Chapter 3
Partially Transitive Set of Partial ω-isomorphisms

Let E be a biordered set and let

δ0 = r lt (3.1)

where (r l)t denotes the transitive closure of r l. Since r and l are equivalence relations, so is δ0. By the definition of transitive closure, (cf. [2], p. 14), (e,f) δ0 if and only if there exists a finite ordered subset (e0,e1,,en) of E such that e0 = e,en = f and either eirei+1 or eilei+1 for i = 0,1,,n 1. A subset c of E satisfying this property will be called a δ0-chain connecting e and f (or e0 and en). If c = (e0en) is a δ0-chain, then c1 = (en,en1,,e0) is also a δ0-chain, which will be called the inverse of c. Further if c1 = (e0,,en) and c2 = (e0,,em) are δ0-chains and if en = e0 then c1 = (e0,,en,e1,em) is also a δ0-chain. c will be called the composite of c1 and c2; and c1 and c2 the factors of c. We may denote c by c1 c2. Given two arbitrary δ0-chains c1 and c2 their composite c1 c2 is defined if and only if the extremity of c1 (the last member of the ordered set c1) is the same as the origin of c2 (the first member of c2). This definition of composite can be extended to more than two δ0-chains and this operation is associative whenever it is defined.

By the definitions of δ0-chain, it is clear that (e,f) is a δ0-chain whenever erf or elf. We call the chain (e,f) an elementary chain. If e = f. (e,f) is trivially a chain. If c = (e0,e1,,en) is a δ0-chain, then (ei,ei+1) is an elementary chain for every i such that i = 0,1,,n 1 and (e1,ei+1) is a factor of c. Thus we have

c = (e0,e1) (e1,e2)(en1,en). (3.2)

Also if c1 and c2 are δ0-chains such that c1 c2 is defined, then c21 c11 is also defined and

(c1 c2)1 = c 21 c 11. (3.3)

If (e,f) is an elementary δ0-chain, define

ϵ(e,f) = Iω(e)  if e = f τr(e,f) if erf τl(e,f) if elf. (3.4)

We call ϵ(e,f) the elementary transformation associated with the elementary chain (e,f). If c = (e0,,en) is any δ0-chain, the chain transformation associated with c is defined by

ϵ(c) = ϵ(e0,e1) ϵ(e1,e2) ϵ(en1,en). (3.5)

Thus

ϵ(c1 c2) = ϵ(c1) ϵ(c2) (3.6)

whenver c1 c2 is meanigful.

By (Theorems Remark 2.11 and Remark 2.11*) it is clear that if c is a δ0-chain connecting e0 to en, then ϵ(c) is an isomorphism of the biordered subset ω(e0) onto ω(en). Also if (e,f) is an elementary chain, the transformation ϵ(f,e) is the inverse of ϵ(e,f) (and vice versa) and consequently

ϵ(c1) = (ϵ(c))1. (3.7)

Let c = (e0,,en) be a δ0-chain. Then for hωre0 we define a δ0-chain hc by

hc = (h,e0,e 1,,e n) (3.8)

where e0 = hτr(e0) and ei = ei1ϵ(ei1,ei),i = 1,2,,n.

By axiom (B1), e0rh and by (β1) and (β1), ei1rei or ei1lei according as ei1rei or ei1lei. Therefore hc is a δ0-chain and the factor (ei1,ei) of hc is either trivial or of the same type as (ei1,ei) . If c1 = (e0,,ei),i = 1,2,,n then it follows by induction that

ei = e 0ϵ(c i) andeiωe i.

Similarly if kωlen, we define the δ0-chain ck by

ck = (e0,,e n,k) (3.9)

where en = kτl(en) and for i = 1,2,n,eni = eni+1 = ϵ(eni+1,eni).

As before it can be shown that ck is a δ0-chain in which eiωei such that the factor (ei1,ei) is either trivial or of the same type as (ei1,ei). Since enωen we can define enc1 by (3.8) and then enc1 = (en,en1,e0). Therefore

ck = ((k,en) e nc1)1 (3.10)

and hc may be expressed in terms of chains of the form ck in a similar way.

