(RxZ) (XxS) -- ---- XxYxZ -- ----> XxZ

cl.T

RoS

It is evident that T RRoS whence clT C RoS. But (RoS,j') is the

intersection of all closed subsets of XxZ through which Y factors.

Thus (RoS,j') C (clT,j). Hence (RoS,j') (clT,j).

1.35. Example. With the hypothesis of Example 1.34, T and clT do not

necessarily coincide.

Proof. Let X = Z be the closed unit interval with the usual subspace

topology induced from the real line. Let Y be the closed unit interval

with the discrete topology. Let R = {(x,y): y = x} considered as a

subspace of XxY. Let S = {(y,z): 0 < y < } considered as a subspace

of YxZ. It is easy to see that both R and S are closed in XxY and YxZ

respectively.

Clearly T = {(x,z): 0 < x < } and c3T = {(x,z): 0 < x < }

whence T clT.

1.36. Example. In the category Top, the composition of relations is
-2
not necessarily associative.

Proof. Let X = Z be the closed unit interval with the usual subspace

topology induced from the real line. Let Y be the closed unit interval

with the discrete topology. Let R = {(,)} considered as a subspace

of XxX. Let Sbe{(x,y): y = x} considered as a subspace of XxY and let

Tbe{(y,z): 0 < y < } where T is considered to be a subspace of YxZ.

Hence, each together with its inclusion map is a relation since each

of R, S, and T is a closed subspace of XxX, XxY, and YxZ respectively.

It follows that RoS = {(,)} and that (RoS)oT = 0. But

SoT = {(y,z): 0 < y < } and from this it follows that

Ro(SoT) = {(,z): zeZ}. Hence Ro(SoT) # (RoS)oT.

1.37. Remark. At first glance, the results of Examples 1.34, 1.35 and

1.36 seem to be pathological, thereby casting doubt on the usefulness

of the categorical definition of composition of relations (1.26). How-

ever, this should cause no more anxiety that1 does the fact that the

set theoretic union of two subgroups of a group is seldom a subgroup.

Furthermore, the results 1.31, 1.38, 1.39, 2.4, 3.1, 3.6, 3.9,

3.10, 3.12, 4.22, 5.20, 5.23, 5.25, 5.26, 5.27, 5.34, 6.13 and 6.27

seem to indicate that this definition (1.26) yields nice theorems which

re-enforces its appropriateness.

1.38. Theorem. Let (R,j) be a relation from X to Y and let (S,k) be a

relation from Y to Z. Then (RoS)-1 and S-loR-I are isomorphic relations

from Z to X.

Proof. The following products shall be used:

(Xxy,T1~ 2), (YxZ,1, 2), (RxZ,pl*,o2*), (XxZ,Pl,p2), (XxS,1 ,51 2)

(Xx(YxZ),lT '2*), (XxyxZl, 1'2,iT3), and ((XxY)xZ,pl,p2).

The notation " over a projection morphism shall denote the

projection morphism of that product object where the product is taken

in reverse order; i.e., the projections of YxX are r1 and 72 and the

projections of ZxY are TfI and T2.

Consider the following diagram. It will be shown to be commutative.

ZxR-1

zP 2,' 1 >
RxZ *y (XXY)xZ '> -- "P ZX (Yxx)/

V01 IV )

(RxZ)CI(xxS) X. : xyxZ >~ -- Zxy>rX

X2 02 02

x XS 1-1') [X(YXyZ) t~--------s (Zxy) cX

k*xl1

"X,

37

01 = < `Ip2',2P2> and 02 = <171 T*, 2 27*,'7 2

11 = P1P> = P2"

^1 <3'2 1 = 3l = P2

2 1 p 1> 71P2 = I <7r2,17T>p1 = 72PP.

^72

1 30]p]> = '2P2pl> = 7T2<'r2,7I>PI = n1PI.

I 1 1 2' < -2I' 311 2 1 >(D 1 7 12 1 2 2T12*
" 3<13'T2,1> = 31 2 = 2 22.

i 02<2 1,T1 *> 2= 2 1 T" 2, 1> T<, T l*> f2< 2,']>7n2* = 1712*.

2 <7T 3 ,1T r> = 12 1 2 2 = 2

3 0222" ,7 i*> = 2T1*T<2',12f>72*,71*> 71= '

3n22 = 212 = 2

Thus 01
p1 > = 1 and 02 = I2 <2, >>i7r*,71*> .

0l

p >(jxl) = P2(jx1) = p *.
P(ij*X

p *

= *.
1 2 1 1 2 12

2 2 1 2 2 1 21p
p (lxj*-) = J*p2*

= J-PI*".

Thus >(jx]) = (Ixj:)

1<2' 2 I>*>(1xk) = < 2' > ,2*(1Xk) kp2 = k*T

Il(k*xl) ,1> = k*pl

= k*-T2.

2 2<2 T *>(lxk) = 1*(1xk) = PI.

2(k*xl) = 2 = ~1

Thus <.T2*1*>(lxk) = (k*xi).

Hence the diagram is commutative.

Consequently by the definition of intersection, there exists a

unique morphism & such that the following diagram commutes.

Y

It is easy to see that the following diagram commutes.

Y X
(RxZ) r (XxS) >---------- XxxZx

(S-lxX)( (ZxR-1) <--- ~I'3 Z xX

S-1oR-1

Since (RoS,6) is the intersection of all extremal subobjects

through which y factors, then (RoS, ) < (S oR ,a ).

-1 -
Hence there is some morphism p such that a'p = B. Consequently,

ip=
S = ts' = t t and the following diagram commutes.

S- oR-1

-2 i \ 2.'1> a2

SB
RoS XxZ -------------- ZX

(RoS)-1 Y'

Since ((RoS)-1,B*) is the intersection of all extremal subobjects

through which B factors then ((RoS)-1,B*) < (S-loR-1,a').

Now applying the above result to (S-1,k*) and (R-1,j*), it fellows

that ((S-loR-1>)-1,a') < ((R-1)-o(S-1)-1,B#) E (RoS,Z) (1.11) whence

(S-loR-1,a') < ((RoS)-1,6*) (1.12), so that

1.39. Corollary. Let (R,j) be a relation from X to Y. Then (RoR-1,j#) is

a symmetric relation on X and (R-IoR,j') is symmetric on Y.

Proof. ((RoR-1)-1,j#*) E ((R-1)-1oR-1,j) (RoR- ,j#) and

((R-loR)-1,j*) = (R-1o(R-l)-I,j) = (R-IoR,j") (1.38 and 1.11).

1.40. Proposition. Let (R,j) be a relation from X to Y and leL (S,k) and

(T,m) be relations from Y to Z. Then

(Ro(S(T),e) < ((RoS)(1(RoT),6).

Proof. By Proposition 1.5 there exist canonical isomorphisms:

i: (RxZ) (Xx (S (' T) ----- (RxZ) .( (XxS) 0 (XxT)

y: Xy(S AT) r--------- > (XxS) 1(XxT).

Consider the following commjutative diagrams.

X2
S TT ---- S----

A1 O^ k

T -------- yz
m

ix1
RxZ -------(XxY)Z

'B 01

xx (sr (T) >-S---- ------R (Xxs) xT) (X T) ----- ---- (yxZ)

Let (T,B) be the epi-extremal mono factorization of <, ,w3->y. Thus

the codomain of T (domain of S) is Ro(S{ T). Since this is the intersec-

tion of all extremal subobjects through which yI factors it follows
that (Ro(SfT),6) (RoS,61) and (Ro(SX.T).) < (RoT,62). Thus

(Ro(SAT),8) < ((oS)A)(RT),C) (1xa19).

SECTION 2. CATEGORICAL CONGRUENCES

2.0. Remark. Lambek [16,pg 93 presents the following definitions for

dealing with rings which have identities.

More general than homomorphism is the concept of homo-
morphic relation. Thus let 6 be a binary relation between rings
R and S, that is essentially a subset of the Cartesian product
RxS, then 6 is called homomorphic if 000, 191, and rl0s1, r20s2
imply (-rl)6(-sl), (rl+r2)6(sl+s2), (r r2)6(s1s2). Of course a
similar definition can be made for any equationally defined
class of algebraic systems.

A homomorphic relation on R (that is, between R and itself)
is called a congruence relation if it is an equivalence rela-
tion, that is reflexive, symmetric, and transitive.

Lajibek notes that a symmetric transitive relation is not neces-

sarily reflexive, but is a congruence on a subring. He also notes that

a reflexive homomorphic relation is a congruence. This latter result

is due to the fact that all homomorphic relations are difunctional

(see 5.22).

We will generalize all of these results. However it must be noted

that in the category Rng a congruence is an equivalence relation and

conversely. Thus we shall obtain the result that if (R,j) is a symme-

tric transitive relation on an object X then R is an equivalence rela-

tion on an extremal subobject of X. However, this result must be post-

poned until Section 3 (see 3.4 and 3.10).

Also the result that the reflexive diiunctional relations are pre-

cisely the equivalence relations must be postponed until Section 5.

In this ?ecticn it will be shcwn that a (categorical) congruence

is a (categorical) equivalence relation and that congruences (when '

has coproducts) are determined by (categorical) quotients (2.12).

If f is a set function from a set X to a set Y then the set

{(x1,x2)EXxX: f(x1) = F(x2)}

is called the congruence (sometimes kernel) determined by f. It will

be shown that (categorical) congruences have behavior similar to that

of the above set (2.8, 2.10, 2.11, and 2.12).

2.1. Definition. If (R,j) is a subobject of XxX then (R,j) is called a

congruence if and only if there exists a morphism f with domain X such

that (R,j) is the equalizer of fnl and fr2.

j 1 f
R ----- XxX .--X > Y
T2

If g is a morphism with domain X then the equalizer of girl and gn2

denoted by (cong(g),i ) is called the congruence generated by g.

2.2. Remark. If X is a -object then (Ax,i ) is the congruence gener--

ated by 1X'

2.3. Remark. It is easy to see that (R,j) is a congruence on X if and
-1
only if (R ,j*) is a congruence on X.

2.4. Theorem. If (R,j) is a congruence on X then (R,j) is an equivalence

relation on X.

Proof. Since (R,j) is a congruence on X there exists a miorphism f with

domain X such that (R,j) is the equalizer of fir, and firi. Recall that

(AX,i ) is the congruence generated by I whence iL Xir = 1XT2iX. Thus

fTi = fn2Si so by the definition of equalizer there exists a morphism
X -Y

A from AX to R for which jX = iX. This implies that (AXiX) < (R,j) so

that (R,j) is reflexive.

To see that (R,j) is symmetric, observe that

fTrlj fr2 = f1r1jT^J Tr2j. Thus flrj*-T f72j*T so that

since T is an epimorphism it follows that flj* = fj*. Hence, from the

definition of equalizer, there exists a morphisnm from R-I to R for

which jn = j*. This implies that (R-",j*) < (R,j) so that (R,j) is sym-

metric.

Consider the following products: (XxX,Tl,Tr2), (XXXxX,712,T 23),

((XxX)xX,plp2), (Xx(XxX),P1,p2), (RxX,1 *,T2*) and (XxR,I,i2). To see

that (R,j) is transitive, consider the following commutative diagram.

jx1X
RxX >>---- (XxX)xX

Y
(RxX) ri (XxR) ------ ------------------- XxxxX

XxR --------------- Xx(XxX)
1Xxj

Let (T#,j#) be the epi-extremal mono factorization of y.

Recall that the codomain of T# (domain of j#) is RoR.

Next, it will be shown that fiy = f-2y = fY-.

y = f 302(]xj)X2 = fT22(1Xxj).2 = fT2ji2,2 = fljt2A2 =

f-l 2(PX:<)X)2 = f 022(lXxj)'2 = f 2y.

fly f= 10i(jx-lx)A f1pI1(jxlX)i f1 jTl*'/ =- f2J 1i*l

fr2p1(jxly)XA = f 12il(jx1X)X1 = f 2y.

Thus fT,;~i- 'i, 3>"' = fy = fTy fT 2<,3>y; so fTrj/T# fi#2

Again, since T# is an epimorphism, it follows that firT3j = frr2j#. By

the definition of equalizer there exists a morphisrm k from RoR to R for

which jk = j#. This implies that (RoR,j#) < (R,j) so that (R,j) is tran-

sitive.

2.5. Theorem. The intersection of any finite family of congruences on

any -object is a congruence.

Proof. Let {(Ei,ei): iI)} be a finite family of congruences on X. Then

there exist morphisms fi with domain X such that (Ei,ei) is the equal-

izer of fini and fi72 (2.1). Let the codomain of each fi be denoted Yi.

Consider the morphism from X to TTYi and consider
iel isI
the intersection ( ( Ei,e).
ieI

It will be shown that (f1 Eie) is the equalizer of 7rl and
iel iel
T .2'
iel

e 71 f i
SEi-- XxX ~~_. X ---- "-- Y

Xi ei fi V Pi

Ei Yi
ie

First observe that: pjlne = fj e = f 72e = p r2e for
ici isI
each jIe. Thus 7Te 7 e.
iel iel

Now if g is a morphism from W to XxX such that rr2g
irI ieI
then fj7Tg = p 1g = PjTr2g = fj2g.Thus by che definition of
iel iel
equalizer there exist morphisms ki from W to Ei so that eiki = g for

each ilI. Thus by the definition of intersection there exists a morphisn

k from W to (\ Ei such that ek = g. This implies that (f Ei,e) is the
ieI icI
equalizer of 72.
ieI ieI

2.6. Proposition. If is complete then the intersection of any fam-

ily of congruences on any -object is a congruence.

Proof. Repeat the proof of 2.7 assuming I to be infinite.

2.7. Proposition. Let Q be the family of all congruences on X and let

(fQ,p) be the intersection of this family. Then ( O and Ax are isomor-

phic relations on X.

Proof. If (E,e)eQ then (Ee) is an equivalence relation and hence is

reflexive (2.4). Thus (xX,ix) < (E,e). Hence (Ax,iX) < (r(Q,p). But

(Ax,ix) is a congruence; hence (i'P,o) < (AX,iX).

2.8. Proposition. Let f be a -morphis- from X to Y. Then f is a

monomorphism if and only if AX and cong(f) are isomorphic relations on X.

Proof. Since (cong(f),if) is an equivalence relation (2.4) it is reflex-

ive and hence (Ax,iX) < (cong(f),if). If f is a monomorphism then

fTlif = fn2if implies that "lif = T2if. Hence there exists a morphism k

for which iXk = if and consequently (cong(f),if) < (Ax,iX).

Conversely, suppose that (cong(f),if) (Ax,i') a and a and are

morphisms having domain Z and codomain X such that fa = f. Consider

the morphism from Z to XxX. fT = fa = f, = fa2 so that

there exists a morphism X from Z to AX for which iX = . Thus

a = 1:1 = nlijXX = 2 = B. Consequently a = 8 so that

f is a monomorphism.

2.9. Definition. A -morphism f from X to Y is said to be constant if
ad only if for all pairs of mrphiss a
aud only if for all pairs of morphisms Z TLT2ET^. X, fa = f6.

2.10. Proposition. Let f be a morphism from X to Y. Then f is constant

if and only if (cong(f),if) F (XxX,1X).

Proof. If f is constant then fir = f2r so that fIrlx = f2X 1 Thus
2-X 1 AA 2 XXX
there exists a unique morphism k from XxX to corig(f) for which ifk = 1XxX

whence if is a retraction. But since if is an equalizer, it must be an

isomorphism (0. 4 ) so that (cong(f),if) and (XxX, 1XxX) are isomorphic

relations on X.

Conversely, suppose that (XxX,1XxX) (cong(f),if) and that a and

6 ere morphisms with common domain,and codomain X. Consider from

Z to XxX where Z is the common domain of a and B. Since

falXx, = f21X, it follows that fir = fT2 so that

fa = fT1 = fi2 = fB. Thus f is a constant morphism.

2.11. Proposition. If f from X to Y, g from Z to Y, and h from X to Z

are ( -morphisms such that f = gh then (cong(h),ih) < (cong(f),ir).

Furthermore if g is a monomorphism then (cong(h),ih) (cong(f),if).

Proof. Since hrlih = h 2ih it follows that ghrlih = ghT2ih so that

fTlih = fT2ih. Thus there exists a morphism k from cong(h) to cong(f)

for which ifk = ih. Whence (cong(h),ih) < (cong(f),if).

If g is a monomorphism then f'lif = fn2if = ghl1if = ghr2if

implies that hTlif = hi 2if. Thus there exists a morphism k* from cong(f)

to cong(h) for which ihk* = i,, whence (cong(f),if) < (cong(h),ih). Con-

sequently (cong(f),if) = (cong(h),ih).

2.12. Proposition. If ( has coequalizers and f is a I -morphism from

X to Y and if (f*,Z) is the coequalizer of Tiif and r2if then cong(f)

and cong(f*) are isomorphic relations on X.

