = e e = e.

This proves that a(x,x) a(x,x) is the identity map of ω(e). Similarly it can be shown that a(x,x) a(x,x) is the identity map on ω(f). Hence

a(x,x) = (a(x,x))1.

(8.5) Let x Re, and e,e ωr(e). If ele then

xσr(e,e) = e(xσr(e,e)).

Proof. Choose x i(x;Le) and let e1 = eτl(e) and e1 = eτr(e). Thus by (R2)

xσr(e,e) = xσr(e,e 1),xσr(e,e) = xσr(e,e 1).

By axioms (R3)xσr(e,e1) and xσr(e,e1) are -equivalent inverses of xσl(e,e1). Thus the required result follows from the fact that e(xσr(e,e1)) is also an inverse xσl(e,e1) in He,ear(x,x), the product e(xσr(e,e1)) being defined in S.

(8.6) If h1 and g1 are as in (R6), g1ωrh1.

Proof. Since h S(e,f) and g ωl(e) ωr(f), we have gτl(e)ωrhτl(e). Since a(x,x) is an ω-isomorphism we have

g1 = (gτl(e))a(x,x)ωr(hτl(e))a(x,x) = h 1.

Remark 8.2: Let x,y,e,f,x,h and g be as in axiom (R6). Then g1ωrh1

h Hyσr(f,h),xσl(e,h) andg Hyσr(e,g),xσl(f,g).

Then the partial transformation σr(h1,g1) and the products (xσl(e,h)),(yσr(f,h)) and (xσl(e,h))(yσr(f,h)) are defined.

Proof. Since h S(e,f) and g ωl(e) ωr(f),

gτl(e)ωrhτl(e).

Hence

g1 = (g)al(x,x) = (gτl(e))a(x,x)ωr(hτl(e))a(x,x) = (h)al(x,x) = h 1.

By the definition of σl(e,h),xσl(e,h) Lh and hence h Lxσl(e,h). Similarly it can be seen that h Ryσr(f,h). Consequently

h Hyσr(f,h),xσl(e,h),

and g Hyσr(f,g),xσl(e,g).

Remark 8.3: If eωr(e)(eωl(e)), then

eσr(e,e) = eτr(e)(eσl(e,e) = eτl(e)).

Proof. Since σr(e,e) is idempotent preserving eσr(e,e) is an idempotent of Re. Choose (e,e) i.

Then

ar(e,e) = τr(e) a(e,e) = τr(e)

by axiom (R1). Hence for eωre

ea(e,e) = eτr(e).

Again by axiom (R2), since ereτr(e) we have

eσr(e,e) H e,eτr(e) = Heτr(e),eτr(e).

Thus eσr(e,e) is the idempotent of Heτr(e),eτr(e). Consequently

eσr(e,e) = eτr(e).

The bracketed statement can be proved dually.

Theorem 8.4. Let = (S,a,σr,σl) be a regular system. Then the partial operation of S may be extended by means of a,σr and σl so as to make S a regular semigroup. Conversely, if S̄ is the regular groupoid of a regular semigroup S,as,σsr and σsl can be defined in such a way that s = (S̄,as,σsr,σsl) is a regular system. The regular semigroup obtained by extending the operation of S¯ by as,σsr and σsl coincides with S.

Proof. For x,y S define

x y = (xσl(e,h))(yσr(f,h)) (8.2)

where e E(Lx),f E(Ry),h S(x,y) and (xσl(e,h))(yσr(f,h)) denotes the product in S. This product is defined in S since h Hyσr(f,h),xσl(e,h).

If x i(x),y i(y),h1 = hal(x,x) and h2 = har(y,y), then by axiom (R2),

x y Hh1,h2. (8.3)

By axioms (R3), (R3)* and Lemma 2.6, the right hand side of (8.2) does not depend on the choice of e E(Lx) and f E(Ry). Hence to prove that (8.2) defines a binary operation on S, it is necessary to show that for h,hS(e,f),

(xσl(e,h))(yσr(f,h)) = (xσl(e,h))(yσr(f,h)). (8.4)

First suppose that hlh. Then hτl(e) = hτl(e), and hence by (R3)*,

xσl(e,h) = xσl(e,hτl(e)) = xσl(e,hτl(e)) = xσl(e,h).