Lemma 3.1. Let c = (e0,,en) be a δ0-chain in E. Then if hωre0 and kωlen, we have

ϵ(hc) = τr(h,e 0) ϵ(c) andϵ(ck) = ϵ(c) Tl(e n,k). (3.11)

Proof. Let f ω(h) and f0 = fτr(h,e0) = fτr(e0). Then by (3.5) and (3.8)

fϵ(hc) = (f0)ϵ(e0,e 1), ϵ(e n1,e n).

Now if e0re1, then e0re1 and by Theorem 2.11,

(f0)ϵ(e0,e 1) = (f 0)τr(e 1) = ((f0)τr(e 1))τr(e 0τr(e 1)) = ((f0)τr(e 0))τr(e 1) = (f0)τr(e 0,e1).

If e0le1, then (f0)ϵ(e0,e1) = f0τ(e0,e1) and therefore in either case

f0ϵ(e0,e 1) = f 0ϵ(e0,e1).

Inductively,

(f0)ϵ(e0,e 1) ϵ(e 1,e 2),,ϵ(e i1,e 1) = (f 0)ϵ(e0,e1) ϵ(ei1,ei).

Hence

(f)ϵ(hc) = (f0)ϵ(e0,e1) ϵ(e1,e2) ϵ(en1,en) = (f0)ϵ(c) = (f)τr(h,e 0) ϵ(c).

This proves the first equality in (3.11). The second can be proved similarly or may be deduced from the first using (3.3), (3.6), (3.7) and (3.10).

Corollary 3.2. Let c be a δ0-chain in E and hωe0(kωen). Then ϵ(hc)(ϵ(ck)) is the domain resriction of ϵ(c) (range restriction of ϵ(c)) to ω(h)(ω(k))

Proof. Since hωe0, e0 = hτr(e0) = h and it follows that τr(h,e0) = τr(h,h). Clearly τr(h,h) ϵ(c) is the domain restriction of ϵ(c) to ω(h). The bracketed statement can be proved dually.

Theorem 3.3. Let E be a biordered set and let 0 denote the set of all chain transformations of E. Then we have the following.

  1. Each ϵ 0 is a partial isomorphism.
  2. If ϵ 0 then ϵ1 0
  3. Let ϵ,ϵ0, range(ϵ) = ω(e) and dom(ϵ) = ω(f). Then if either eωf or fωe, ϵ ϵ0
  4. For e,f E,eδ0f if and only if there exists ϵ 0 such that dom(ϵ) = ω(e) and range(ϵ) = ω(f).

Proof. Statements (a) and (b) follow from the definition of chain transformation and Theorem 2.11. Statement (d) follows from (a) and the definition of δ0.

To prove (c) let ϵ,ϵ0. Then there exist δ0-chains c1 and c2 such that

ϵ = ϵ(c1) andϵ = ϵ(c 2).

Let range (ϵ) = ω(e) and dom(ϵ) = ω(f). Then e is the extremity of c1 and f is the origin of c2. If eωf by Lemma 3.1 we have,

ϵ ϵ = ϵ (Tl(e,e) ϵ) = ϵ(c1) (τl(e,e) ϵ(c 2)) = ϵ(c1) ϵ(ec2)

where ec2 is defined by (3.8). Now c1 ec2 is defined according to the definition of compositions of δ0-chains and hence by (3.6)

ϵ ϵ = ϵ(c ec 2).

Thus ϵ ϵ0. When fωe we can similarly prove that ϵ ϵ0.

An isomorphism of ω-ideals ω(e) and ω(f) of a biordered set E wiil be called a partial ω-isomorphism (or ω-isomorphism when no confusion is likely) of E. If α is an ω-isomorphism of E we define eα anf fα respectively by

domα = ω(eα),rangeα = ω(fα). (3.12)

We also denote by 1 the set of all ω-isomorphisms of E. Clearly every chain transformation is an ω-isomorphisms.

Hence

0 1.

Also, if α 1, and if g ω(eα), it is clear that

α|w(g) = τr(g,g) α

is also an ω-isomorphism. Similarly if g ω(fα). the range restriction of α to ω(g) (which is α τl(g,g)) is an ω-isomorphism. Hence if α,β 1 and if fαωeβ (or eβωfα) then

α β = α (τr(f α,fα) β) 1 (α β = (α τl(e β,eβ)) β 1).