Proof. Since fnif fiT2if then by the definition of coequalizer there

exists a morphism k* from Z to Y for which k'f* = f. Since

f*li = f2if it follows that flif = k*f*rri = k*f*if =

fr 2i ,. Thus there exists a morphism k from cong(f*) to cong(f) for

which i k = i Consequently (cong(f*),if,) : (cong(f)i ).
f f C f
Now since f* is the coequalizer of r if and i2 i then

f*7lr i f2r i Hence there exists a morphism k' from cong(f) to

cong(f*) for which i fk' = i Consequently (cong(f),if) <

(cong(f*),if,).

2.13. Proposition. If is complete and 0 is a family of congruences

on X generated by morphisms f: X ---->Y and if (,r1,p) (A ,iX) then

the unique morphism 0 from X to TT Y such that rfO = f, is a monomor-

phism.

Proof. Observe that for each f, fir i = T~f, r i = 2i = f i 0. Thus
-- 10 r 2f 10f
f
it follows that (cong(),i ) < (cong(f),if) for all X f--- Hence
(cong(0),i ) < ((n,p) (1.19). Since ((),p) (A,iX) < (cong(0),i)

(2.4) it follows that (f ,p) (A ,iX) = (cong(O),i). Thus 0 is a

monomorphism (2.8).

2.14. Corollary. If is complete and 0 is a family of congruences

on X generated by morphisms f: X--- Y and for some g: X --- Yg,
f 0
g is a monomorphism, then the unique morphism 0 from X to TT Yf such

that f 0 = f is a monomorphism.

Proof. Since g is a monomorphism then (cong(g),i ) = (A ,iX) (2.8). Thus
g X X
(C0I,p) < (A,i X) by the definition of intersection. But

(AX,iX) < (f[I,p) (2.4 and 2.6). Consequently (6X.iX) = ((In,p) and the

result follows from Proposition 2.13.

SECTION 3. CATEGORICAL EQUIVALENCE
RELATIONS AND QUASI-EQUIVALENCE RELATIONS

3.1. Theorem. If {(E.,4 ): icl} is a family of equivalence relations on

a -object X then their intersection (rf E.,4) is an equivalence rela-
iel
tion on X.

Proof. Since (AX,i ) < (E., .) for each iel it follows that
X k 1 1
(A ,i ) < ( Ei.4) (1.19). Hence (\n E.,<) is reflexive.
iel iel
Since (fl E.,4) < (E.,V.) for each iel and since each (E.,i.) is
isl
symmetric it follows that (( \ Ei)-l,*) < (E -1, i*) < (E.,,.) for each
I 1 1 1
iEI
isl (1.12). Thus ((ft E.)-1,4*) < (r\ Ei,4) and hence (r Ei,r) is sym-
ilI iel iel
metric.

Since ((~l Ei,) < (Ei,4i) for each ilI then
iEI
((C E.)o(R E.),#/) < (E.oE'.,4#) for each ilI (1.30). And since
idI icI
(EioEi',i#) < (Ei,4i) for each icI it follows that

((f, Ei)o( ( Ei),#) < (fn E. ,) (1.19) whence (f Ei, ) is transitive.
ieI iel iIe iil
Thus it is an equivalence relation.

3.2. Definition. A quasi-equivalence (R,j) ou X is a relation on X which

is both symmetric and transitive.

This term is due to Riguet [221 ; however, Lambek  calls

symmetric transitive relations subcongruences. Wiile this term sub-

congruence is appropriate in the categories Grp and Ab, it does not seen

co be appropriate in more general categories. MacTane I calls such

relation .s sy1-Jcric idemporents.

3.3. Proposition. If (A,a) is an extremal subobject of X then (AxA,axa)

is a quasi-equivalence on X.

Proof. Consider the products (AxA,pl,p2) and (XxX,Tr', T2). Since a is an

extremal monomorphism then axa is an extremal monomorphism (0.20) and

hence (AxA,axa) is a relation on X.

Consider the following commutative diagram.

axa < il ,Il>
AxA >---- ---- XxX >-- XxX

T (axa)*

(AxA)-1

Since 7l <;T2',~>(axa) = 2(axa) = ap2;

ir (axa) = ap1 = ap2;

7r,2 <2'l>(axa) = l (axa) apl; and

,2(axa) = ap2 = apl

then it follows that (axa) = (axa). But is an iso-

morphism hence an epimorphisn and axa is an extremal monomorphism;thus by

the uniqueness of the epi-extremal mono factorization of <72',71>(axa)

(0.13), ((AxA)- ,(axa)*) = (AxA,axa). Thus (AxA,axa) is symmetric.

To see that (AxA,axa) is transitive, first consider

(((AxA)xX)tA(Xx(AxA)),y) where y is the unique extremal monomorphism

induced by the indicated intersection. It will next be shown that

(AxAxA,axaxa) and (((AxA)xX),1 (Xx(A';A)),y) are isomorphic as extremal sub-

objects of XxxxX. To show this it will be shown that (AxAxA,axaxa) is

precisely tha intersection of ((AxA)xX,0 ((axa)xlX)) and

(X:-(A,-A),02 (l x (ax;a)).

Consider the products: (XxXxX, 1, 2, 3), (Xx(AxA),plp2),

((AxA)xX,I, 2), (Xx(XxX),l '*r *), ((XxX)xX, 11 2*) and

(AxAxA,p 1,p2'p3

Observe the following qualities.

S1] ((axa)x) 1,> = ((axa)xl))<

,a3 > =

1T(axa)1P'P2> ,ap3> = l (axa)2

ap1 = ap1 = (axaxa).

T2 1((axa)xl x)<

,ap3 >

3, 1((axa)xl x

,ap3>

102(lx (axa))>

T202 (2Xx(axa))>

r302 (x( (axa))>

-= 2TrI((axa)xl )<

,a 3>

T2(a-a)< P'P2> = aP2

ap2 = T2(axaxa).

2- *((axa)xl)1',p2>,ap3> =

1XP2<'ap3> = lXap3 = ap3 = '3(axaxa).

= Ir*(1Xx(axa))
1Xp = 1 ap1 = aP1 = l(a>aaxa).

SiTr2*(1Xx(axa))
7l(axa)p2> = l(axa) =

ap1 = ap2 = T2(axaxa).

= T2'r2*(lx(axa))> =

2(axa) = ao2 = ap3
77,(axaxa).
.3

Thus by the definition of product the following diagram conmlutes.

(axa)xlX
(AxA) xX ---------- (XxX) xX

<<'1 2>,ay axaxa
AxAxA ,---------------- ----------- XxXxX

,P23 Xx (AxA) -------- Xx(XxX)
lxx(axa)

Now, if (W,6) is a subobject of XxXxX so that there exist mor-

phisms y and y such that 0 ((axa)xl )y = 6 = (1 Xx(axa))y then con-
1 2 1 X i 2 X 2
side the morphism

,ap >E = y .
1 2 3 1
Again since Q ((axa):.l ) = 0 (1 x(axa))y it follows that
X 2 X 2
P1 2 1= *(Xx(axa))y2 = 2 0 (1 X(axa))y2 1 11((axa)x1 )yX =

T Ift 1*((axa)xl )y = Ir (axa)pl = ap

Hence pl> = ap a = ap 1 Y= 1
1 3> 1 111 1 2
Since p 2 P = p2 = P ahd p2P2 = = P2Y 2 2 it

follows that F p y2. Hence p > =

= Y
2 3 2 2 2 3 2 3" 2
Consequently 1 2 3 2
From this it follows that (axaxa) = 6. Since (axaxa) is a mono-

morphism the morphism & is unique. Thus it has been shown that

(AxAxA,axaxa) is the intersection of (Xx(AxA) ,2 (1 x(axa))) and

((AxA)xX,01((axa)x]X)). It next will be shown that the following diagram

commutes.

axaxa < Tr1,3
AxAxA ------ XxXxX --------------- XxX

-

"'AxAA A

(Ti,j#) is the epi-extremal mono factorization of (axaxa).

Now r1(axa) = ap1 api = 71(axaxa) and

[2(axa) = ap2< i,P3> = ap3 = 72<1 T, 3>(axaxa). Thus the above

diagram commutes..

Since ((AxA)o(AxA),j#) is the intersection of all extremal sub-

objects through which 3>(axaxa) factors then

((AxA)o(AxA),j#) < (AxA,axa)

whence transitivity is obtained.

3.4. Canonical Embedding. Let (R,j) be a relation on X. Let (Tl,jIl

(c2,J2), and (-3,j3) be the epi-extremal mono factorizations of T1j,

21j, and T2j* respectively. Let RX, XR. and XR-1 denote the domains of

ji, j2, and j3 (codomains of TIp2 and T3) respectively.

R -------------- -----> xX ----------- X
TI
TTr

j' "X2 3
R --------- --------c- XxX -------------- X

J 12
R ---------- XXX X

SXR

In the categories Set, FGp, Grp, Ab, (RX,j ) may be taken to be

the set {xcX: there exists ycX such that (x,y)ER} together with the in-

clusion map. Similarly, in these same categories, (XR,j,) may be taken

to be the set {ycX: there exists xcX such that (x,y)eR} together with

the inclusion map and (XR-,j 3) may be taken to be the set

{xcX: there exists yCX such that (y,x)cR-1} together with the inclusion

map.

In the categories Top1, and Top2 the extremal subobjects (RX,j ),

(XR,j2), and (XR-,j 3) of X have precisely the same underlying sets as

above endowed with the subspace topology induced by the topology of X.

See Section 4 (4.1, 4.2, and 4.3) for a more detailed discussion.

It is easy to see that in the category Set, a symmetric, transi-

tive relation on a set X is an equivalence relation on a subset of X.

Recall the discussion in Section 2 (2.0) of the remarks of Lambek who

obtains the similar result for homomorphic relations on rings with

identity. This result we wish to generalize. In order to do thils we must

first be able to pick out the subobject.of X on which the relation is an

equivalence relation.

Referring to the above diagrams, since T is an isomorphism (1.9)

and sialce the eui-extremal mono factorizaticr is unique (0.18) it is

clear that (RX,j1) = (XR-1,j3). That is, there exists an isomorphism k

from RX to XR-1 such that j3k = jl (se,- 4.4 and 4.5).

Consider the product (RXxXR,l,' 2). Also, consider the morphism

(jlxj2)< r1'2> from R to XxX. Since

7i(jlxj2)<9lT2> JlP1 = JiTI = 2Tij and

^T2( 1xj2) = .222< 1,T2> = j2T2 = 72j it follows from the defini-

tion of product that (jlxj2) = j. Note, is an extremal

monomrorphism since j is an extremal monomorphisr (0.16).

Now suppose that (R,j) is symmetric on X. Then it follows that

there exists an isomorphism a so that ja = j* (1.13). Thus 72ja = 7T2J*

and this together with the fact that a is an isomorphism and the unique-

ness of the epi-extremal mono factorization implies that

(XR,j2) (XR-l,j3). That is, there exists an isomorphism B so that

J2 J3. Thus it is routine to see that the following diagram commutes.

T1

a XXX X
R 1-.- XR

2 12
SXxX --------------------- X

R 1 --------- R

R i

Consider the following products: (XRxXR,pI,p2) and (XxX,~1I ,2).

Letting 4 = (kkxlxR) then (2j2x2) = j, since

11(j2xj2)p = J2P1 = j2BklP = j2BkTl = J1lT = IrTj and

2(J2xj2) = J2o02 j21XRP2 = 2T2 = 2J. Thus the following

diagram commutes and the relation (R,i) on XR shall be called the

canonical embedding of R into XRxXR.

R *-RXxXR

Skx1XR

XRxXR

I j2xj2

XxX

3.5. Lemma. Let (R,j) be a symmetric relation on X. Then (R,i) is a

symmetric relation on XR.

Proof. Suppose that (R,j) is symmetric on X and let (R,4*) be the inverse

of (R.~) on XR. Then Y = i*T*. It is easy to verify that

p. Thus since (R-l,j*) is the intersection of

all extremal subobjects through which <72,r1>j factu-rs there exists a

morphimn X from R-1 to R so that (j2xj2)W*X = j*. But j*'T = <2,7T1.>j

whence (j2xj22)T* = (j2xj2) = <2,T>j = j*T. So

j*'T = (1(j2xj2) ) ((2xj2)) *)T*. Since (j2xj2)i* is a mDonomorphism

it follows that XT = T*. Recall that T and T*- both are isomorphisms (1.9).

Rence .\ is an isomorphism.

Recall that by the definition of symmetry (1.10), there exists a

morphism c from R-1 to R so that ja = j*. Thus

(j2xj2),"-1 = j 1 = j j= (j2xj2)~. But since (j2xj2) is a monomor-

phism this implies that *Xca- = Thus since X and a are isomorphisms,

we have (R,4*) E (R,4). Hence (R,i) is symmetric on XR.

3.6. Lemma. If (R,j) is a quasi-equivalence on X then (R,i) is a

quasi-equivalence on XR.

Proof. In view of Lemma 3.5 it need only be shown that (R,p) is transi-

tive on XR. To that end first consider the following diagram. It will

be shown that there exists a morphism X such that the diagram commutes.

Ox1XR
RxXR ------------ (XRxXR)xXR

xP 01 -
II6 "-
S(RxXR) (XRxR) XRXPRxXR -------- XRxXR

2 } x2
XRxR ,------ >XRx (XRXR) j 2xj2

X l J2xj2xj2

RxX :-- X -- (XXX) xx

SI
S(RxX) t, (Xx

XxR -- -------- Xx (XxX)
1xj

.--------..- XxX

Clearly <, 3>(j2xj2xj2) = (j2xj2)<1 ,P3>. Also,

01 (jxX) (1Rxj2)1 01 = j01 ((j2xj2) jxj2)X =1

O ((j2xj2)xj2)(x1XRlX) = (j2xJ2xj2)61(x1XR)Xl and

02(1Xxj)(j 2x1) )2 = 2(j2xj)X2 = 02(j2x(j2x, ) )2 =

O2(j2x(j2xj2))(IXRpx)X2 (j2xj2xj2)02(lXRX1)X2

as can be verified in a straightforward manner. Thus the diagram above

is commutative, and in particular,

ol(jx1x)(1Rxj2)l = (j2xj2xj2)' = 92(1Xxj)(j2x1R)x2. Hence, by the

definition of intersection there exists a unique morphism X such that

yX = (j2xj2xj2)6.

Let (RoR,i#) denote the composition of (R,p) with (R,u) on XR. Let

(RoR,j') be the composition of (R,j) with (R,j) on X. Then

6 !T# where T# is an epimorphism and <1l, 3>y = j'T' where r'

is an epimorphism. But since yX = (j2xj2xj2)6, it follows that

<'1f,3>yX = (2xj2)6 so that (j2xj2)0#T# = j'T'A. Hence the fol-

lowing diagram commutes.

(RxXR) C (XRxR) -> RoR

(RxX) n (XxR) XRxXR

T' j2Xj2

RoR -----------------'> XxX
1

Since T# is an epimorphism and j' is an extremal monomorphism, by

the diagonalizing property (0.19) there exists a unique morphism & such

that j'S = (j2xji2)# and CT# = T'X. But this says that

(RoR,(j2xj2)i#) < (RoR,j'). Since (R,j) is transitive (RoR,j') < (R,j)

hence (RoR,(j2xj 2)#) < (R,j) =: (R,(j2xj2)i). Hence there exists a mor-

phism o from RoR to R such that (j2xj2)io = (j2xj2)y1#. Again, j2xj2 is a

monomorphism so that ipo = p# which says that (RoR,4#I) < (R,y) hence

(R,y) is transitive.

3.7. Theorem. If (R,j) is a relation from X to Y and 7rlj is an epi-

morphism then (RoR-l,j#) is reflexive on X.

Proof. It will first be shown that the following diagram commutes.

RxX -------- ---> (XxY) xX
<1RKlj> 01

R -- -- > (RxX) r (XxR-l) 1 ;- XXYxX

< 2

XxR -------------------- Xx(YxX)

Consider the following products: (XxY,7Tl,r2), (XxYxX,T~1,7r2, 73),

(RxX,p1,o2), (XxR-l,pl*,p2*), ((XxY)xX,l*,'2*), (Xx(YxX), 1i 2.),

(YxX, 1, 12), and (XxX,-T1, 2).

Now,

1 (jxlX)= il1*(jxlx) r ljPl
2O1(jxl) = T27Tl*'(jx = 2" -

730 (jxl)1R, 1R ,3IJ> = T2*(jxlx)<1R,'1>J = ; 2< R iP> = Tr13

102(iXxj) = iI(Ixxj*)> = P1*Tj,T> = Tlj.

72 2(1XxjT)<1]jT> = (T 2(IXXj*) = ,IJ*j * =

^Tfj*' = 11j = r2J

B302(1Xxj*)<3)j.T> = 722(iXxj*)<' .,T> = 22j*T = 2<2,lI>j = 1J"

Thus by the definition of pr-'uct the diagram commutes. Hence

there exists a morphism E so that X1E = <1R,'Ij> and X2E = .