Further by axiom (R4),

h(yσr(f,h)) = yσr(f,h).

Thus

(xσl(e,h))(yσr(f,h)) = (xσl(e,h)(h(yσr(f,h)))).

Now xσl(e,h) Lh = Lh and so, the products (xσl(e,h))h and h(yσr(f,h)) are defined in S. Since S is associative and since h is a right identity of xσl(e,h), we see that (8.4) holds.

If hrh, dually we can prove that (8.4) holds. If h and h are any two members of S(e,f) by Lemma 2.7, there exists hS(e,f) such that hlhrh. Hence combining the two cases we see that (8.4) holds in general.

For x,y S, if the product xy is defined in S then Hy,x contains an idempotent e (say) and by Lemma 2.4, S(e,e) = {e}. Therefore by axiom (R1) and (R1)*

x y = (xσl(e,e))(yσr(e,e)) = xy.

Thus the operation in S, whenever it is defined, coincides with that given by (8.2).

If eωre, by Lemma 2.5, eS(e,e). Hence by axiom (R1)* and (R2) we obtain

e x = (eσr(e,e))xσr(e,e) (8.5) = e(xσr(e,e)) = xσr(e,e).

Dually if eωle

x e = xσl(e,e). (8.5*)

Thus (8.2), (8.5) and (8.5*) together give

x y = (x h)(h y).

We shall now prove that the operation defined by (8.2) is associative. Let x,y,z S, e E(Lx)y Hf,f and g E(Rz). Choose h1 S(e,f), h2 S(f,g) and let y = i(y;f,f). Then a(y,y) is an ω-isomorphism of ω(f) onto ω(f) and by axiom (R5)

a(y,y)1 = a(y,y).

Hence if h1 = h1ar(y,y) and h2 = h2al(y,y), then by Lemma 3.9, there exist h S(e,h2) and hS(h1,g) such that

har(y,y) = hτl(f).

Now by axioms (R3)* and (R5) we have

yσr(f,h) = yσl(f,hτl(f)) = yσl(f,h).

Therefore by (8.5) and (8.5*)

h y = y h. (8.6)

Since

h2ωf = fal(y,y) = fa(y,y)1

and h1 S(e,f), we have

h ωl(e) ωr(h 2) ωl(e) ωr(f).

Hence by axioms (R3), (R6) and (8.2)

(xy)σr(h 1,har(y,y)) = (xy)σr(h 1,hτl(f)) = (xy)σr(h 1,h) = (xσl(e,h))(yσr(f,h)).

Therefore by (8.5) and (8.5*) we get

(x y) h = (x h)(h y).

Dually we have

h (y z) = (y h)(h z)

and so

(x y) z = ((x y) h)(h z) = ((x h)(h y))(h z) = ((x h)(y h))(h z).

Now the products (x h)(h y),((x h)(h y))(h z) and (y h)(h z) are defined in S. Thus

((x h)(y h))(h z) = (x h)(y h)(h z) = (x h)(h (y z)) = x (y z).

We have proved that S with the operation defined by (8.2) is a semigroup. Let this semigroup be denoted by S(). Evidently two elements x,y S that are ()-equivalent in S are also ()-equivalent in S(). An idempotent in S is also an idempotent in S(). This implies that S() is a regular semigroup.