Definition 3.1. Let E be a biordered set and T 1. Then T will be called a partially transitive set of ω-isomorphisms of E (or a partially transitive set on E) if and only if T satisfies the following axioms.

  1. 0 T
  2. α T α1 T
  3. If α,β T and if either fαωeβ or eβwfα then α β T.

If T is a partially transitive set on E, then we call the relation

δ(T ) = {(e,f) E2e α = e  and fα = f  for some α T} (3.13)

the transitivity relation of T, T a transitive set on E if δ(T ) = E2 and E a uniform biordered set if there exists a transitive set on E. A relation δ on E is said to be regular if there exists a partially transitive set T on E such that δ(T ) = δ.

Associated with a biordered set E, we have the following sets.

PT (E) = {T|T is a partially transitive set on E} (3.14) Re(E) = {δ :   for some T PT (E),δ(T ) = δ} (3.15) T(E,δ) = {α 1|α δ} (3.16) PT (E,δ) = {T PT (E)|δ(T ) = δ} (3.17)

where δ is a given relation on E . These sets are ordered by inclusion.

Theorem 3.3 and the discussion preceding Definition 3.1 give,

0,1 PT (E). (3.18)

By Theorem 3.3, δ0 = δ(0) and if

δ1 = δ(1)

we have

δ0,δ1 Re(E). (3.19)

Lemma 3.4. Let E be a biordered set. Then we have the following.

  1. PT (E) is a partially ordered set containing a greatest and a least member that is,
    T PT (E) 0 T 1.
  2. T PT (E) T T(E,δ(T ))
  3. For δ Re(E).
    1. δ is an equivalence relation
    2. δ0 δ.
    3. (e,f) δ if and only if for some α E1 ̃, eα = e,fα = f and α δ.

    Conversely if δ is a relation on E satisfying (c1), (c2), and (c3), then

    T(E,δ) PT (E) andδ(T(E,δ)) = δ. (3.20)
  4. The mapping
    δ : Tδ(T )

    is an order preserving mapping of PT (E) onto Re(E).

Proof. Statement (a) follows from definition (3.1) and (3.18).

Let α T. In order to obtain (b), by (3.16), it is enough to prove that α δ(T ). Let (g,h) α so that gα = h. Then by (T1), τr(g,g) T and since g ω(eα) by (T3), τr(g,g) α T. But if α = τr(g,g) α, then eα = g and fα = (g)α = gα = h. Hence (g,h) δ(T ).

To prove (c) suppose that δ Re(E). Then there exists T PT (E) such that δ(T ) = δ. For every e E,τr(e,e) E0 ̃ T by (T1) and hence by (3.13). δ is reflexive. Axiom (T2) implies that δ is symmetric and the transitivity of δ follows from (T3). Thus (c1) is obtained. Results (c2) and (c3) follow from (a), (b) and (3.13).

Now suppose that δ satisfies (c1), (c2) and (c3). Then if ϵ E0 ̃, by (b) and (c2), ϵ δ0 δ so that by (3.16) ϵ T(E,δ). Hence T(E,δ) satisfies (T1). If α T(E,δ),α δ and hence α1 δ1 = δ (since δ is an equivalence relation). Similarly if α,β T(E,δ) and fαωeβ (or eβωfα) then τr(fα,fα) β δ(α0τl(eβ,eβ) δ). Therefore

α β = α (τr(f α,fα) β) δ δ = δ, α β = (α (τl(e β,eβ)) β δ δ = δ).

Hence T(E,δ) PT (E). Now suppose that (e,f) δ. Then by (c3) there exists α , such that eα = e,fα = f and α δ. Then by (3.16), α T(E,δ). Hence (e,f) δ(T(E,δ)) by (3.13). On the other hand if (e,f) δ(T(E,δ)), there exists α T(E,δ) such that eα = e,fα = f. Then from the fact that α δ we get (e,f) δ. Hence δ(T(E,δ)) = δ. The statement (d) is evident from (3.13) and the definition of regular relations.