From the above it is easy to see that

1l< ,1T3>YE = TlY = 1Tlj = T = 3Y 2YE. Recall that (Ax,iX) is

the equalizer of T1 and T2 hence there exists a morphism such that

ix = <11l,73>YE.

Let (RoR ,j#) be the indicated composition of relations and let

T# denote that epimorphism for which j#T#/ = y. Thus, combining

the above results, <7Tij,>Tj> = YE -= j#T#E = iX.

Since (Ax,ix) and (X,) are isomorphic as extremal subobjects

of XxX (1.21), there exists an isomorphism X such that <1,1x>X = iX.

Consequently, Xt = iX = Y = .

Now 7ITX = l XX = X} = Xrl<7T J,7 rj> = 7rlj and by hypothesis

Trlj is an epimorphism; thus, since is an isomorphism,it follows that

( must be an epimorphism.

Thus has ((,iX) as its epi-extremal mono factcrization.

But this means that (Ax,ix) is the intersection of all extremal subob-

j.cts of XxX through which factors (0.21). Recall that

= j#t#/E, thus (A ,iv) < (RoR- ,ji#) which was to be proved.

3.8. Corollary. If (R,j) is symmetric on X then (RoR,ipV), the composi-

tion of (R,i) -,ith (R,.) on XR, is reflexive on XR.

Proof. Since (R,j) is symmetric on X then (Ri) is symmetric on XR (3.5)

hence (O,,) 1 (R-1. :*) (1.13). Referring to the diagram in (3.41) fol-

lowing the c0 'inition of the canonical embedding it is immediate that

pl is an e; -rphism since plI = Bkt- and each of B, k, and T1 is an
-1 ,.
epimorphisr. Thus (RoR ,1) (RoR,/#) is reflexive on XR (3.7).

3.9. Corollary. If (R,j) is a quasi-equivalence on X then (R,j) is an

equivalence relation if and only if lrlj is an epimorphism (respectively

if and only if r2j is an epimorphism).

Proof. If (R,j) is an equivalence relation then (R,j) is reflexive and a

quasi-equivalence. Thus by Proposition 1.24, 7rj and T2j are retractions

hence epimorphisms.

Conversely, if 1lj is an epimorphism then applying the theorem
-1
and Proposition 1.30, (A ,iX) < (RoR ,j#) < (RoR,j') < (R,j) so that

(R,j) is reflexive and hence is an equivalence relation. (If T2j is an
-1 -1 -1 -1
epimorphism then (Ax,i ) < (R oR,j#*) < (R oR ,j'*) < (R ,j*) and
-1
(R ,j*) E (R,j).)

3.10. Corollary. If (R,j) is-a quasi-equivalence on X then (Rd:) is an

equivalence relation on XR.

Proof. (R,i) is a quasi-equivalence on XR (3.6) and (RoR,ijI#) is reflexive

on XR (3.8). Thus (AXR, XR) < (RoR,y#) < (R,w) whence (R,',) is reflexive

and thus is an equivalence relation on XR.

3.11. Proposition. If (R,j) is a quasi-equivalence on X then (R,j) and

(RoR,j') are isomorphic relations on X.

Proof. By Corollary 3.10 (R,i) is an equivalence relation on XR whence

(RoR,ir#) and (R,,) are isomorphic relations on XR (1.32). Recall that

there exists a mcrphism E such that the following diagram commutes (3.6).

(RxXR)ri (XRxR) --------------;> RoR

T'X ( 2Xj 2)

RoR -;- XxX
j'

Thus (RoR,(j2xj2)i#) < (RoR,j'). But as mentioned above

(RoR,i#) (R,t) hence there exists an isomorphism A# such that

I^,#jt = So by the definition of the canonical embedding (3.4),

j'gX# = (j2xj2) #lx = (j2xj2)y = j. But this implies that
(R,j) < (RoR,j'). Thus since (R,j) is transitive, (R,j) (RoR,j')

which was to be proved.

3.12. Proposition. Let (R,j) be a relation on X. Then (R,j) < (AX,iX)

if and only if R is symmetric on X and (R,i) < (AXR,iXP).

Proof. If (R,j) < (AX,iX) then there exists a morphism a such that

j = iXa. Thus iTl = rliXa = T2ixa = 72j whence

n <,f2,Tl>j = T2j = IrFj = Tr2<'r2,Ttl>j. Thus by the definition of product

j = j. Consequently the epi-extremal mono factorization of

j is (1R,j) and so (R,j) (R-1,j*); i.e., (R,j) is symmetric.

Recall that j = (j2xj2)d (3.4). Thus

lj FTl(j2Xj2)' = j2p14 and 72j = 2(j2xj2)) = j2P2P'. But "lj = '2j
hence j2P = j2P2'. Since j2 is a monomorphism it follows that

pl -= p21,'. Recall that (AXR,iXR) is the equalizer of pI and p2. Hence
there exists a morphism; B such that iXR5 = 4j'. This implies that

(R,,) < (AXRiXR))

Conversely, if R is symmetric and (R,') < ('XR,iXR) then there

exists a morphism 8 such that d = ixRP; hence

P1 = P liXR = P2iXRB = P2. Since (j2xj2) = j, we have

1 j r "Q(j2xj2) = j 2P1 = J2P2 7T 2'(j 2xj 2) = F2j. Thus r1j = Tr2j

so that there exists a morphism a such that j = iX&. This means that

(R,j) <. (Ax, ix).

3.13. Definition. Let (R,j) be a relation on X. Then R is said to be

a circular relation if and only if RoR < R-1

This notion is due to MacLane and Birkhoff  exercisee 3,

page 14).

3.14. Proposition. Let (R,j) be a relation on X. Then R is a circular

relation if and only if R1 is a circular relation.

Proof. If R is circular then RoR < R-1. Thus

R-loR-1 = (RoR)-1 < (R-1)-1 R (1.38, 1.;2 and 1.11). Hence R-1 is

circular.

Conversely, if R1 is circular then by the above, (R-1)'1 =R is

circular.

3.15. Theorem. Let (R,j) be a relation on X. Then R is an equivalence

relation on X if and only if R is reflexive and circular.

Proof. If R is an equivalence relation then R is reflexive. Since R is

transitive and symmetric, RoRR R-' hence R is circular.

Conversely, if R is reflexive and circular then R-1 is reflexive

(1.17) and R-1 is circular (3.14). Hence

R R'1oA.. < '-loR1 < R (1.31 and 1,12) whence R is symmetric. Thus

R R-1 (1.11).

Now RoR R-1oR-1 < R hence R is transitive. Thus R is an equiva-

lence relation.

SECTION 4. IMAGES

4.1. Definition. Let (R,j) be a relation from X to Y and let (A,a) and

(B,b) be extremal subobjects of X and Y respectively. Consider

(Rir(AxY),y) and (Rfi(XxB),6). Let (T ,a) and (? .2) be the epi-extremil
1 2
mono factorizations of i y and r 6 respectively. Denote the domain of a
2 1
by AR and the domain of B by RB. Thus the following diagrams commute.

R

^ Y 2

axly
Rt(Axy) ----> Xxy c Y

2 AXY ARIrw,

R

R 1 '(XxB) >-.------ -- XxY -----> X

XWX ^ RB

4.2. Remark. Since (X,1 ) and (Y,I ) are extremal subobjects of X and Y
X Y
respectively, then (R '(XxY),y) (R,j) and (RA (XxY),6) 5 (R, j) whence

(XR,&) is precisely the extremal subobject (XR.,j) used in the canonical

nQboddiin (3.-'). Since X = Y in 3.4 then also (RY,3) is precisely

(RX,j ) used inl 3.4.

4.3. Example. In the category Set, for (A,a) < (X,1 ), (B,b) < (Y,1 )

and (R,j) < (XxY,1 ),
XxY
AR j{yY: tL exists acA such that (a,y)ER}

RB {xeX: there exists bcB such that (x,b)eR}.

This is easily seen since R(\(AxY) L {(a,y): acA, (a,y)eR} and

R(I(XxB) {(x,b): beB, (x,b)eR}, and AR is the set of all second terms

of elements of R((AxY) and RB is the set of all first terms of elements

of Rf(XxB).

In the category Top AR and RB have precisely the same underlying

sets as above. They are endowed with the subspace topology determined by

the topology of XxY.

In the category Top2, AR and RB have" precisely the same underlying

sets as in Top for it is easy to verify that AR and RB are closed sub-

sets of X and Y respectively. Recall that the image of a morphism in Top

is the closure of the set theoretic image (0.15).

4.4. Theorem. If (R,j) is a relation from X to Y and (A,a) is an extremal

subobject of X then (AR,a) and (R-1A,B) are isonorphic extremal subobjects

of Y.

Proof. Consider the following commutative diagrams.

R

A

RAxY) -------------- XxY --- -------------Y

AxY _axl,,

R1

6 iT
R- 1 (YxA) ------- -- YxX -'--Y--

1 xa

Y
YxA RA

j <7f iT>
j R ------- Xx ------ YxX

R-1

It can be shown in a straightforward manner that

(ax] ) = (1 xa)

2 1 Y Y 21
where pl and p2 are the projections of AxY. Hence

(1 xa)

= (axl )X = y = j1 = j* .
Y 12 1 2 2 1 Y2 2 1 2 1 1 1
Thus by the definition of intersection there exists a morphism 5 such

that SS = <2 ,Tr >Y = j*TX (1 xa)

X Hence
21 1 Y 2'1 2
f 6 = f Y = y = ai But fi 6 = Bi C. Thus, since (AR,a) is
1 1 21 2 1 1 2
the intersection of all extremal subobjects through which T.2y factors

(0.21), it follows that (AR,o) < (R-1A,B).

Similarly, it follows that

< >2'1 *X = -> 16 = j (axl,])2,-X whence -h4rc
<2' 1 2 1 3 (ax] )

exists a moiphisnm such that y'* = -16. Then

2 < ,T7 >-16 = f 6 = ST2 = y* = aT ;*. Again, since (R1A, 8) is the
ntrscton f all exremal subobjecs 2
intersection of all extremal suhobjects through which h ,1 6 factors,
1

(R-IA,B) < (AR,a). Consequently (R-IA,B) E (AR,a).

4.5. Corollary. If (R,j) is a relation from X to Y and (B,b) is an extre-

mal subobject of Y then (RB,B) and (BR-1,a) are isomorphic as extremal

subobjects of X.

Proof. Recall ((R-1)-1,j#) E (R,j) (1.11). Letting (R-l,j*) play the role

of (R,j) and (E,b) the role of (A,a) in the theorem, the following is

obtained: (BR-1,Q) ((R-)-1B,B/#) E (RB,B).

4.6. Corollary. If (R,j) is a symmetric relation on X and (A,a) is an

extremal subobject of X then (AR,a) and (RA,B) are isomorphic as extremal

subobjects of X. (In particular, (XR,j ) and (RX,j ) are isomorphic as

extremal subobjects of X as was shown directly in 3.4.)

Proof. Recall that (R-1,j*) E (R,j) (1.13). Hence by the theorem

(AR,a) (R-1A,B) = (RA, ).

4.7. Proposition. Let (Al,al) and (A2,a2) be extremal subobjects of X and

(R,j) be a relation from X to Y. If (Al,al) < (A2,a2) then

(A1R,al) < (A2R,a2).

Proof. By hypothesis there exists a morphism v so that a2P = a1. Thus,

there exists a morphism ( such that the following diagram commutes.

R('(A1xY) Y xy
A AlXYA axl xy

1 -Y 2
R (A,2xy) ------. .. .-- ..-X XxY

Thus w2Y1 = i2Y25 whence, because (A1R,al) is the intersection of

all extremal subobjects through which T2rY1 factors and r2y2C factors

through (A2R,a2), (AIR,al) < (A2R,a2) which was to be proved.

4.8. Proposition. Let (Bl,bl) and (B2,b2) be extremal subobjects of Y and

(R,j) be a relation from X to Y. If (Bl,bl) < (B2,b2) then

(RBi,B1) < (RB2,B2).

Proof. (RB1 ) 1*)E (B1R(-1,12) < (B212) (RB22) (4.5 and 4.7).

4.9. Proposition. Let (R,j) and (S,k) be relations from X to Y and (A,a)

be an extremal subobject of X. If (R,j) < (S,k) then (AR,a) < (AS,a).

Proof. In a manner similar to that in the proof of 4.7 one can establish

the existence of a morphism \$ such that the following diagram commutes.

R ---------- S

j t k

RTA(AxY) S A(axy) --- Xxy

axi

Axy

Hence the following diagram commutes.

R /(Axy)

S 1 (AxY) ---------- XxY

Thus, since (AR,a) is the intersection of all extremal subobjects

through which yT, factors, and 2 factors through (AS,a), it follows that

(AR,cL) < (AS,5) which was to be proved.

4.10. Proposition. Let (R,j) and (S,k) be relations from X to Y and (B,b)

be an extremal subobject of Y. If (R,j) < (Sk) then

(RB,B) < (SBj).
Proof. (RB,8) (BR-1,B*) < (BS-1,~*) E (SB,8) (4.5, 1.12, and 4.9).

4.11. Proposition. Let (R,j) be a relation from X to Y and let (Al,al)

and (A2,a2) be extremal subobjects of X. Then

((AlfA^R:A2 ) <_ (AIRAA2R,5).

Proof. Since (AljnA2,a) < (A1,al) and (A1CfA2,a) < (A2,a2) it follows

that ((A,! A2)R,a) < (A1R,a1) and ((A(fA2)R,a) < (A2R,a2) (4.7). Thus

((AiCA2)R,a) < (A1RflA2R,Q) (1.19).

4.12. Proposition. Let (R,j) be a relation from X to Y and let (B.,bl)

and (B2,b2) be extremal subobjects of Y. Then

(P.(B1^ B2),B) < (RB1 nRB2,B).

Proof. (R(B nB2),B) ((B B2 )R-1,B*) < (BiR-lnB R-1,*)

(RB1nRB2,B) (4.5 and 4.11).

4.13. Proposition. Let (RI,jl) and (R2,32) be relations from X to Y and

let (A,a) be an extremal subobject of X. Then

(A(R1AR2),a) < (AR,( AR2, ).

Proof. It is clear that there exist morphisms 1, and C2 such that the

following diagram commutes.

R1 N(AxY) "Y

(r.,fl R2) (AxY) )x ----> xxY

'R2-A (Axy)

69

Thus rr2y = 1r2Yl1 = T2Y2 2. Again since (A(R1(\R2),a) is the inter-

section of all extreral subobjects through which ir2y factors it follows

that (A(R(.R2) ,) < (AR1,51) and (A(R1 FR2),a) (AR2,a2). Hence

(A(RI AR2),a) < (AR1 AAR2, ) (1.19).

4.14. Proposition. Let (R1,jl) and (R2,j2) be relations from X to Y and

let (B,b) be an extremal subobject of Y. Then

((R AR )B, ) < (RB 2R B,B).
% 1 2 1 2
Proof. ((RflR2)B,6) E (B(RI.R2))-1, *) < (BR3-~I BR-1 ,1) =
1 2 1 2 i 2
(R1 BR 2B,6) (4.5 and 4.13).

4.15. Proposition. Let (R,j) be a relation from X to Y then (R,j) and

(R1 (RYxY),y) are isomorphic as extremal subobjects of Xxy.

Proof. Consider the following commutative diagrams.

R 1) > Xxy
RY ------ XxY

T 1 T 1

RY ---

ji

RYxY ,

- R fA(RY7Y) :>--------------------------->^ XxY

R

Since (jlxl ) = = j, there exists a morphism E

such that yv =- j
R

RYxy
.^^.. yxY YXiy

R n---( R ((RYxY) XXY

P, R

Thus (R,j) < (R n(RYxy),y). Clearly the reverse inequality holds

so that (R,j) E (R(T(RYxY),y).

4.16. Proposition. Let (R,j) be a relation from X to Y. Then (R,j) and

(Rr%(XxXR),6) are isomorphic relations from X to Y.

Proof. Analogous to the proof of 4.15.

4.17. Corollary. Let (R,j) be a relation from X to Y. Then (R,j) and

(Rt'(RY^-XR),6) are isomorphic relations from X to Y.

Proof. (R.j) : (RO(RYxY),v) E (R(l(XxXR),'6) (4.15 and 4.16). But since

(RY,jl) and (XR,j2) are extremal subobjects of X and Y respectively it

folJows that ((RYxY)A(Xx'R),a) E (RYxXR,). Thus

(R,j) E ((Rr;(RYxY))n(Rn(X\-XR)),B) E (RA(RYxXR),).

4.18. Proposition. Let (R,j) be a relation from X to Y and let (A,a) be

an extremal subobject of X. Then (AR,a) ((RYAlA)R,&).