Conversely let S be a regular semigroup and S̄ be its regular groupoid. For (x,x) i, we define as(x,x) as the ω-isomorphism given by Lemma 4.11. Then it is clear that as is a mapping of i into T(E,δ) where in this case E = E(S),δ = δ(S). Further for eωre(eωle) and x Re(x Le) define

xσsr(e,e) = ex(xσ sl(e,e) = xe). (8.7)

If

σsr = {σ sr(e,e)eωre},σ sl = {σ sl(e,e)eωle}

we then show that

s = (S̄,as,σsr,σ sl) (8.8)

is a regular system.

Axioms (R1) and (R1)* are clearly satisfied. Also if eωre(eωle),x Re(x Le) and if x i(x) then by Lemma 4.11.

ea sr(x,x) = (eτr(e))a s(x,x) = x(ee)x = xex,

and similarly

ea sl(x,x) = xex.

By Lemma 4.1 (Lemma 4.1*) x translates Le(Re) to an () class of S and therefore ex Re(xe Le). Also ex Lxex(xe Rxex). Hence axioms (R2) and (R2)* hold. Axioms (R3), (R3)*, (R4) and (R4)* are easily verified. Axioms (R5) and (R5)* follow from the fact that for every (x,x) i,as(x,x) and as(x,x) are mutually inverse ω-isomorphisms (cf. Lemma 4.11) and the equality

ex = x(xex)(xe = (xex)x)

where eωre(eωle).

To verify axiom (R6), assume that x He,e,y Hf,f1,x = i(x;e,e1),h S(e,f) = S(x,y) and g ωl(e) ωr(f). Then

h1 = hasl(x,x) = xhx

and

g1 = gasl(x,x) = xgx.

Therefore by (4.8) and (8.7) we have

((xσl(e,h))(yσr(f,h)))σr(h 1,g1) = g1((x h)(hy)) = (xgx)(xy) = xgy = (xg)(gy) = (xσl(e,g))(yσr(f,g)).

Axiom (R6)* can be verified similarly.

We have now proved that s is a regular system. With respect to this system the product x y of x,y S is given by

x y = (xσl(e,h))(yσr(f,h))

where e E(Lx),f E(Ry) and h S(x,y).

Then by (8.7) and (4.8)

x y = (x h)(h y) = xy

where the product on the right is that of S. Therefore the new operation coincides with the operation in S.

Hence the theorem.

Definition 8.2. If is a regular system, we call the regular semigroup S() constructed in the above theorem, the regular semigroup of . Conversely, given the regular semigroup S, the regular system s defined by (8.8) is the regular system of S.

Lemma 8.5. Let = (S,a,σr,σl) be a regular system. Then E(S()) = E(S) and S = S()¯ where S()¯ denotes the regular groupoid of the regular semigroup S()¯.

Proof. Let S = S(), Es = E(S()) and E = E(S). Denote by ωr,ωl and τr,τl etc. the quasi-orders and partial translations of E and by ωsr,ωsl,τsr,τsl etc. those of Es. Since Es is the biordered set of idempotents of the regular semigroup S,ωsr,ωsl,τsr,τsl etc. are quasi-orders and partial translations defined by (1.10), (1.11) and (4.1). For x,y S the product of x and y in S, if it is defined, will be denoted by xy. The product of the same elements in S is denoted by x y.

Since the operation of S is the extension of the partial operation in S, it is evident that

E Es.

On the other hand suppose that x Es. Choose e E(Lx),f E(Rx) and h S(e,f). Then using the fact that x is an idempotent in S, (8.5) and (8.5*), we obtain

x = (xσl(e,h))(xσr(f,h)) = (x h)(h x) = x h x.

By the definition of product in S, we obtain from above that

h xxe in S,

and therefore

(h x) h = (h x)σr(e,h) L h.

Dually h (x h) Rh and hence h x h Hh,h. Further, since Hh,h contains an idempotent, (h x h)(h x h) Hh,h. But (h x h)(h x h) = h (x h x) h = h x h. Hence h = h x h, and so h x is an idempotent in S. This implies that e and h x are equivalent idempotents in S and therefore

e = e (h x) = (eσl(e,h))((h x)σr(h,h)) = (hτl(e))(h x)( by Remark 8.3 and axiom (R1)).