Let M PT (E). Define

M = {T|T M},M = {T|T M TT}. (3.21)

Similarly define and on Re(E) as follows. If D Re(E)

D = {δ Re(E)δ D δ δ}t, D = {δδ D}t. (3.22)

Theorem 3.5. Let E be a biordered set. Then the sets PT (E) and Re(E) are complete lattices under lattice operations defined by (3.21) and (3.22) respectively.

Proof. Let M be an arbitrary subset of PT (E). Then, if

TM = {T|T M}

it can be seen that TM PT (E) and it is also the greatest partially transitive set contained in every member of M. Consequently TM is the lattice product of M in PT (E). Also for any subset M of PT (E),

1 {T|T M TT}

so that this latter subset of PT (E) is not empty and its lattice product is the lattice sum of M. Thus PT (E) is a complete lattice.

To prove that Re(E) is a complete lattice we first show that for any D Re(E)

δD = [{δδ D}]t R e(E).

Evidently δD satisfies (c1) and (c2) of Lemma 3.4. Suppose that (e,f) δD. Then there exists a finite subset {δii = 1,2,,n} of D and {eii = 0,1,,n} of E such that e0 = e,en = f and ei1δiei,i = 1,2,,n. Then by (c3), there exists αi E1 ̄ such that eαi = ei1fαi = ei and αi δi δD. Then

α = α1 α2 αn δDn = δ D

eα = e and fα = f. Hence δD satisfies (c3) also. Thus δD Re(E). Since δD is the lattice sum in the lattice of all equivalence relations on E, it is the lattice sum in Re(E). Then as before it can be shown that D exists and is in Re(E). Hence Re(E) is also a complete lattice.

Theorem 3.6. Let E be a biordered set. Then

δ : T δ(T ) andT : δ T(E,δ)

are order preserving and join preserving, and order preserving and meet preserving mappings of PT (E) onto Re(E) and Re(E) into PT (E) respectively such that

T δ = IRe(E).

Proof. We have already proved that δ is an order preserving mapping of PT (E) onto Re(E) and by (3.16), it is clear that T is an order preserving mapping of Re(E) into PT (E). The equality T δ = JRe(E) follows from the fact that δ(T(E,δ)) = δ for every δ Re(E). Hence it remains to prove that δ is join preserving and T is meet preserving. Therefore if M PT (E) and D Re(E) we need show that

δ(M) = δ(M) and T(D) = T(E,D) (3.23)

where δ(M) = {δ(T )|T M} and T(D) = {T(E,δ)δ Re(E)}.

For convenience, we call a finite ordered subset c = (α1,,αn) of E1 ̃, a composable chain if fαi = eαi+1 for all i = 1,2,,n 1. Every singleton set can also be regarded as a composable chain (in this case the condition stated above is satisfied vacuously). If c = (α1,,αn) is any composable chain, we call

α(c) = α1 αn

the composite of c and if c is a singleton, the composite of c is the unique element of c. Let

M = {TT M}

and

M̄ = α(c)c  is a composable chain in  M. (3.24)

Then if α M̄, there exists a composable chain c such that α(c) = α. If c = (α1,,αn) is a composable chain in UM and if

c1 = (α n1,α n11,,α 11)

it is clear that c1 is also a composable chain in M and that

α(c1) = (α(c))1.

Hence M̄ satisfies axiom (T2) of definition (3.1). Since E0 ̃ M M̄, it satisfies (T1) also.

Now let c = (α0,α1,,αn) be a composable chain in M. Then

eα(c) = eα0 andfα(c) = fαn.

For e ω(eα0), define

e0 = e ei = (ei1)αi1,i = 1,2,,n

and

αi = τn(e i,ei) αi,i = 0,1,,n.

Then for i = 1,2,,n,fαi1 = ei = eαi, and hence c = (α0,,αn) is also a composable chain in M. Further

α(c) = τr(e,e) α(c) = α(c)ω(e).

Thus the domain restriction of the composite of a chain is again the composite of another chain. Similarly we can show that the range restriction of the composite of a chain is again the composite of a chain. These results are trivial when c is a singleton.