Proof. It follows from Proposition 1.5 that RI(((RYAA)xY) and

RPf((RYxY)A(AxY)) are isomorphic relations from X to Y. By Proposition

4.15. (R,j) and (Ro(RYxY,y) are isomorphic relations from X to Y. Thus

RI~(AxY) and R.i((RYAA)>:Y) are isomorphic relations from X to Y. Conse-

quencly by the definition of image (4.1), (AR,a) and ((RYftA)R,&) are

isomorphic as extremal subobjects of Y.

4.19. Corollary. Let (R,j) be a relation from X to Y. Then ((RY)R,a) and

(XR,j,) are isomorphic ns extremal subobjects of Y.

Proof. Let (X,1,) play the role of (A,a) in 4.18.

4.20. Corollary. Let (R,j) be a relation from X to Y and let (B,b) be an

extremal subobject of Y. Then BR-1 and (Br.XR)R-1 are isomorphic as ex-

tremal subobjects of X.

Proof. Immediate.

4.21. Proposition. Let (R,j) be a relation from X to Y. Then (RoR-1,j#)

and (RoR-71,(RYxX),y) are isomorphic relations on X.

Proof. Consider the following diagram.

RYxX

CTix1

xljxl
X1 01
y
(RxX) ( (XxR-1) ---- ---.-, XxyxX ---> XxX

02

XxR-1 XX(YXX)
T xj*

RoR-1

To see the diagram is commutative it need only be observed that

(jxl)(Qrxl) = < lT3>01(jY1). To show this note that

(jxl1)(Qlxl) = (jl'-xl) = (Tljxl) and

aT Cl(jxl) = 1 1(jx) !jpl Ir r(jxl),

7r0 1(jxl) = 3C1(jxl) = p2 = T2(rrjxl).

Thus, since (RoR"1 ,j#) is the intersection of all extremal sub-

objects through which <'7"l,3>y factors,it follows that

(RoR-l,j#) < (RYxX,j xl). Whence (RoR-],j#) < (RoR-1fI (RYxX),?).

4.22. Theorem. Let (R,j) be a relation from X to Y. Then (RY,jl) and

((RoR-)X,B) arc isomorphic as extremal subobjects of X.

Proof. Consider the following products: (XxYxX, 1,2, 3),

(Xx(YxX),p1,'2), ((XxY)xX,Pl *,P*), (RxX,PlP2), and (XxR-1, 1, 2).

Referring to the diagram in the proof of 4.21 it is easy to see

that: irlY = 1( = (jxl)Aj = il*(xl)X = r1jp1X1. Thus

<1 'l~' 3 = "> jll = jlT 1 1l

RoR

(RoR- )x >>

o;'XxX

1

Since <7- ,I3>' Y = j#i//' and rTlj# 5, the following diagram

cc;Lmu:es.

-Cl p

iT#
(R (RoR-1)X

XI I

RY .--

But since has the diagonal property (0.19) and Tr# is an epi-

mcrphism and j is an extremal monomorphism then there exists a morphism

: such that ji = B and pl1?1 = T,. Thus ((RoR-1)X,B) < (RY,j).

Next it will be shown that the following diagram is commutative.

. .----.----- ----

jxl
RxX ---------- (XxY)xX

<1R'i l

R --- --X)R (RxX)l(XXR-1) .- -- xxyxX

2 02

XxR-1 Xx(YxX)
ixj*

= lP1*(jx1)<1Rlj> = UiP1
2P1 (jx 1)1 lj> = 2j 1~1R j> = 2j.
= 2P ':(jxl) = T2JPl = 2""

= P2*(jx)<1Rlj> = P2 = !j.

= \ 2(!xj*)< ,lj, > = IT J"2 = --j =j

= T2p2(]xj*) = 7T2j*2 = I2j*T =

T2< 2',1i j = 1j .

Consequently there exists a morphism * such that the above diagram

commutes and such that yS* = .

Thus <';1,lr,>Y\$: = and hence the following diagram is

commutative.

(RxX),(R (KRx)--- -------- XxyX

.T I F l1

.______ (' ' X
S(RR 1)X R_----- --. X
R ----i;"R -~----'--"" '

1

2e01(jxl)<1Rnlj>

30 ] (jxl)

"1 2(lxj*)< l 3," >

,2 '02xj*)

i3O2 (lxj-)

74

Since (RY,j1) is the intersection of all extremal subobjects

through which irj factors, it follows' that (RY,j1) ((RoR-1)X,B). Thus

(RY,j1) ((RoR-1)X,B) which was to be proved.

4.23. Corollary. Let (R,j) be a relation from X to Y. Then (XR,j2) and

((R-loR)Y,8) are isomorphic as extremal subobjects of Y.

Proof. (XR,j2) (R-lX,j3) E (((R-1)o(R-1)-1)Y, ) = ((R-loR)Y,P)

(4.4, 1.11 and 4.22).

SECTION 5. UNIONS

5.1. Definition. If {(Ri,ji): icI} is a family of relations from X to Y

then let ((FRi,j) be the intersection of all relations (i.e., extremal
icI
subobjects of XxY) "containing" each (Ri,ji) (where containment is in the

sense of "factoring through" as noted in Remark 0.12). (' Ri,j) shall
ieI
be called the relation theoretic union of the family {(Ri,ji): iIl}.

5.2. Examples. In the category Set the relation theoretic union is the

usual set theoretic union together with the inclusion map.

In the category Top the relation theoretic union is the usual set

theoretic union endowed with the subspace topology determined by the top-

ology of XxY together with the inclusion map.

In the category Top2 the relation theoretic union is the closure

of the set theoretic union together with the inclusion map.

In the categories Grp and Ab the relation theoretic union is the

subgroup generated by the set theoretic union of the relations.

5.3. Proposition. Let {(Ri:ji): isl} te a family of relations from X to

Y, let (LRi,k) denote the usual categorical union of subobjects, let
iel
(o.j) be the epi-extremal mono factorization of k and let the codomain

of a (domain of j) be denoted R. Then r and ( JRi are isomorphic rela--
iel
tions from X to Y.

Proof. Since (~Ri,j) is the intersection of all extremal subobjects
igI
containing each (R.iji) and each (Rj,ji) < ( iRi,k) and
iel
(CjRlk) (Rj) and since j is an extiemal monomorphism then
icI

(UtRi,j) < (R,j).
icd

Since (U R ,k) is the intersection of all subobjects which "con-
isI
tain" each (Ri,ji) then (UR,k) < (\jR ,j). Since j is an extremal
isl ieI
monomorphism and (R,j) is the intersection of all extremal subobjects

which "contain" (U R,k) then (R,j) < (Q)R ,j). Thus
ieI iEI
(R,j) E (U-R ,j)..
ieI

5.4. Remark. Notice that by the definition of relation theoretic union,

if (R ,j1), (R2,j2), and (S,k) are relations from X to Y and if

(R!,Jl) (S,k) and (R2,j) < (S,k), then (P R2j) < (S,k) (cf. 1.19).

.5.5 Proposition. Let (R ,j ), (R ,j ), (Sk ) and (S ,k ) be relations
= 1 1 2' 2 1 1 2 2
from X to Y. If (R ,ji) < (R2 j2) and (Slkl) < (S2,k2) then

(R 1- S1,) < (R2JS2k).

Proof. (R ,j ) (Rj ) < (R2' S,k) and (S,k ) < (S2,k) < (R2 S2,k)
_of I ] 2,<2 -2 1' 1 < 2' 2) 2 2

whence (r ..s ,j) (R2 L S2 k) (5.4).

5.6. .ma.'.rk. The following proposition can be strengthened with the ad--

ditional hypothesis that the category has finite coproducts (5.34);

however. J i s included here because it is of interest in its own right.

5.7. Proposition. Let (R,j) be a relation from X to Y and let (S,k) and

(T,m) be relations from Y to Z. Then ((RoS) U (RoT),B) (Ro(S')T), ).

Proof. Consider the following commutative diagrams.

s ----__ k

^s '^^ 7; '~^"~----------
S

S T -----
S- ''"-"."-. Z
rPp

RxZ ---------------> (XxY)xZ
S7A jxZz
x 3 z

In ,\ - ~

(RxZ)[1Xx(SQ}T) >>

Xxyxz

-0

'4 \
Xx(S ) T) ----------- Xx(YxZ)

1XXa

RxZ )'>- (XxY)xZ

jxj1

x5

RxZ f XxS

A6

XxS *I -'- V XX(YxZ)
lxxk

RxZ >, ---- -(Xxy)xZ
7" jx1Z 31-
Y3

R'Z rlXxT

02

XxT >- Xx(YxZ)
1Xxm

Yl <1, 7T3>
RxZ rXx (S T) )---- --- --- XxyxZ -------- XxZ

I Ro(S T)

Y2

T2 RoS '' -

1401
>- IXyxYZ

r'-02

XxyxZ

r.

41-------------------- -----------------

~-------------------- ----------,

--

-.

R.xZfIXxT ->--- XxyxZ XxZ
Y3 < 1'f 3 3>

SRoT '

RoS _

(RoS) ) (RoT) ------ XxZ
X
RoT B3

Since (S,k) < (S( UT,a) and (T,m) < (S kT,a) it readily fol-

lows that ((RxZ)I}(XxS),y2) < ((RxZ)A(Xx(S J T)),y1) and that

((RxZ) n(XxT),Y3) < ((RxZ) n(Xx(S (QT)),yl). Thus there exist imorphis.ns

.! ard 2 such that Yl1, = Y2 and Y12 = Y3. Hence

<~3,'3>' 1 iI Y2 and YiS2 = < ,3>Y3

But (RoS,e2) is the intersection of all extremal subobjects through

which y factors and since Y2 = Y1S1 = 6B'

wv have (RoS,B2) < (Ro(SUT T),B ). And since (RoT, 3) is the intersect-

ion of all extremal subobjects through which <-1, 3>Y3 factors and since

<'.'3 3- 13Y = 122 it follows that

(RoT, 3) < (Ro(S(V T),BI).

Whence ((RoS) k (RoT),P,) < (Ro(SLj T), p) (5.4).

5.8. Proposition. Let (T,m) be a relation from Y to Z and let (R,j) and

(S,k) be relations from X to Y. Then

((RoT)r.a(SoT),R) < ((R.orJS)oT,5).

Prciof. Anaclogour: to the proof of 5.7.

5.9. Lemma. Let {(Ri,Ji): iEl} be a family of relations from X to Y. Then

(*Ri)-"l,j*) (Ul (i)-1,k).
iel iel

Proof. Consider the following commutative diagram.

is]

j <2.2, > 1
Ri ------------ Y >----- -------> YxX

I ( R )-1 /

R. k

R -1 (Ri)-1
X.* iel

Since (Ril,ji.*) is the intersection of all extremal subobjects

through which Ji factors it follows that

(R -1,j*) < ((iRi)-1,j).
i i EI
ieI

Thus

Now (Ri,ji)

"l-i =i

(-(Ri -1)k) < ((()Ri)-1,j*) (5.4).
iel iel

< ()(Ri-1),<2,l>'-1k) since
ieI
<2' 1ji*T = i.. Thus

((ljRi),j) < ((1.(Ri-1), ,>-k)
iel iel
from which it follows that

((JRi)-,j*) < (r J(Ri' 1),<2,l><2,7 k)
iel icI

whence

(((2Ri )-1 *) = (f (R-) ).
iEl i il

5.10. Corollary. Let {(Ri,ji). iIl} be a family of symmetric relations

on X. Then ( URi is a symmetric relation on X.
icI
Proof. It is clear that for each icl, (Ri,ji) E (Ri-1,ji*) (1.13). Thus

(U Ri,j) =- ( (Ri-1),k) < (( Ri)-1,j*) (5.9).
icI iEI iTl

5.11. Proposition. If (R,j) is a reflexive relation on X and (S,k) is

any relation on X, then (R(t)S,m) is reflexive on X.

Proof. Since (R,j) is reflexive, (Ax,i ) < (R,j). Thus

(AX,iX) < (R,j) < (R:JS,m) hence (R(IS,m) is reflexive on X.

5.12. Definition. Let (R,j) be a relation on X. Consider the relation

(R-U R-I,j#).

R *

Let (T.i,X) be the epi-extremal mono factorization of Tlj/#. The

domain of X (codomain of T#) shall be denoted by XR.

XE R --------------- X
X

According to the notation of Section 4, XR is also denoted by

(RR'XR-1)X.

5.13. Examples. In the category Set, XR 1 (R(R-1)X = XRU RX.

That is, XR = {xeX: there exists yeX such that (x,y)ER or (y,x)ER}.

In the category Top1, XR is the same set as in Set endowed with

the subspace topology determined by the topology of X.

5.14. Proposition. Let (R,j) be a relation on X and let (RX,jl) and

(R-1X,j3) be the images of Rlj and Tlj* respectively. Then

(RX~ R-1X,a) < (XR'X) = ((RlU-R-1)X,X).

Proof. Consider the following commutative diagram.

R RX

jR j T
R (R-1 --------- XxX ------------- X

R

Since 1,j = Tlj#XR = XTI#R = JiT1 and (RX,j1) is the intersection

of all extremal subobjects through which Tlj factors then

(PX,jl) (XR,X). Similarly, it can he shown that (R-1X,j3) < (XR,X),

whence (RXi'R-1X,a) < (XR,X) (5.4).

5.15. Proposition. If (R,j) is a relation on X then (R R- ,j/#) is sym-

metric on X and (XR,X) (X(R )R-1),j2).

Proof. ((R-R'-])-1 ,j*) (Rjql)(R-l)-',j) = (R-i1 R,j#) (5.9 and

1.11). Thus (RLiR-1,j#) is symmetric so that

(XpX) = ((R)R ')x,x) E (X(R -1),j2) (4.6).

5.16. Proposition. Let (Rj) be a relation on X and let (Ax iR ) be the

diagonal of YjxXR. Then (Ax R (XXX)iX)- (AX n (XRxXR),p) where

(A iR(X R
(X X .XX)*
R R R XX

Proof. Consider the following commutative diagram.

XR 7T*
A ----------- XxX ---- X
R T2 2

I A X xx \X

I t i x xX xX

SR R R
bx' ~-------- X'xx "'---------G
iX 2

SinceR =l "," th=t =" Thus,
Observe that E1(XxX)i 2 XX)I.X Xh2 *iX r2(XXX)iXR. Thus,

since iX is the equalizer of 71 and 2, there exists a unique morphism

so that iyX = (XXX)"iX
XR
Thus, since (AX SRXRxXR,p) is the intersection of (Ax,iy) and

(XRnXR,)XX), there exists a morphism B so that pB = ix = (XXX)iX Con-

RR R
sequently (A;R(X ^x) -- < (A(x (XRXRp) ,p)"

Since irlX = Tr2iX it follows that -orp = TliX" = r2iXa i 2p, but

(XxX)A = p so that 71(XXX)X = TT2(XxX)X. IhNence XTI*X = Xv2*X. Recall

that X is an extremal monomorphism, hence a mnonomorphism; so that it

follows that irl*X = R22X. Since (A ) is the equalizer of i* and

fo* there exists a morphism a such that ix o. = A.

Thus (XXX)iX a = (XxX). = p, which means that

(AXP(XRXXp)IpP) < (Ay ,(XxX)j iv). Ihencf
R
(Ax ('(X

5.17. Definition. Let (R,j) b2 a relation on X. Then (R,j) is called

quasi-reflexive if and only if (A ,(X X X
R ^R
quasi-reflexive provided that there exists a morphism X such that the

following diagram commutes.

IXR Tr
A >- -- X xX X
XR R R 2" R

R XxX ,,X
Ji 2

5.18. Proposition. If (R,j) is a reflexive relation on X, then (R,j) is

quasi-reflexive on X.

Proof. If (R,j) is reflexive then (RLU)-R,j#) is reflexive (5.11);

hence, 7,J# is a retraction (1.24). Thus rTlj# is an epimorphism; so that

if (T~,X) is the epi-extremal mono factorization of Tlj#, rlj' = XT# so

that X is an epimorphism as well as an extremal monomorphism. Hence X is

an isomorphism (0.17). Thus (XRx) (X,1X) whence (Ax ,iXR) (A,i ).

Thus, since (Ax,iX) < (R.j), (R,j) is quasi-reflexive.

5.19. Proposition. Let (R,j) be a relation on X. Then (R,j) < (A ,i ) if

and only if (R,j) < (A, ,(XxX)i ).
R R
Proof. Suppose that (R,j) < (AX,i ). Then there exists a morphism a such

that j = iXa. Thus iJrj = rliXa = v2ixc = r2j; whence

iJ = I2J = Tij = 2j Consequently the unique epi-extre-

mal mono factorization of j is (1 ,j), and (R,j) (R-j*)

(3.12).

But, since (R,j) (R-!,j*), then (R()R-i,ji/) (R,j), and

(X,,X) ((R(~jR-)X.x) (RXj ). Thus, since (R,jl) E (X ,x) = (XR,j)
R L
ft follows that (R,j) < (X. x >xx) (3.4).
R I R

However, it has been shown that (AXR,(xxx)iX ) and (A,((XRxXP),p)

are isomorphic relations on X (5.16). So that since (R,j) < (A ,i ) and

(R,j) < ((XRxxR),XXX) it follows that

(R,j) (Ax((XRxXR),p) E (AXR (XXX)iX ) (1.19).