Therefore erhτl(e) and hence e = hτl(e). This implies that hx and dually we can show that hx. Hence x Hh,h and therefore x = h. Thus x E.

This proves that E = Es.

Let e,f E,eωrf. Then by Lemma 2.5 e = eτr(f) S(f,e) and by (8.2),

f e = (fσl(f,e))(eσr(e,e)).

Now by Remark 8.3 fσl(f,e) = e. Since ere in E, by axioms (R1) and (R3),

σr(e,e) = σr(e,e) = I Re = IRe.

Thus eσr(e,e) = e. Hence

f e = ee.

But ee = e in S and hence

f e = e.

This implies by (4.1) that eωsrf. Consequently ωr ωsr and dually ωl ωsl.

Further if eωrf, by Remarks 8.3, 8.5 and (4.1),

eτr(f) = fσr(f,e) = e f = eτ sr(f).

Hence

τr(f) = τ sr(f)|ωr(f).

Now for e,f E suppose that eωsrf. Then for h S(f,e),

e = f e = (f h)(h e).

Since e,f,h E,

h e = hτsr(e) = hτr(e),

and

f h = hτsl(f) = hτl(f).

Since (f h)(h e) is the product in S, h ele and hrh e. But h eωe and hence h e = e, i.e., hre. Since (f h)e is defined in S and (f,h)e = e, f hre. This implies that

f hlhrerf h,

and so f h = h. This shows that

hτl(f) = h,

and hence hωf. Therefore eωrf.

Thus ωr = ωsr E2 and dually ωl = ωsl E2. Suppose that h S(e,f), where e,f E. Since h is an idempotent in S,

e h f = (e h) (h f).

Now the product of (e h) and (h f) is defined in S and hence by (8.2), (8.5) and (8.5*)

(e h) (h f) = (e h)(h f) = e f.

Therefore by Theorem 4.4, h Ss(e,f). Thus we have proved that E is a biordered subset of Es.

Now let x Es and let e Lx E, f Rx E. Then by (8.2),

x = x x = (x h)(h x),

where h S(e,f).

Now (x h)(h x) is the product in S and hence x h Hh1,x and h x Hx,h. Since x,e,f are idempotents in S and elxrf, e f Hef and e f = (e h)(h f). Hence e h Hh,ef. But since e h = eσr(e,h), e h Lh. This implies that e h = h and dually h f = h. Consequently e f = h and erhlf. But since hωle, (h,e) ωl (ωr)1 and hence h = e. Similarly h = f. This implies that h = x. Therefore x E, and thus E = Es.

Now the sets underlying S and S()¯ are the same. Also product of x and y is defined in S()¯ if it is defined in S. Conversely if x y is defined in S()¯, Hy,x contains an idempotent. Since this idempotent is an idempotent of S, xy is defined in S also. Thus S and S()¯ denote the same regular groupoids.

Theorem 8.6. Let be a regular system. If S = S() then s = .

Proof. Let = (S,a,σr,σl). By (8.2), (8.5) and (8.7), if eωre and x Re, then

e x = xσr(e,e)

and

xσr(e,e) = e x.

Hence xσr(e,e) = xσsr(e,e).

Consequently σr = σsr and dually σl = σsl.

Now let (x,x) i,xx = e,xx = f. By the definition of the mapping S, the domain of S(x,x) is ω(e) which is the same as the domain of S(x,x) (cf. Lemma 4.11). Now let e ω(e). Then if

f = eS(x,x) = eSr(x,x)

by axiom (R5)

xσr(e,e) = xσl(f,f).

Hence by (8.5) and (8.5*)

ex = xf.

Since fωf, we obtain from the above equality that

f = ff = x(xf) = xex = eS s(x,x),

Therefore for all e ω(e)

eS(x,x) = eS s(x,x).

By Lemma 8.5, S = S̄ and hence

= s.