If c1 = (α0,α1,,αn), c2 = (α0,,αm) are composable chains, and if fαn = eα0 then

c1 c2 = (α0,,αn,α0,,α m)

is evidently a composable chain and

α(c1 c2) = α(c1) α(c2).

Hence if α(c1) α(c2) M̄ and if fα(c1)ωeα(c2) then

α(c1) α(c2) = α(c1) τr(f α(c),fα(c1)) α(c2) M̄.

In the case when eα(c2)ωfα(c1), we see by similar arguments that α(c1) α(c2) M̄. Thus M̄ satisfies (T3) also and M̄ PT (E).

Since M M̄ by the definition of lattice sum we have M M̄. On the other hand if c is a composable chain in M, axiom (T3) implies that α(c) M and so M̄ M. Thus

M = M̄. (3.25)

We can now prove that δ(M) = δ(M). Since δ is order preserving, the fact that δ(M) is the lattice sum of equivalence relations in δ(M) implies that

δ(M) δ(M).

On the other hand if (e,f) δ(M), there exists α M such that eα = e,fα = f and α δ(M). Now by (2.25), there exists a composable chain c = (α0,,αn) in M such that α(c) = α. Since for all i,αi M there exists Ti M such that αi Ti. Hence eαiδ(Ti)fαi, where eα0 = e and fαn = f. Therefore

(e,f) δ(T0) δ(T1) δ(Tn) δ(M).

This proves the required equality.

It remains to prove that T(D) = T(E,D).

Let δ̄ = δ(T(D)). Since the mapping δ is order preserving, δ̄ δ for all δ D and since T is also order preserving, T(E,δ̄) T(E,δ). Thus

T(E,δ̄) {T(E,δ)δ D} = T(D).

But by Lemma 3.4(b), T(D) T(E,δ̄) and therefore T(E,δ) = T(D). For all δ D since δ̄ δ, we have by the definition of lattice products δ̄ D. Hence T(E,δ̄) T(E,D). On the other hand, if α T(E,D),α δ for all δ D and hence α T(E,δ) for all δ D. This implies that α T(D).

This completes the proof.

Before concluding this section, we prove three lemmas needed in Chapter Lemma 5.

Lemma 3.7. Let e E. If e1,e2 ωl(e) and e1reα then

τr(e 1,e2) τl(e 2,e2τ(2)l) = τl(e 1,e1τ(2)l) τr(e 1τl(e),e 2τl(e)).

Proof. Let h ω(e1). Then h ωr(e2) and (h)τr(e1,e2) τl(e2,e2τl(e)) = ((h)τr(e2))τl(e2τl(e)). Now hτr(e2)ωle2τl(e)ωe and hence by (B2),

(hτr(e 2))τl(e 2τl(e)) = ((hτr(e 2))τl(e))τl(e 2τl(e)) = ((hτr(e 2))τl(e 2))τl(e)

since τl(e) is a bimorphism. Hence

(h)τr(e 1,e2) τl(e 2,e2τl(e)) = (hτl(e))τr(e 2τl(e)).

But hτl(e)ωe1τl(e) and hence by (B2)

hτr(e) = (hτl(e))τl(e 1τl(e)) = hτl(e 1τl(e)).

Therefore

(h)τr(e 1,e2) τl(e 2,e2τl(e)) = (hτl(e))τr(e 2τl(e)) = (h)τl(e 1,e1τl(e)) τr(e 1τl(e),e 2τl(e))

which is the required equality.

Lemma 3.8. Let e,f,g1 and g2 satisfy the conditions of axiom (B5). Then

αl(g 2,e) αr(g 2τl(e),g 1,τl(e)) αl(g 1,e)1 = αr(g 2,f) αl(g 2τl(f),g 1,τr(f)) αr(g,f)1.

where, for e1ωre2(e1ωle2)

αr(e 1,e2) = τr(e 1,e1τr(e 2))(αl(e 1,e2) = τl(e 1,e1τl(e 2))) (3.26)

and αr(e1,e2)1 denotes the inverse of αr(e1,e2).

Proof. Since e,f,g1 and g2 satisfy conditions of axiom (B5), there exist g2,g2 and g1 such that statements (a), (b) and (c) of Lemma 2.7 are satisfied.