Conversely, if (R,j)< (A ,(XxX)iX ) then since
R R
'IT(XXX)ixR = X'*iXR = X12 *ix = 2(X)iXR it follows that
R R R R
(AXP'(xXX)ix ) < (Ax ,i). Consequently the result (R,j) < (Axi ) fol-

lows from the transitivity of < (0.2).

5.20. Theorem. If (R,j) is a relation on X then R, RoA and A oR are
R R
isomorphic relations on X.

Proof. It will be shown that RoA and R are isomorphic relations on X;
X
the proof for A oR and R is analogous and is omitted.
rR
The following products shall be considered: (XRxYE r"I*,2*),

(XxXxX,rP ,,2,3), (XxX,7T,l12), ((XxX)xX,i ,f 2), (Xx(XxX), i1 J ),

(RxX,5p,,2) and (XxAXR 1* 2*).

Consider the following commutative diagram.

jx1X
RxX ------- (Xx) xX

r. Y I 01
Y 4Z < 1 2,T3>
RxX XxA., ------ --- XxXxX ------> XxX
RR)
X2 09
i Xx(xxx)ixR
SXXA ^------------ P'>Xx(X>) -
XR

Next it will be shown that < T,,i'2> = <' ,'i>y.

T < ,T, 2>Y = i1Y = TJ<~I1,7T3>y.

T2<7T2>y = 2Y = 202(1XX(XXX)iXR)2 2(1XX(XXX)iX )X =

T1 (XxX)jxP2*X = X"liXR P2*-2 = X2*XRP*2 =
1 XR 2 2 X R 2 2 R
X7T *iXR 2* 2 = 2(XxX)iXR P2-2 N2 2 (Xx(XxX) iXR 2

3 32(1X(XXX)iXR)A2 = 73Y = T2<7,73>Y-

Thus by the definition of product <7 1,)N3>y = <71,T2>y.

Now consider <71,'r2>y. It will be shown that Y = jl1l-

T<71T,,2>Y = :FlY = T l01(jx1)A1. Thus Tly lTlm1 (jXl)1 = lJl

T 2<1, 2>Y = I2Y = "201(jxl)1= )2~ (jx1)X1 = 7j2jP11. Hence, by the

definition of product, y = jXl1l.

Since (RoAxR,a) is the intersection of all extremal subobjects

through which <-1, 3>y factors and 3>Y = <1-l,T2>Y = J1Pil it

follows that (RoAX,a) <_ (R,j).

For the reverse inequality consider (RQJR- ,j#). Since

(RLJR-,ji'/ ) is syrmmietric, ((R:)]R-I)X,ji#) and (X(RC R-1),j2#) are

isomorphic extremal subobjects of X (5.15).

x! x
R ^/R-I y------------->- xX ---- --- X

R 1 J '-^< ~- 1 X
x(RIIR 1) '

RR
-XR

Thus if (G#,X#) is the epi-extremeal mono facrorization of T2j#,.

there exists an isomorphism ( such that g, E/#. So in particular,

S7Tj = Q1 #x =- XR #AR"

Let a be the isomorphism for which iXRO = <1XR ,XR> (1.21).

It next will be shown that the following equality holds:

<7Ji2J,2J> = 01 = 02(lx(XX)ix )

so that the following diagram is commutative.

01

jx1X
A------^ -----

RxX

(XxX)xX

--------- XxXxXX

i i
1

1 -,
RxX % XxAXR

I X
AX 2
XxA .---
XR

51 2(1X(XXX)iXR )<1rj ,a R>

7202 (lX(XX)iXR) <7tlJ,oG~ XR>

0,2(IXx(XXX)iXR )<1j,Gc AR >

SXx(XX)
-4. Xx (XxX)

1XX(XXx)iXR

= i1(lXX(XXX)iR ) =

P1* = 7lJ.

= r17 2(1Xx(Xxx)iR ) -

T1(XXX)XR P2 <7t j,oX#R> =
R
XPi XRP2* = XPliR oG/#XR

XR R#R
X011#XR = X1#XR >= T2j.

7 2'2(lxX(XXX)ixR) =

T2(XXX)iXRP2*<1 P Iil R> =

XP2iiX P2,* j,oS#XR> = XP2< 1XR >S R =

X1 XR#X = XW#\R = TT2j

0i1 = 'l l = fl'"

5201 = T2fl = 1r2j.

303j' ,< J> = 2i2<> 2J> = ,2j-

sv ----------

Thus by the definition of product:

= l = 02(1XX(XXX)iXR )

But it is also true that

ol = O1(jx)<1,rTj> (1.31).

Hence by the definition of intersection there exists a unique morphism

E such that yE = . Thus

<1,sf3>yE = <1, 3>< lij, 2ji2j> = = j.

Hence j = atC; and consequently (R,j) < (RoAXR,a).

5.21. Corollary. Let (R,j) be a relation on X. Then R-1, AX oR- and

R-oAXR are isomorphic relations on X.

Proof. Recall that since (RU' R-,j#) is symmetric (5.15), (XR-1,X*)

and ((RL4R-1)X,j1#) are isomorphic as extremal subobjects of X (5.15

and 5.9). Also (X ,x) and ((Rt UR-1)X,j,#) are isomorphic as extremal

subobjects of X (5.15) whence,

(X~IX! ,.X*x*) E (XRxXR,XXX) and hence (AXR_1' (*xX*)iX.1) and

(AXR,(XX)i XR). But by the theorem

(R- *) =- (AXR-oR-'1,*) E (R-oAxR-1 ,a*).

Consequently,

(R-',j*) (AX oR,a#) (R-loAXR,a#).
R R

5.22. Definition. Let R be a relation from X to Y. Then R is said to be

difunctional if and only if Ro(R-loR) < R and (RoR-1)oR < R.

The term difunctional relation is due to Riguet [22J.

5.23. Proposition. Let R be a relation from X to Y. Then R is difunc-

tional if and only if R-1 is difunctional.

Proof. If (R,j) is difunctional then since (Ro(PR1-oR),k,) < (R,j) we

have ((R-!R)oR-1, k) E ((R-loR)-loR-'1,k) ((Ro(R-oR))-1,k *)<(R ,j)
1 1 1

This follows from 1.38, 1.11, and 1.12. Also since ((RoR1)oR,k2) < (R,j)

it follows that

(R7-o(RoR-1),k2#) (RIo(RoR-1)-1,k2) E (((RoR- )oR)-I,k2*) < (R1 j*).

Thus (R-l,j*) is difunctional.

If (Rl,j*) is difunctional then since ((R-1)-1,j#) (Rj) (1.11)

and since, applying the above to (R-1,j*), ((R-1)-1,j#) is difunctional

it follows that (R,j) is difunctional.

5.24. Proposition. If R is a relation on X then R is a quasi-equivalence

(3.2) if and only if R is quasi-reflexive and difunctional.

Proof. If (R,j) is a quasi-equivalence then (R,j) is symmetric hence

(R:^)R-,j#) = (R,j). Thus (XR,x) (RX,j1) E (XR,j2) (5.15 and 4.6)

from which it follows that (A R(XxX)i~ ) (XR'(j2xj2)iXR1
1 XR
Since (R,j) is a quasi-equivalence then (R,y) is an equivalence

relation on XR (3.10) so ( ,X'iXR) < (R,i). Hence

(AXR,(j2xj2)iR) < (R,j) so that (R,j) is quasi-reflexive on X. Since

(R,j) is symmetric and transitive then

(Ro(R-loR),kl) < (Ro(RoR),k1) < (RoR,j') < (R,j). Similarly,

((RoR-1)oR,k2) < ((RoR)oR,k2) < (RoR,j') < (R,j). Hence (R,j) is difunc-

ticnal.

Conversely if (R,j) is quasi-reflexive and difunctional then

(A ,(XxX)i ) < (R,j) so that (A ,(XxX)i ) < (R-',j*) (1.16 and 1.12).
-XR XR XR XRR
Thus (RcR,j') < (Ro(A oR),k) < (Ro(RloR),k, ) < (R,j) and

(R-1,*) < (AXRo(R-loAXR),k) < (Ro(R-loR),kl) < (R,j) (5.20 and 1.30).

Thus (R,j) is both transitive and symmetric hence a quasi-equivalence.

5.25. Proposition. Let R be a relation on X. Then R is an equivalence

relation if and only if R is reflexive and difunctional.

Proof. If (R,j) is an equivalence relation on X then (R,j) is reflexive

and a quasi-equivalence on X. Thus (R,j) is difunctional (5.18 and 5.24).

Conversely if (R,j) is reflexive and difunctional then (R,j) is

quasi-reflexive and difunctional (5.18) hence (R,j) is a quasi-equiva-

lence (5.24). Since (R,j) is also reflexive it must be an equivalence

relation.

5.26. Theorem. Let (R,j) be a relation from X to Y and (RX,jl) and

(XR,j2) be the usual images (3.4). Then R, RoAXR, and ARXOR are iso-

morphic relations from X to Y (cf. 5.20 and 1.31).

Proof. Consider the following commutative diagrams.

J2
XR -------------- Y

T2 T2

R ------- Xx

Ti Trl

RX >~1------X

A -----RX ------- XX ___----XX

cL iRX------ t ___________- R
X P12
A >39- --- XxX X
+ X P2
a 2
xji PJ
RX & P
A -. RXxRX ___ RX
RX p2"

Observe that pl(jiXjl)iRX lPliRX = jlP2*i[ = P2(jlxjl)i;

thus there exists a morphism a such that iXa = (j xjl)iRX; i.e.,

(ARX,(' l l)iRX) (AX,ix)

90

Thus: (ARXoR,k2) < (AxoR,j') E (R,j) (1.30 and 1.31).

To see that (R,j) < (ARXoR,k ) consider the following commutative

diagram.

R -------------Xx(XxY) --:-- XxXxy

I 2 2' 3

q X

XxY xy------- -- X
1Xxy

Recall that (ARX,i) E (RX,) (1.21) hence there exists

an isomorphism o such that <1RX,1lRX> = ipRX.

Consider also the products (ARXxY,1,62), (XxR, 1,T2), and

((Xxx)xy, l, ,2)

It will next be shown that the following diagram is commutative.

(J I j )i RX1y

A XY x

R

(XxX)xy

-3> XXXXy

^C)2

,T1j,"I2j> = 02

XXi

1,19l((j Ijl)iRjx ly) P 1 ((j 1 xj ) iRXX y) =

Pl QJ 1 RX 1 1 / IL
j I P*< 1RiX N, 1> .= jlT = I j li

q20!o ((jlxjl)iRxXly )<^T1,72'j = P2 l1((jXjl)iRxXIy) =

P2(jlXjl)iRXPl = jP2*iRXoT1 = jlP2*<1RX,'RX> =

Jl1T = T'li

31301((jlxjl)iRXXly) = x2((jlXjl)iRXXly)<0T1,F2j>

ly12 T23.

r1i02(1Xxj) = f1(IxXj)<7rlj,lR> = i = Trlj

fl202(1XXj) = lrlT2(lxxj) = >lj r2< lj,'R> = T!J"

T302(1Xxj) = r2i2(1Xxj) 2= ljJ2<7Tlj,lR> = 2j

1092 = ftl = Tlj.

T2 2< TiJ,j> = '7Tf2 = Trj.

T392 = T2T2<-lj j> = r2J"

Consider the following commutative diagram.

(JlXjl)iRXX1
ARXxY >---------- (XxX) xy

'1 13
",
ARXxYA(XxR ----- XxXxY ------------~XxY

A 1 xj R k2

ARxOR

By the definition of intersection, there exists a unique morphism

( from R to (A -
R< "--

92

Thus 'Y3 = < = <"ljI2j>= j. Hence k2 = j;

that is, (R,j) < (ARXoR,k2). Hence (R,j) (ARXoR,k2).

Similarly it can be shown that (R,j) E (RoAXR,kl).

5.27. Theorem. Let (R,j) be a relation from X to Y. Then

(Ap,,(j1xj1)iyX) < (RoR-1,j#) and (AXR,(j2xj2)iXR) < (R-loR,j').

Proof. Consider the following products: (XxX,p1,P2)., (RXxPX,*pl ",2*),

(XxY,1T ,2), (RxX,pl,p2), (XxR-1,-1',2), ((XxY)xX, ~1,2),

(Xx(YxX),'IT*, 2*) and (XxYxX, 1, 2,)3).

Also consider the following commutative diagrams.

R - XxY

T'I jj T

RX ----------- X

j <2T2,7I1>
R --- ------- XxY )---- YxX

R- I

Next, the following diagram will be shown to be commutative.

jxlX
RxX .------------~ (XxY)xX
< R, IJ-> f A X -

SY <
R ----~-(RxX) (XxR-) ----- ------ XXYXX ---------> XX

X2 2 ^
( \ I Xri* 71.

,- i

- -J

To see this, it need only be observed that

o1(jxlX) = 2(lXxj*)<-Tj,T>.

.i01 1(j1Xx) = Tii (jlX)<1R',7lj> = 1iJP : Tlj.

20! (jxIX)<1R' l j> = T21 I(jXlX)<1R',Ij> = 2jpl = T2j

S30 (jX1X)<1RF"1j> = i2(jxlX) = 2<1R'R1j> = rlj-

!G2(1Xxj*)< T> = i*(lXxj) = p1< 1J,T> = Tflj.

02o2(1xXj*)<7!IjT> = Tr712*(xj-j*) T 1j*P2 = .lj*T =

1 <2,1T]>J = T2j.

3502(lXxj*) = T 2 2(1xxj*) = 2j = 7lj.

Thus the diagram is commutative and

01(jlX1)<1Rrj> = 02(lxXj*) =

Hence by the definition of intersection there exists a morphism E

such that yE = . Clearly <-,- 3>YE = thus

= j#r#Z.

It next will be shown that if a is that isomorphism for which

= iRXO then (oTl,(jlxjl)iRX) is the epi-extremal mono factor-

izauion of <'Trj,'"lj>

R ------------- XxX

1 x 1jR
RX p----------s A
RX ApRX
o

1i (jl1l)iRX = JiP1RXoi = jlPT 1'*<1RX, RX> = J1"1 = T1iJ

p2 31xj.i)iRXOTI = JlP2*iROTl = j1P2*<1RX-,Py >- Jl' = li

Thus the diagram above commutes and (orl,(j1xj1)iRX) is the epi-ex-

tremal mono factorization of (0.18). Since (ARX( 1xjl)iRX) is

the intersection of all extremal subobjects through which <'ij,Tlj> fac-

tors and since <7lj,Trlj> = j#T#E it follows that

(ARX,(j1xjl)iRX) < (RoR-1,j#) which was to be proved. The proof that

(AXR,(j2Xj2)iXR) < (R-1oR,j') is similar.

5.28. Theorem. If (R,j) is a relation from X to Y then R is difunctional

if and only if (RoR-)oR E R = Ro(R-loR).

Proof. If R is difunctional then

(R,j) (RoAXR,kl) < (Ro(R-1oR),kl) < (R,j) (5.26, 5.27 and 5.22).

Similarly,

(R,j) = (ARoR,k2) < ((RoR-1)oR,k2) < (R,j).

The converse is immediate from the definition (5.22).

5.29. Remark. Let ( be a locally small quasi-complete category having

(finite) coproducts. It is noted in passing that if has arbitrary

products; i.e., is complete, then 1 is also finitelyy) cocompletc 9 9.

Recall that the unique epi-extremal mono factorization of a morphism is

obtained by taking the intersection of all extremal subobjects of the

codomain of the morphism through which the morphism factors (0.21). Also

recall that if the intersection of all subobjects of the codomain of the

morphism through which the morphism factors is taken, then the unique

extremal epi-mono factorization is obtained (0.21). Finally recall that

if {(Ai,ai): isl} is a family of subobjects of a -object X, then the

subobject (lkJAi,a) is obtained by taking the intersection of all sub-
iE
objects of X which "contain" each (A.,a.).