Remark 8.7: If R and Rs denote the classes of all regular systems and regular semigroups respectively, then by Theorem 8.4 the correspondence S() is a surjection of R onto Rs. Again by Theorem 8.6 this correspondence is also an injection. By defining morphisms of regular systems suitably, it is possible to make the correspondence (referred to above) an equivalence of categories of regular systems and regular semigroups.

Theorem 8.8. Let = (S,S,σr,σl), = (S,S,σr,σl) be two regular systems and let ϕ : S S be a homomorphism (an isomorphism). Then ϕ can be extended to a homomorphism (an isomorphism) of S() into (onto) S() if and only if the following condition and its dual are satisfied: For every eωre and x Re

(xσr(e,e))ϕ = xϕσr(eϕ,eϕ). (8.9)

Proof. Let x,y S, e E(Lx),f E(Ry) and h S(e,f). Then by (8.2)

x y = (xσl(e,h))(yσr(f,h))

where the product on the right hand side is defined in S hence if ϕ is a homomorphism of S into S, then

(x y)ϕ = (xσl(e,h))ϕ(yσr(f,h))ϕ.

By the definition of homomorphisms of regular groupoids ϕ|E(S) is a bimorphism and hence hϕ S(eϕ,fϕ). eϕ E(Lxϕ) and fϕ E(Ryϕ). Hence if satisfies (8.9), then

(x y)ϕ = (xϕσl(eϕ,hϕ))(yϕσr(fϕ,hϕ)) = xϕ yϕ

i.e., ϕ extends to a homomorphism.

Conversely if ϕ is a homomorphism and if eωre, x Re, then eϕωreϕ,xϕ Reϕ and

(xσr(e,e))ϕ = (ex)ϕ = eϕxϕ = xϕσr(eϕ,eϕ).

Statement (8.9)* can be verified dually.

If ϕ is an isomorphism it is clear that its extension is also an isomorphism. Hence the bracketed statement now follows.

By Theorem 5.17 and the definiton of strictly regular groupoid we notice that the regular groupoid S̄ of a regular semigroup S is strictly regular if and only if S is strictly regular. Similarly S is an inverse semigroup if and only if S̄ is an inverse groupoid. Thus we have the following theorem:

Theorem 8.9. The regular semigroup S() of a regular system is strictly regular if and only if the regular groupoid S of is strictly regular. Similarly S() is an inverse semigroup if and only if S is an inverse groupoid.

Remark 8.10: If S is an inverse groupoid, axioms for regular systems containing S can be simplified considerably. In this case for each x S, there exists unique x such that (x,x) i so that the mapping S(x,x) (in this case we can write x1 instead of x) may be regarded as a mapping of S into T(E,δ) where E = E(S). Hence we may write this mapping as S(x) instead of S(x,x1). The fact that E is a semilattice implies that the mappings Sr(x,x1) and Sl(x,x1) defined by (8.1) reduce to S(x) and S(x1) respectively. Further, axioms (R3) and (R4) are superfluous when S is an inverse groupoid.

If we define an inverse system = (S,S,σr,σl) as a regular system whose regular groupoid is an inverse groupoid, then we can state the axioms for inverse system as follows (here except for the changes introduced above, all other notations are those of definition (8.1)).

  1. For every e E, S(e) = Iω(e) and σr(e,e) = IRe.
  2. eωe,x Re xσr(e,e) He,eS(x).
  3. x S,xx1 = e,x1x = f,eωe and f = eS(x) implies
    xσr(e,e) = xσl(f,f)

    and

    S(x)1 = S(x1).

  4. Let x He1,e y Hff1 and gωef. If h1 = (ef)S(x1) and g1 = gS(x1) then ((xσl(e,ef))(yσr(f,ef)))σr(h 1,g1) = (xσl(e,g))(yσr(f,g)).

Theorems corresponding to Theorems 8.4 and 8.6 can also be proved for inverse systems in a similar manner.