Hence

g1τl(e) = (g 2τr(g 1))τl(e) = (g 2τl(e))τr(g 1τl(e)).

Therefore by (3.26) and by (c) of Lemma 2.7, we have

(e)αl(g 2,e) αr(g 2τl(e),g 1τl(e) αl(g,e)1) = (eτl(g 2τl(e))τr(g 1τl(e))τl(g 1))

for every e ω(g2). Since τl(e) is a bimorphism, by axiom (B2) we have

(e)τl(g 2τl(e)) = (eτl(e))τl(g 2τl(e)) = (eτl(g 2))τl(e)

and

((e)τl(g 2τl(e))τr(g 1τl(e)) = ((eτl(g 2))τl(e))τr(g 1τl(e)) = ((eτl(g 2))τr(g 1))τl(e).

Hence

(e)αl(g 2,e) αr(g 2τl(e),g 1τl(e)) αl(g 1,e)1 = (((e)τl(g 2))τr(g 1))τl(e)τr(g 1)) = ((eτl(g 2))τr(g 1))τr(g 2τr(g 1)) = ((eτl(g 2))τr(g 2))τr(g 1) = (eτl(g))τr(g 1) = (e)τl(g 2,g2) τr(g 1).

We can prove dually that

(e)αr(g 2,f) αl(g 2τr(f),g,τr(f)) αr(g 1,f)1 = (e)τr(g 2,g2) τl(g 1).

The lemma now follows from axiom (B5).

Lemma 3.9. Let α . Then for every e,g E,h1 S(e,eα) and h2 S(fα,g), there exist h S(e1,(h2τl(eα))α1) and hS((h1τr(eα))α,g) such that

(hτr(e α))α = hτl(f α).

Proof. Let h1 = (h1τr(eα))α and h2 = (h2τl(fα))α1. We first prove that

S(h1,h2) S(e,h 2),S(h 1,h 2) S(h1,g). (3.27)

Suppose that h S(h1,h2) and hS(e,h2). By the definition of h2, h2 ω(eα) and hence

h ωl(e) ωr(e α).

Therefore,

(hτl(e),hτr(e α))ωr × ωl(hτl(e),hτr(e α)).

Thus by axiom (B5), there exists h E such that

hrhωlh 1,hτr(e α) = hτr(e α) andhτl(e)rhτl(e).

Further, by axiom (B2), we have

hτr(h 2) = (hτr(e 2))τr(h 2) = (hτr(e α))τr(h 2) = hτr(h 2).

Hence hS(e,h2) ωl(h1) and therefore

h,h ωl(h 1) ωr(h 2).

Since h S(h,h2) and hS(e,h2) we have

hτr(h2) = hτr(h2), hτl(h1)ωrhτl(h1) andhτl(e)ωrhτl(e).

Then

hτl(h 1) = (hτl(e))τl(h 1)ωr(hτl(e)τl(h 1)) = hτl(h 1),

which implies that hS(h1,h2).

From the facts that hS(e,h),h ωl(e) ωr(h) and by axiom (B5) we obtain that there exists h such that

hlhωrh hτl(e) = hτl(e) andhτr(h2)lhτr(h2).

Since hτr(h)lhτr(h2),hωrh we have

hτr(h 2)lhτr(h 2)  andhτr(h 2)ωrhτr(h 2).

Thus

hτr(h 2) = hτr(h 2) hrh hτl(e)rhτl(e) = hτl(e) h S(e,h2).

Consequently S(h,h2) S(e,h2). The other inclusion relation in (3.27) can be proved dually. Now to prove the lemma, choose h S(h1,h2). Then by (2.5)

hτr(e α) S(h1,h2) τr(e α) S(h1τr(e α),h2).

Since α is a partial ω-isomorphism, we have

(hτr(e α))α S((h,τr(e α))α,h2α) = S(h 1,h 2τl(f α)).

But S(h1,h2τl(fα)) = S(h1,h2)τl(fα) and hence there exists hS(h1,h2) such that

(hτr(e α))α = hτl(fα)

and hence the lemma.