Â£._i ,> y*X Observe that tt 1 >^ = J*tX, and y = k^xX^. Thus by the definition of intersection: (R A S,ifi) <_ (f'AS" 1 ^), However since r* is an isomorphism, (R A S,^) E ((R A S)" 1 ,\p*) ; whence ((RASj-^f'O < (R~V\ S" 1 ,^). To obtain the reverse inequality, note that by the definition of intersection (R -1 A S -5 ,) <_ (RAS,i|j) since jr -1 *=

4> and kT -1 X,

<^. Thus (R~ ! A S -] ,

(+>) <_ (R A S ,^) . Whence (R~V\ S -1 ,4>) < (RA S, ^) = ( (R A S)" 1 ,ip*) . Consequently: (R^AS1 ^) E ((R/IS)" 1 ,^). 1.19 Remark. Ic is clear from the definition of intersection (O.G ) that if (R,j) Â£ (S,k) and (R,j) 1 (T,m) then (R,j) < (SAT.n). 1.20. P roposition . Let (R,j) be a relation on X. Then RAA^, R'^Aj,, and R AR~ f\ Â£Â„ are isomorphic relations on X. P roo f. Consider the following commutative diagram. Â— >. R-l S/iR'^lA,,X*X

PAGE 32

26 Note that since t^ equalizes t^ and 7iÂ„ , i v = hr 0-J 6) and also that <7i 2> tt 1 >" 1 = <7T 2 '^i > > ie -> <7T 2 ,TI 1 > = *XxX 0.6). Observe that jx _1 X^ = ~ 1 j*X^ = <7 '2' T, l > ~ li X A 3 = <7r o ' ir i >i x A 3 * Consequently jx -1 X, = iÂ„A^ = iyX,. Thus by the definition of intersection: (R1 OA x ,i x A 3 ) <_ (Rf\A x ,i x X 2 ). Also observe that j*rA = j X, = i A . Whence 1 2' 1 J "l 2*1 X 2 j*xX, = i v AÂ„ so that by the definition of intersection: A l J X A 2 Thus: (RHA x ,i x A 2 ) < (R1 nA x ,i ) .A 3 ) (ROA x ,i x >. 2 ) = (R~ 1 AA x ,i x A 3 ) Clearly (RflR~ ! H A^ i^) < (Rn^.i^). But (Rr>A x ,i x A 2 ) < ( R_1 AA x ,i x A 3 ) and (RA^,!^) <_ (RA^,^). Thus: (ROA x ,i x A 2 ) < (RAA x ,i x A 2 )A(R _1 nA x ,i x A 3 ) = (RAR^AA^i^g) Hence, (RrtA x ,i x A 2 ) = (ROR-ip k A x ,i x A 6 ) . <1 XÂ» 1 X > 1.21. Lemma . Tf X is a r -object and X Â— > X*X is the unique morphism h such that ti h = it h = 1Â„, then (X, <1 , 1 >) and (A v ,i v ) are 1 iA A A A A isomorphic relations on X. Proof . Since tt <1 , 1 > = tt <1 , 1 > and i is the equalizer of ir. and tt , IaXzaa a 1 2. it follows that (X,) < (A x ,i x ). Since tt^I^j^ l y , is a section, hence an extremal monomcrphism. Clearly, ^ ] <] -v' 1 x >Tr 1 i x = l v n i Â± x = * lh and tt 2 <1 x ; 1 >tt i = lyTji^r = i,i x = T1 2 i x* I!ence ' 'Â°y ' ne definition of product, ir 1 t x X Tius (A x ,i x ) < (X,) 1.22. Exam ple, In the categories _Set_, Top . . Top ? , Grp_, Ab, and Rng , <] r\ >: x XxX can be defired by <1 ,1 >(x) = (x,x) eX>
PAGE 33

27 1.23. Remark . It is also easy to see that up to isomorphism of extremal subobjects (X,<3y,ly>) (and thus (Ay,'iy) also) is the equalizer of each cf the following sets of morphisms: {T7j ,Tr 2 }, {<1 x ,1 x >ttj , <1 x ,1 x >tt 2 }, { <1 x> 1 X >7T l' L XxX*> t<1 X' 1 X >T! 2' X XxX } ' apd {<1 X ,1 X > 1 T 1 ,< J X ' 1 X >TT 2 ' X XxX } ' 1.24. Proposition . If (R,j) is a reflexive relation on X then tt-^j and tt^ j are retractions. Proof. Since (X,) < (A x ,iy) < (R,j) there exist morphisms a and 3 such that i y a = < lv>lv > and j3 = iv. Thus ly = 1T i lx > = ^l^X = ifijBcx. Thus TT-ij is a retraction. Similarly iroj is a retraction. 1.25. Remark , Consider the following products: (XxY,p 1 ,P2)> (Y-Z,p j_ ,j?2) , (XxYxZ,?-^ J ^2' : "3^ Â» (( XxY ) xZ j^i >^2^ and (Xx(YxZ) ,1T^*,TT2*) . Tt ^ s cas > Lo see. there exist isomorphisms Â°1 = fl 2 > and 2 = Pi 7r 2*>P2 7T 2' A ' > (XxY)xZÂ»-Â» XxYxZ {Â£r Oo -Â« X.x(YxZ) such that ^iQi = p i 'it i , ^o^l r " p 2^1' ^3^1 = ^2 and "1^2 ~ ^l** T, 2Â®? = Pi ^2' and TioOo = Pi 1 * 2*' 1.26. Definition . Let (R s j) be a relation from X to Y and (S,k) be a relation from Y to Z. Consider the following intersection. s-RxzysÂ•^ XXSV l*j Y Ixk Â•Â•? (XxY)xZ 9i ^^2 Let denote that unique morphism from X x YxZ to XxZ such that a^<7r1 , :; io> = :rj and C 2 < ^1'^ 3 ; ~ ^3 wn; re ~1 and a 7 arG the projections cf XxZ to X and Z respectively.

PAGE 34

28 Let (i',j') be the unique epi-extremal mono factorization of Y> and let tlie codomain of t' (domain of j") be denoted by RoS. The relation (RoS,j') is called the c omposition of R and S. 1.27. Ex amples . In the categories Set , Grp, Ab, and Top-, the composition of R and S is isomorphic to the set {(x,y): there exists a yeY such that (x,y)eR and (y,z) eS}. This is the usual set theoretic composition of relations (which incidentally is not the usual notation for the composition of functions when they are considered as relations) . In the category Top? , the composition of R and S is the closure of the above set. 1.28. Def inition. If (R,j) is a relation on X then R is said to be transitive if and only if (RoR,j") < (R,j). A relation on an object X is said to be an equivalence relation if and only if it is reflexive, symmetric, and transitive. 1.29. Exa mple s. In the categories Set and Top , , transitive relations and equivalence relations are the usual set theoretic transitive relations and equivalence relations together with the inclusion maps. In the category Top_Â„, equivalence relations are closed sel theoretic equivalence relations. In the categories Grp , and Ab, equivalence relations are subgroups of the catesian product which are set theoretic equivalence relations. 1.30. Proposition , Let (Rijj-i) and (R^j^) be relations from X to Y and let (S,,k.) and (SÂ„,k ? ) be relations from 7 to Z and suppose (S. 1 ,j 1 ) < (R 2 ,j 2 ) and (S 1 ,k 1 ) <_ (S 2 ,k 2 ). Then (F^oS^j) <_ (R 2 oS 2 ,k).

PAGE 35

29 Proof . Since (R-,^) < (R 2 ,J 2 ) and (S^k-^ <_ (S 2 ,k 2 ) it is immediate that (RjXZ.j-jXl) Â£ (R 2 x Z,J 2 x1 ^ and ( XxS iÂ» lxk i) 1 (XxS 2 ,lxk 2 ) whence ((R^OO (XxS 1 ),y 1 ) < ((R 2 xZ)0 (XxS 2 ),y 2 ). Consequently there exists a morphism a such that the following diagram conmutes. j 2 xl > (XxY)xZ (R 9 xZ) H (XxS ) .__/__ (r 1 xz> r, (xxs 1 ) *-^x Xx(YxZ) lxk, Thus iT 3 >Y 2 a = ^i Â»T r 3 > Y^' Since (R.oSpj) is the intersection of all extremal subobjects through which <~ ,tt 3 >y factors (0.21) and since ^j.i^Yi factors through (R 2 oS 2 ,k) it follows that (R-^S^j) <_ (R 2 oS 2 ,k) which was to be proved . 1.31. Theorem . Let (R,j) be a relation from X to Y then RoAy, R, and A v oR are isomorphic relations from X to Y. Proo f. First consider RoAy From the definition of composition of relations the following commutative diagram is obtained. ixl RxY Â»(RxY) H (XxAy) Â» Â— A, --* (XxY)xY Xx A Y Â»~ -> Xx(YxY) 0, XxYxY Ixi-

PAGE 36

30 Recall that (A , i ) is the equalizer of the projections p 1 and p from Y>Y to Y. It will next be shewn that y = y. Let p , and p' be the projections of XxA to X and A respectively, and let it * and ir * be the projections of Xx(YxY) to X and Yxy respectively. Then P 1 <7T 1' TT 9 > ^ = *]Y = P 1 <7; 1 > TT 3 > YP 2 < ^1'^2 >Y = V = ^ 2 2 (1>Y. Hence <^,,v Â„>y = YLet. p, and pÂ„ be the projections of (XxY)xy to Xxy and Y respectively and let p * and p * be the projections of RxY to R and Y respectively. Since ^ j 3 P x " x = TjPj^jxl)^ = it.^ Cjxl)X ] = T^y = TT 1 <7T 1 > T1 7 > Y, T, 2^ P 1* X 1 = 1T 2 P 1^ >1 ' )X 1 = 7T 2Â°1 ^ Xl ^ X l = ^2 y = 7T 2 Y ' 3rid y = y = j'x', then the following diagram commutes. and (RxY) H (XxA. ) RxY Pl * -> R fe Y V* XxY fc> RoA, Thus since (RoA , j ") is the intersection of all extremal subobjects through which y = y factors (0.21) it fellows that (RcA^j') 1 (R,j). To see that (R,j) <_ (RoA , j ') consider the following commutative diagrams ,

PAGE 37

31 < JÂ» "2J > (XxY)xY V5~ P2 v YÂ» XxY Â— -;s> XxYxY V -** XxY -& Y *-> RxY Y Recall that (A ,i ) = (Y,<1 1 >) (1.21), thus there exists a rcorphism c?:Y -Â» A y such that i y c = . >. XxA

PAGE 38

32 It now will be shown that the following diagram commutes, J*1 Y r*YÂ» Â— >(XxY)xY 1 ^iJ.^TT J> Xx A >^ Y 1 xi -** Xx(YxY) X Y Tr 1 1 (jxl) = 7r 1 p 1 (jxl) = ^ JP 1 * <1 R Â» 1T 2 3 * > = "l^R = V ^ 2 1 (J x D<1 r ,t: 2 j> = TT 2 P 1 (jxl)7 r 2 J> = Tr 2 J Pl * = ^ j 1 R = u 2 J ^ 1 Q 1 < j> TI 2 J > = T1 1 P 1 < J s Tf 2 J > = V* ^ 2 i < -i' 7T 2 : ' > = 7T 2^I < ^ ,7, 2 J> = 7T 2^" i ] 2 (1 = 7l 1 "(l x i Y ) = P 1 <1T 1 J,0^ 2 J> = TTjj. Tf 2 2 (lxi Y ) P 1 1T 2 *(lxi Y )< 1 T 1 J,0TT 2 J> = p jlyPg^l 3 > 01F 2^ > = P l i YÂ° 7T 2^ = P 1 <1 Y' 1 Y >TT 2 J ' = ^2^ = ^2^' ^ 3 Q 2 (lxi Y ) = P 2 ir 2 *(lxi Y ) = P 2 1 Y^2 <71 1 J " ' aTr 2 J > p 2 i Y 07T 2^ = P 2 <1 Y' 1 Y >7T 2^ = ^'V = ^ ' Thus by the definition of intersection there exists a unique morphism E, from R to (RxY)fÂ»(XxA ) such that Y Y5 C 1 = 1 (jxl Y ) = 2 (lxi Y )< 1 r 1 J s 0Tr 2 J>. Thus <7f. ,tF->yC = < T' 1 Â»^o >0 i < J > 7T tJ > ' But since 7T l 9 l = *l l < -3Â» 7r 2 : * > = 1r l p "i < J' 7T 2 :i> = "^ and TT 2 < ^1'^3 > Â°1 < ^ ' TT ?J > = T '3 e L < ~' ,T '?? > = p2 < ^' 1T 2i> = ^ jt follows frora tbs definition of product that 0Â„ = j.

PAGE 39

33 Thus j = Y5 = j't'E; whence (R,j) Â£ (RoAÂ„,j'). The proof that (R,j) = (A oR,j'") follows from analogous arguments. A 1.32. Proposition. If (R,j) is reflexive and transitive on X then (RoR,r) = (R,j). Proof. Since (R,j) is transitive then (RoR,j') Â± (R,j)Since (R,j) is reflexive then (A Y ,iÂ„) < (R,j). Thus (R,j) < .(RoA j") < (RoR.j') <_ (R,j) (1.31 and 1.30). Hence (RoR,j') = (R,j). 1.33. Remark . As has been remarked in (1.27), if (R,j) is a relation from X to Y and (S,k) is a relation from Y to Z then in the categories Set, Ab, Grp, and Top , (RoS,j') may be taken to be the set {(x,z): there exists a yeY such that (x,y)eR and (y,z)e.S) together with the inclusion map j'. Thus the categorical definition of composition (1.26) yields in these special concrete categories the usual set theoretic composition. A similar remark can be made about the definition of the inverse relation. That is, the categorical definition yields the usual set theoretic definition in the categories Set , Ab, Grp , and Top to only mention a few. Indeed, the categorical definitions were obtained by analyzing the situation in the set theoretic case. However, in the category Top ? of Hausdorff spaces and continuous maps the extremal monomorphisms are the closed embeddings which leads to the. following consequences. 1.34. Example . If (R,j) is a relation from X to Y and (S,k) is a relation from Y to Z for TopÂ„-objeccs X,Y, and Z. Let T be the following set. {(x,z): there exists a veY such that (x,y)e:R and (y,z)eS} Then (RoS,j') " (clT,j) where "el" means closure with respect to the top-

PAGE 40

34 ology of XxZ (c.f . 1.27) . Proof. Recall that the extremal monomorphisms are the closed embeddi rigs, thus RoS is a closed subset of XxZ. Clearly the following diagram commutes. (RxZ)fUXxS) RoS XxZ It is evident that T CEoS whence clT C RoS. But (RoS.j") is the intersection of all closed subsets of XxZ through which Y factors, Thus (RoS,j') C (clTJ). Hence (RoS,jO = (clT.j). 1.35. Exa mple. With the hypothesis of Example 1.34, T and clT do not necessarily coincide. Proof . Let X Z be the closed unit interval with the usual subspace topology induced from the real line. Let Y be the closed unit interval with the discrete topology. Let R = {(x,y): y = x} considered as a subspace of XxY. Let S = {(y 5 z): < y < h} considered as a subspace of YxZ. It is easy to see that bath R and S are closed in XxY and YxZ respectively. Clearly T = [(x,z): < x < h] and clT {(x,z): <_ x < h} whence T CÂ£ clT.

PAGE 41

35 1.36. Example . In the category Top the composition of relations is not necessarily associative. Proof. Let X = Z be the closed unit interval with the usual subspace topology induced from the real line. Let Y be the closed unit interval with the discrete topology. Let R = {(h,h)} considered as a subspace of X*X. Let Sbe((x,y): y = x} considered as a subspace of X*Y and let Tbe{(y,z): < y < %} where T is considered to be a subspace of YxZ. Hence, each together with its inclusion map is a relation since each of R, S, and T is a closed subspace of X*X, X*Y, and Y*Z respectively. It follows that RoS = {(%,%)} and that (RoS)oT = 0. But SoT = {(y,z): <_ y <_ h) and from this it follows that Rc(SoT) = {(%,z): zeZ}. Hence Ro (SoT) i (RoS)oT. 1.37. Remark. At first glance, the results of Examples 1.34, 1.35 and 1.36 seem to be pathological, thereby casting doubt on the usefulness of the categorical definition of composition of relations (1.26). However, this should cause no more anxiety than does the fact that the set theoretic union of two subgroups of a group is seldom a subgroup. Furthermore, the results 1.31, 1.38, 1.39, 2,4, 3.1, 3-6, 3.9, 3.10, 3.12, 4.22, 5.20, 5.23, 5.25, 5.26, 5.27, 5.34, 6.13 and 6.27 seem to indicate that this definition (1.26) yields nice theorems which re-enforces its appropriateness. 1.38. Theore m. Let (R,j) be a relation from X to Y and let (S,k) be a relation from Y to Z. Then (RoS) -1 and S -1 oR -1 are isomorphic relations from Z to X. Proof . The following products shall be used: (X^ttj .rr.,), (Y p 1 *,p 2 *), (XxZ,p 1 ,p 2 ), (X*S ,P i ,6 2 ) ,

PAGE 42

36 (XxCYxZ),^*,^*), (XxYxZ,7i 1 ,^ 2 ,Tf 3 ), and ( (XxY)xZ ,p 1 ,p 2 ) . The notaLion " * " over a projection morphism shall denote the projection morphism of that product object where the product is taken in reverse order; i.e., the projections of YxX are tij and tt 2 and the projections of ZxY are ft ^ and tt 2 . Consider the following diagram. It will be shown to be commutative. RxZ *y (RxZ)O(XxS) Â»Xx S *>l x xk ZxR> (XxY)xZ V> WZx(YxX) -> XxYxZ >=?0. -> Xx(YxZ) W <'(! i , TTp Â» IT q> <7T 2 *,77 , *> f e, Â« -**> ZxY xX o 2 -**(ZxY)xx ^pg.pj: S -1 xX > v k*xl.

PAGE 43

37 0, =

and 9 = . 'j ^i > " ]^2 > "2^2 j_ " 1 > "2 1 ' 2 ^l ]

= Pi < P2>

p l > = p 2' i 1 0 1 = Vl = V ^2 e i

P l > = 7T lP2

P l > = Ti 1 <-iT 2 ,TT 1 >P 1 = T^pj ^y K ~'^ jTTo >^1 >0 1 = TT 2 1 = TT 2 |:> 1* ^3 l

p ] > = Tr 2 p 2

p l > = 7T 2 <1T 2 ' TT 1 > ^1 = ^l^l Tf 3 0 1 = ij^Oj = TjPj. TT, <71~ , TTÂ« ,71", >0Â„ = Tt^O^ = Tr^TT^*. TT 2 ^ 2 = n 2 Oy = Tr 2 2 = TfiTTn*' 71 3 2 << ''2'^l >Tr 2*' TT l* > = 7T 2* << ^2'^l >1I 2* , '' T l' <> = """l' i 3 0 2 = ^1Â°2 = *!** Thu s 1 p 1 > =

0 1 and '^^3 , ^2 > ^3 >0 2 = 2 < iT i >ir 2*Â» Tr l "> P

p >(jxl) = P (jxl) = P *. 12 2 1 ' 1 J 2 2 P (lxj*)

P *

= P= ^2

P 1 > ^ ><1) = Â°l^ Xl) = JP l' V = J* Tp l 5 p (lxj*)

= j*p *

= j*Tp *. Thu S

P 1 >(jx]) = (IXJ*)

1 f 1 Â«Tr 2 ,tr 1 >TT 2 *,Trj*>(lxk) = 7r 2 *(lxk) = <;? 2'^l >kp 2 = k * Tp V Tf 1 (k*xl) p i > = k*p 1 k *^P 2 -

PAGE 44

38 ft 2 <U 2 ,!:,7T l* >(1Xk) = TI 1 " : ( lxk ) = Pj/ if 2 (k*>Â«l) p 2 = ^1* Thus <.T 2 *, TI 1 * > (l x k) = (k*>a). Hence the diagram is commutative. Consequently by the definition of intersection, there exists unique morpbism Â£, such that the follovzing diagram commutes. (Rxz)H (x>. s) t-y V (s _i xx)n (z x r _i ) >->~> XxYxZ S -& zxyxx * It is easy to see that the following diagram commutes (RxZ)H (X*S)>*(S } xx) C\ (ZxR l ) ~> XxYxZ <7Tj ir 3> cS Â— > ZxX S _1 oR l Xxj

Since (RoS,&) is the intersection of all extremal subobjects through which y factors, then (RoS.P) <_ (S _1 oR _1 ,

(0 . Hence there is some morphism u such that a'u = B. Consequently,

a'y =

f3 u'u = B*x* and the following diagram commutes

PAGE 45

39 RoS Â»S'^oR -1 (RoS) --IT Since ((RoS) -1 , 3*) is the intersection of all extremal subobjects through which

= a Pl

= ap 2 ; (ax a )* Tr 2 (axa) = tt (axa) ap^; and Tr 2 (axa)

= ap 2

= a Pl then it follows that (axa) (axa) . But

is an isomorphism hence an epimorphisn and axa is an extremal monomorphism; thus by the uniqueness of the epi-extremal mono factorization of (axa) (0.18), ((AxA) -1 , (axa)*) e (AxA, axa). Thus (AxA, axa) is symmetric. To see that (AxA, axa) is transitive, first, consider (((AxA)xX) A (Xx(AxA) ) ,y) where y is the unique extremal monomorphism induced by the indicated intersection. It will next be shown that (AxAxA,axaxa) and (((AxA)xX)f\ (Xx(AxA)) , y) are isomorphic as extremal subobjects of X x X x a. To show this it will be shown that (AxAxA,axaxa) is precisely the intersection of ((A x A)xX,Q. ((axa)xl )) and 1 A (Xx(AxA),0 2 (l x(axa)).

PAGE 56

50 Consider the products: (XxXxX,^ ,v 2 ,*,) , (Xx(AxA).p ,p ), ((AxA)xX,3 lS 3 2 ), (Xx(XxX), 1 r 1 *, ir2 *), ((XxX)xX,ft * * *) and (AxAxA,p 1 ,p 2 ,p 3 ). Observe the following equalities. TT 1 1 ((axa)xl x )Â«p 1 ,p 2 >,ap 3 > = ir 1 ^ 1 *((axa)xl x )<

> ap 3 > = Tr 1 (axa)p 1 Â«p 1 ,p 2 > ,ap 3 > '= ^ (axa)

= ap 1

= apj = it. (axaxa). i 2 1 ((axa)xl x )Â«p 1 ,p 2 >,ap 3 > = t^tt 1 *((axa)xl x )Â«p ] ,p 2 > ,ap~ 3 > 7T 2 (a-.a)

= ap 2 < Pl ,p 2 > = apÂ„ = ttÂ„ (axaxa) . Tf 3 e i (( aya ) x l x )<

Â» a P 3 > = ^ 2 *((a^a)xl^)Â«p 1 ,p 1 >, a p^ > = l x P2<

> a p3 > = 1 x a ^3 = a ^3 = ^3 (axaxa) ~> 1 02( 1 x x ( axa ))< a P 1 >

> = Tr 1 *(l x x(axa))> = ^x ap l = a Pl = iri(axaxa) ^2 e 2^ I X x ^ axa ^ > = Tr i 7r 2*^ 1 x x ^ axa ' ^ > = TT 1 (axa)p2> = ti j (axa)

= api = a Po = TT^(axaxa). ^3Â®2 (l x x (axa))P 3 >> = 7T 2 7r 2*^X x ( a> ' a )) ca P 1 > < P2'P3 >> = Ti2(axa) = ap2P~3 > = a f>3 ~ tt^ (axaxa) . Thus by the definition of product the following diagram commutes.

PAGE 57

51 :

,ao,> V V 2 AxA x A V*; (XxX) xX axaxa Xx(AxA) >v=~ l x x(axa) ^=Xx (XxX) Now, if (W,6) is a subobject of XxXxX so that there exist morphisms y and Y such that (faxa)xl )y =6 = (1 x(ax a ))Y then con12 1 x 1 2 X 2 sider the morphism = C from W to AxAxA. It will 1112 112 2 2 be shown that <

,ap >Â£ = y and >Â£ = Y Â• 12 3 1 12 3 2 Since p 1

? = p^ = Pj^Yj and P 2

P 2 >Â£ = P~ 2 C = P 2 Pi Y i it follows that p 1 Â«p 1 ,p 2 >,ap 3 >C =

C = PjY^ Now since 0j ( (ax a ) x 1 x )y 1 = 2 U x x (a x a) )y 2 it; follows that P 2 Y, = fi 2 '' ; ((axa)xl x ) Yi = i 3 1 ((ax a )xl x )Y 1 = '" 3 9 2 (l y x (axa) )> 2 = tt 2 it 2 *(1 x( a x a ))Y 2 = TT 2 (ax a )p 2 Y 2 = ap 2 P 2 Y 2 Whence \$ 2 y 1 = 3 2 Â«P ] ,P 2 > ,aP 3 > ? = ap 3 C = ap 2 P 2 Y 2 Thus <

,ap >l = Y Â• 12 3 1 Again since ((fixa)xl )y = 0(1 x(axa))y,, it follows that X A 1 Z A *P X Y 2 = " 1 *(l x x (axa))Y 2 = ^(l^Caxa)^ = ^((a^xl^ = Trft *((axa)xl )y. = ir (axa)p y, = ap p y Â• 11 a1 1 xl lli Kence p >5 = ap^ = a P 1 P 1 Y 1 = P^Since p

C = P 2 ? = P 2 P 2 Y 2 ahd P 2 < " P 2'^3 > ^ = P 3 ? = P 2 P 2 Y 2 Lt follows that E, = p y . Hence p C = P Y.~2 3 2 2 2 ) 2 j 2 3 2 Z Consequently
PAGE 58

52 ((AxA)xX,3 i ((axa)xl x )) . it next will be shown that the following diagram commutes. axaxa A*A*A Â»Â• XxXxX -> XxX *" (AxA) o (AxA) AxA "f (t#,j#) is the epi-extremal mono factorization of <~ri . tt 3 > (axaxa) Now tt 1 (axa) = api = api = tt x (ax a x a ) and Tf 2 (ax a )

= ap 2

= ap 3 = tt 2 ( a x a x a ) . Thus the above diagram commutes.. Since ((AxA)o(AxA) , j#) is the intersection of all extremal subobjects through which (axaxa) factors then ((AxA)o(AxA),j#) <_ (AxA.axa) whence transitivity is obtained. 3.4. Canonical Embedding . Let (R,j) be a relation on X. Let (i^jj), (t 2 ,j 2 ). anc * ^ T 3'J3^ ^e t ^ ie epi~extremal mono factorizations of Tiij, TT ? j, and tt 2 j* respectively. Let RX, XR. and XRÂ™ 1 denote the domains of Jl> J2Â» ar '<^ J 3 (codomains of Tj.t^ , and t 3 ) respectively. XR l 9r-

PAGE 59

53 R >V-*Â» x> anc * (XR -1 ,j 3 ) of X have precisely the same underlying sets as above endowed with the subspace topology induced by the topology of X. See Section 4 (4.1, 4.2, and 4.3) for a more detailed discussion. It is easy to see that in the category Set , a symmetric, transitive relation on a set X is an equivalence relation on a subset of X. Recall the discussion in Section 2 (2.0) of the remark? of Lambek who obtains the similar result for homomorphic relations on rings with identity. This result we wish to generalize. In order to do this we must first be able to pick out the subobject.of X on which the relation is an equivalence relation. Referring to the above diagrams, since t is an Isomorphism (1.9) and since the epi-extremal mono factor izaticv is unique (0.18) it is clear that (RXjj-^) = (XR~*,j 3 ). That is, there exists an isomorphism k from RX to XR" 1 such that j 3 k = jj (see 4.4 and 4.5).

PAGE 60

54 Consider the product (RX*XR, p j , p" 2 ) . Also, consider the morphism Cj i x J2' from R to XxX Since ^i 1 x J2) = JiPiT 2 > = Jixi = TTij and 11 2 (J 1 X J2) T 2 > = J2T2 = ^2J it follows from the definition of product that (j i*j 2)^1 ,T2 > = JNote, is an extremal monomorphism since j is an extremal monomorphism (0.16). Now suppose that (R,j) is symmetric on X. Then it follows that there exists an isomorphism a so that jot = j* (1.13). Ihus i^ja = ttz3* and this together with the fact that a is an isomorphism and the uniqueness of the epi-extremal mono factorization implies that (XRjj;,) = (XR ,33). That is, there exists an isomorphism g so that J2^ =; 33* Thus it is routine to see that the following diagram commutes,

PAGE 61

55 Consider the following products: (XRxXR,Pi ,P2) and (XxXjTTj ,7T 2 ) . Letting i< = (gkxl XR ) then (J2. X J2 )^ = J > since -'\ (J2 X 32H = J2Pl^ = J2Bkp! = J23 k Tl = Jm = tfij and n2(J2 x J2)^ = J2Â°2* = J2 1 x rP2 < Ti ,t 2 > = J2 T 2 = ^2 J Â• Thus the following diagram commutes and the relation (R,i|0 on XR shall be called the canon ical emb edding of R into XRxXR. R -> RX*XR 3kxl XR XRxXR * J2 X J2 XxX 3.5. Lemm a . Let (R,j) be a symmetric relation on X. Then (R,^) is a. symmetric relation on XR. Proof . Suppose that (R,j) is symmetric on X and let (R,i|)*) be the inverse of (R.ifj) on XR. Then

il; = \p*x* . It is easy to verify that Tr l > J (J2 x J2) < P2Â»Pl >1 K Thus since (R~ ,j*) is the intersection of all extremal subobjects through which j factors there exists a morphi-sm X from R"" 1 to R so that (j 2 x j 2)^*^ j*Â« But j 5 ' : T < TT 2 j' rT i > j whence (j 2 xj 2 )^AtÂ« = (j 2 x j 2) J = j*T. So j*T = ((j 2 X J2)^*)^T " C(J2 x J2)" 1 f'*)T*. Since (J2 X J2)4'* is a monomorphism it follows that Xt = t* . Recall that x and x* both are isomorphisms (1.9) Rence X is an isomorphism.

PAGE 62

56 Recall that by the definition of symmetry (1.10), there exists a morphism a from R _1 to R so that jet = j*. Thus (j 2 x j 2) ^'^Aa -1 = j*a -1 = j = ( j 2 * j 2 ) 'I' Â• But since (j 2 X J 2) is a monomorphisra this impli es that ^""Xa"" 1 = xl RxXR M Â» (XRxXR)xXR XRxXR -^ Xx (XxX) VJ

PAGE 63

57 Clearly (j 2 xj 2 xj 2 ) = (j 2 xj 2 )

. Also, G 1 (j>-l x )(l R xj 2 )A 1 Q 1 (j x j 2 )\ 1 = e 1 ((j 2 xj 2 )ijjxj 2 )X 1 01 ((J2 x J2^ x i2^^ xl XR^' V l = (J2 x J2 x J2)01^ xl XR) X l and 2 ( 1 X X J ) ^2 X1 R^ X 2 = Q 2^2 x l) X 2 = 2^2 X (J2 X '2^)^2 = 02(J2 X (J2 X J2))( 1 XR X ^)' V 2 = (J2 x 32 x J2)Â°2( 1 XR xl i') X 2 as can be verified in a straightforward manner. Thus the diagram above is commutative, and in particular, Â©l(j x lx)( 1 R x 32^1 = (J2 X J2 X J2) 6 = Qi^X*!) (J2 xl R^ X 2Hence, by the definition of intersection there exists a unique morphism A such that Y* = (j 2 x j 2 x j 2 )5. Let (RoR,4j//) denote the composition of (R,i|;) with (R,il>) on XR. Let (RoR,j') be the composition of (R,j) with (R,j) on X. Then p~3>6 = ij)#T# where t# is an epimorphism and < ^j > ^3 > Y = J ' t ' where t' is an epimorphism. But since yA. (J2 x J2 x J2^' *Â•*follows that

YX = (j 2 x J2) < Pl sP3 >1 ^ so that (j 2 x j 2 )it'#'r# = j't'A. Hence the following diagram commutes. (RxXR) O (XRxR) 1# Â™Â» RoR y s / (RxX) A (XxR) Â£ / 4># XRxXR / RoR V?-Â— i X-i J 2 J 2 -> XxX

PAGE 64

58 Since x# is an epimorphism and j' is an extremal monomorphism, by the diagonalizing property (0.19) there exists a unique morphism Â£ such that j'C = (j 2 x J2H# and Ct# = x'A. But this says that (RoR, (.i 2 xj 2 )^.:0 _< (RoR.j'). Since (R,j) is transitive (RoR,j T ) <_ (R,j) hence (RoR, (j 2 x j 2H*) Â£ ( R >j) =: ( R > (j 2 X J 2^) Â• Hence there exists a morphism from RoR to R such that (j^^H = Cj 2 X J 2^^ * ^ a ^n ' J2 X J? *" s a monomorphism so that \po = ifi// which says that (RoR, -> (XxY)xX ->(RxX)H (XxR1 ) >* *Â» XxYxX XxR~ J Â»-> Xx(YxX) l Y xj* Consider the following products: (XxYjir, ,ir 2 ) , (XxYxX,^ ,ir 2 , tt 3 ) , (RvX, Pl ,p 2 ), (XxR-l,p 1 *,p 2 *) J ((XxY)xX,ir 1 *,Tr 2 *) 1 (Xx(YxX) ,ff x ,if 2 ) , (YxX,tt 1 ,tt 2 ) , and (XxX,tt 1 ,tt 2 ) . Now, ^l lCj xl x ) <1 RÂ» 7r l3 > = Tr i 7T i A (J xl x) <1 R'"lJ' = TT lJ p l = ir 2 7r l*(J xl x) <1 R' ir lJ > = H 23w 3 e 1 (jxl x ) = ir 2 *(jxl x ) = P 2 <1 r Â»tt 1 J> = T^j.

PAGE 65

59 T7 1 2 (l X x j*) = ffl(l x x j")<7T 1 j,T> = Pl* = lTlJ . 7T 2 2 (l x xj*) = TT l ff 2 (1 X X J >V ' ) <7r l J ' T> = ^ 1 J " P * <1T 1J> T> = TT ^ 3 * T = TTi<1T2,Tr 1 >j = Tr 2 3 Â• TT 3 2 (l x XJ*)<'f] ] -T> = TT 2 ff 2 (i x x j ;!: ) < j,T> = TT 2 j*T = U 2
T < 1 > J = ^ljThus by Lhe definition of p act the diagram commutes. Hence there exists a morphism E so that AjE = < 1^,itiJ > and A 2 E = . From the above it is easy to see that TTiyE = 1'iYE = ttiJ = ^3Y^ = 7T 2 <7f l Â» Tr 3 > Y^Recall that (A x ,i x ) ^ s the equalizer of ttj and tt 2 hence there exists a morphism <{> such that i x ^ = yZ. Let (RoR ,j#) be the indicated composition of relations and let T# denote that epimorphism for which j#T# = y. Thus, combining the above results, = < ^i,'^3 > y^ = j#T#E = ix^' Since (Ay,iy) and (X,) are isomorphic as extremal subobjects of X*X (1.21), there exists an isomorphism A such that A = iy. Consequently, Acf> = iy<{> = < tti,tt3 > yE = . Now tt j A = lyA<Â£ = Ac}) = tti = tt j j and by hypothesis Tijj is an epimorphism; thus, since A is an isomorphisrrij it follows that has (cj>,iy) as its epi-extremal mono factorization. Eut this means that (Ay,iy) is the intersection of all extremal subobjects of X>X through which factors (0.21). Recall that <7V 1 J > fi 1 J > = j#x#E, thus (Ly,Â±v) <_ (RoR _1 ,j#) which was to be proved. 3.3. Coro llary. If (R,j) is symmetric on X ther. (R.OR, ij#) , the composition of (R>i|0 with (Pv. C) on XR, is reflexive on XR. Procf_. Since (R,j) is symmetric on X then (R*i|0 is symmetric on XR (3.5) hence (R,i{0 = (R -1 ,^*) (1.13). Referring to the diagram in (3. A) fol-

PAGE 66

6C lowing the d ! inition of the canonical embedding it is immediate that Pjip is an e t irphism since p^ = 3kx and each of 3, k, and t, is an -1 Â•v epimorphism. Thus (RoR ,i|;) = (RoR, 4)//) is reflexive on XR (3.7). 3.9. Corollary . If (R,j) is a quasi-equivalence on X then (R,j) is an equivalence relation if and only if ir,j is an epimorphism (respectively if ajid only if tt 2 j is an epimorphism). Proof . If (R,j) is an equivalence relation then (R,j) is reflexive and a quasi-equivalence. Thus by Proposition 1.24, ir,j and tt 2 j are retractions hence epimorphisms . Conversely, if ttjj is an epimorphism then applying the theorem -1 and Proposition 1.30, (A ,i ) < (RoR ,j//) < (RoR.j') < (R,j) so that X X (R,j) is reflexive and hence is an equivalence relation. (If tt 2 j is an -1 ' -1 -1 -1 epimorphism then ( A y,i x ) < (R oR,j#*) _< (R oR ,j *) Â£ (R ,j*) and (R~\j*) H (R,j).) 3.10. Corollary. If (R,j) is a quasi-equivalence on X then (R,ij0 is an equivalence relation on XR. Proof . (R,40 is a quasi-equivalence on XR (3.6) and (RoR,^//) is reflexive on XR (3.8). Thus (A YP ,i v D ) < (RoR,^//) < (R.ifi) whence (R,iJ<) is reflexive aK XR Â— Â— and thus is an equivalence relation on XR. 3.11. Prop osition. If (R,j) is a quasi-equivalence on X then (R,j) and (RoR,j') are isomorphic relations on X. P_roo_f. By Corollary 3.10 (R,40 is an equivalence relation on XR whence (RoR,^//) and (R,^) are isomorphic relations on XR (1.32). Recall that there exists a mcrphism E, such that the following diagram commutes (3.6).

PAGE 67

61 t// (RxXR)Ti (XRxR) t'X RoR Â— -l> RoR (J 2 x J 2 )H l V Â— JÂ»XxX , f\S Thus (RoR, (j 2 xj 2 H#) 1 (RoR,j'). But as mentioned above (RoR,tp#) = (R,4<) hence there exists an isomorphism \i\ such that 0\ff = if). So by the definition of the canonical embedding (3.4), j'Â£X# = (J 2 xj 2 )ii;//X// = (j 2 xJ 2 H = j. But this implies that (R,j) <_ (RoR,j'). Thus since (R,j) is transitive, (R,j) = (RoR,j') which was to be proved. 3.12. Proposition . Let (R,j) be a relation on X. Then (R,j) <_ (^x^X^ if and only if R is symmetric on X and (R,i|0 <_ (Axr> i-xii' Â• Proof. If (R,j) <_ (AxÂ»ix) then there exists a morphism a such that j = J-x aThus tt,j = Tjix = 1J 2^X a = ^/J wnence 7i , j = tt 2 j = irjj = Tr 2 j . Thus by the definition of product j = j. Consequently the epi-extremal mono factorisation of <7r 2 ,ir 1 >j is (1r,j) and so (R,j) = (R _1 ,j v ' c ); i.e., (R,j) is symmetric. Recall that j = (j 2 *j 2 )4' (3.4). Thus "lJ = ^l^z^l^^ = 3 2 p lV and ^ = " T 2^2 X J2^^ = J2 p 2^' But "l^ = T ' 2.3 hence j 2 Pii|i = j 2 P 2 '^Â» Since j 2 is a monomorphism it follows that p,i|i = Po'ji. Recall that (Axr^xr) is tne equalizer of p^ and p ? . Hence there exists a morphism g such that ixK.5 = i'This implies that (R,ip) < (Axr.Ixr)Conversely, if R is symmetric and (R,'W Â£_ C^xR'^Xr) tk en there exists a morphism 3 such that il> = ixR^S hence

PAGE 68

62 Pllp = PjixR^ = P2 i XR P = p 2^' Since (J 2 X J 2^ = J> we have ^1J = Ti(J2 x J 2^ = J 2 P l 1 ^ = 3pP2^ = 7T 2^2 X J2^ = lr 2J ' Thus ^ 1 J = ^ so that, there exists a morphisin 2 such that j = iya. This means that (R,j) < (A Xs ix)3.13. Definition . Let (R,j) be a relation on X. Then R is said to be a circular re lation if and only if RoR 5_R . This notion is due to MacLane and Birkhoff ^20 J (exercize 3, page 14) . 3.14. Froposi tion. Let (R,j) be a relation on X. Then R is a circular relation if and only if R is a circular relation. Proof . If R is circular then RoR ^_R -1 . Thus R _1 oR -1 = (RoR) -1 ^(R -1 ) -1 =R (1.38, 1.12 and 1.11). Hence R _1 is circular . Conversely, if R -1 is circular then by the above, (R -1 )" 1 E R is circular . 3.15. Theorem. Let (R,j) be a relation on X. Then R is an equivalence relation on X if and only if R is reflexive and circular. Proof. If R is an equivalence relation then R is reflexive. Since R is transitive and symmetric, RoR <_ R = R L hence R is circular. Conversely, if R is reflexive and circular then R 1 is reflexive (1.17) and R _1 is circular (3.14). Hence R~ : = R~ 1 oA y _
PAGE 69

SECTION 4. IMAGES 4.1. De finitio n. Let (R,j) be a relation from X to Y and let (A, a) and (L,b) be extremal subobjects of X and Y respectively. Consider (11 A(AxY) ,y) and (R H (X*B) , 6) . Let (f ,a) and (f .6) be the epi-extremal 1 2 mono factorizations of it y and ir 5 respectively. Denote the domain of a 2 1 Â• by AR and the domain of 6 by RB. Thus the follovring diagrams commute. RA(AxY) Â» A*Y ARV^' -> . Y 4.?. Remark. Since (X.l ) and (Y,l ) are extremal subcbjects of X and Y X Y respectively, then (RA(XxY),y) = (R,j) and (R f\(X*Y) , -5) = (R,j) whence (XR,oO is precisely the extremal subobject (XR ; jÂ„) used in the canonical embedding (3.4). Since X = Y in 3.4 then also (RY,3) is precisely 63

PAGE 70

64 (RX,j ) used in 3.4 4.3. E xamp le. In the category SeX, for (A, a) <_ (X,l ), (B,b) <_ (Y,l ) A Y and (R,j) < (XxY,l ), AR = ; {ycY: tl exists aeA such that (a,y)t:R} RB = {xeX: there exists heB such that (x,b)eR}. This is easily seen since Rf\(AxY) = {(a,y): acA, (a,y)eR} and RA(XxB) = {(x,b): t>Â£B, (x,b)eR), and AR is the set of all second terms of elements of RfV(AxY) and RB is the set of all first terras of elements of RA(XxB). In the category Top , AR and RB have precisely the same underlying sets as above. They are endowed with the subspace topology determined by the topology of XxY. In the category Top , AR and RB have' precisely the same underlying sets as in Top for it is easy to verify that AR and RB are closed subsets of X and Y respectively. Recall that the image of a morphism in Top is the closure of the set theoretic image (0.15). 4.4. The orem. If (R,j) is a relation from X to Y and (A,a) is an extremal subobject: of X then (AR,a) and (R^Ajg) are isomorphic extremal subobjects of Y. Proof. Consider the following commutative diagrams. Â— > Y -^r AxY ^ r axl v Â£Â» AR *"

PAGE 71

65 R ifMYxA) >Y R>*Â» *^ XxY Â»R 1 <7T 2 ,7T 1 > -*YxX It can be shown in a straightforward manner that (?.X] ) = (1 x a )

2 1 Y Y 2. 1 where p and o are the proiections of A*Y. Hence 1 '2 -" (1 x a )

>. = (a*l )A = y = iA = j A rA . Y 2 12 2 1 Y 2 2 1 2 11 1 Thus by the definition of intersection there exists a morphism Â£ such that 5^ = y = j*tX = (1 *a)

X . Hence ft dÂ£ = fi y -Â• Tt y = cf . But ff 6Â£ = gi: t" . Thus, since. (AR.et) is i 12 1 7 1 12 the intersection cf all extremal subobjects through which it y factors (0.21), it follows that (AR,a) < (R -1 A,3). Similarly, it follows that <7t .71 > _1 j*A = *" 1 5 = jt -1 X (axl )

~*\ whence thare 2' 1 2' 1 J 3 Y 2' 1 4 exists a mciphism E* such that yÂ£* = < it Â»"" >_1 o. Then Tt -1 5 = f 6 =-3t = tt yÂ£* = ax Â£*. Again, since (R'^AjB) is the 1 intersection of all extremal subobjects through which ft 6 factors

PAGE 72

66 (R 1 A,&) Â± (AR,a). Consequently (R 1 A,&) = (AR,c). 4.5. Corollary . If (R,j) is a relation from X to Y and (B,b) is an extremal subobject of Y then (RB,8) and (BR -1 , a) are isomorphic as extremal subobjects of X. Proof. Recall (CR" 1 )" 1 ,j#) = (R,j) (1.11). Letting (R -1 ,j*) play the role of (R,j) and (B,b) the role of (A, a) in the theorem, the following is obtained: (BR -1 , a) = ( (R~ ] ) _1 B, B#) E (RB,B). 4.6. Corollary . If (R,j) is a symmetric relation on X and (A, a) is an extremal subobject of X then (AR,a) and (RA,B) are isomorphic as extremal subobjects of X. (In particular, (XR,j ) and (RX,j ) are isomorphic as extremal subobjects of X as was shown directly in 3.4.) Proof. Recall that (R -1 ,j*) E (R,j) (1.13). Hence by the theorem (AR,a) = (R _1 A,B) = (RA,8). 4.7. Pro position . Let (A^a^) and (A 2 ,a 2 ) be extremal subobjects of X and (R,j) be a relation from X to Y. If ^2,3^) <_ (A2,a2) then (AjR^j) <_ (A 2 R,a 2 ). Proof . By hypothesis there exists a morphism u so that a 2 y = a^. Thus, there exists a morphism E, such that the following diagram commutes.

PAGE 73

67 Thus t^Yi = Tr 2^2^ whence, because (AjR,^) is the intersection of all extremal subobjects through which t^Yi factors and tt 2 y 2 Â£, factors through (A 2 R,a 2 ), (AjR.01}) <_ (A 2 R,a 2 ) which was to be proved. 4.8. Proposition . Let (B^jbj) and (B 2 ,b 2 ) be extremal subobjects of Y and (R.j) be a relation from X to Y. If (B^b,) <_ (B 2 ,b 2 ) then (RB^Bj) 1 (RB ? ,6 2 ). Proof . (RB^B^ = (BjR" 1 ^!*) Â± (B 2 R' 1 ,B 2 *) = (RB 2 ,B 2 ) (4.5 and 4.7). 4.9. Proposition. Let (R,j) and (S,k) be relations from X to Y and (A., a) be an extremal subobject of X. If (R,j) <_ (S,k) then (AR,a) <_ (AS, a). Proof . In a manner similar to that in the proof of 4.7 one can establish the existance of a morphism E, such that the following diagram commutes. RA(AxY) X*Y AxY Kence the following diagram commutes, SH(AxY) 3X*Y Thus, since (AR,a) is the intersection of all extremal subobjects through which * 2 y factors, and ir^y factors through (AS, a), it follows thai (AR,a) <_ (AS, a) which was to be proved.

PAGE 74

68 4.10. Pro position . Let (R,j) and (S,k) be relations from X to Y and (B,b) be an extremal subobject of Y. If (R,j) <_ (S.k) then (RB,3) <_ (SB S 3). Proof. (RB,6) = (BR" 1 , 6*) <_ (BS _1 ,0*) = (SB, 8) (4.5, 1.12, and 4.9). 4.11. P roposition . Let (R,j) be a relation from X to Y and let (Ai,ai) and (A ? ,a 2 ) be extremal subobjects of X. Then ((AjAA 2 )R,a) < (A 1 RAA 2 R,a). Proof . Since (AjAA 2 ,a) < (A^a-,) and (A i r\A 2 ,a) < (A 2 ,a 2 ) it follows that ((A 1 AA 2 )R,a) < (AjR.cxj) and ((A 1 AA 2 )R,a) < (A 2 R,a 2 ) (4.7). Thus ((A 1 AA 2 )R,a) < (A 1 RAA 2 R,a) (1.19). 4.12. Preposition . Let (R,j) be a relation from X to Y and let (B^bj) and (B 2 ,b 2 ) be extremal subobjects of Y. Then (R(B i nB 2 ),6) <_ (RB i nRB 2 ,8). Proof. (R(B i riB 2 ),8) = ((B 1 AB 2 )R" 1 ,B*) < (B 1 R~ 1 AB 2 R~ 1 , 6*) = (RBjARB^f?) (4.5 and 4.11). 4.13. Propo sition . Let (Ri 5 ji) and (R 2 ,j 2 ) be relations from X to Y and let (A, a) be an extremal subobject of X. Then (A(RjAR 2 ).a) Â£ (AR 1 AAR 2 ,a). Proof . It is clear that there exist morphisms ^ and Â£ 2 such that the following diagram commutes. ^ R x A(AxY) (R x nR 2 )r> (axy) Â»Â—

PAGE 75

69 Thus tt 2 y = Tr 2 Y 1 ? 1 = 1T 2 Y 2^2" A Â§ ain since (A(R AR 2 ),a) is the intersection of all extremal subobjects through which irÂ„y factors it follows that (A(R 1 AR 2 ) s a) < (AR^c^) and (AfR^R^a) Â£ (AR 2 ,a 2 ). Hence (A(R 1 AR 2 ),a) < (AR 1 AAR 2 ,5) (1.19). 4.14. Proposi tion. Let (R,,^,) and (R 2 ,j 2 ) be relations from X to Y and let (B,b) be an extremal subobject of Y. Then. ((R 1 AR 2 )B,B) < (RjBARgB.B). Proof. ((R 1 AR 2 )B,6) = (B^/lR,,)'" 1 , g.*) <_ (BR; 1 /^ BR -1 ,g*) = (R BAR B,3) (4.5 and 4.13). 4.15. Prop osit ion. Let (R,j) be a relation from X to Y then (R,j) and (RA(RYxY),Y) are isomorphic as extremal subobjects of X X Y. P roof . Consider the following commutative diagrams. > XxY RYxY RA(RYxY) >\$>Â— ?** X X Y Since (j ^xl ) < t, , ., j> = = j, there exists a morphism Â£

PAGE 76

70 RY*Y < ^l^ 2 i > R Thus (R,j) Â£ (RHCRYxY) ,y) Â• Clearly the reverse inequality holds so that (R,j) = (Rn(RYxY),>). 4.16. Prop osition. Let (R,j) be a relation from X to Y. Then (R,j) and (R A (X*XR) , 6 ) are isomorphic relations from X to Y. Proof . Analogous to the proof of 4.15. 4.17. Corollary. Let (R,j) be a relation from X to Y. Then (R,j) and (R A (RY*XR) , B) are isomorphic relations from X to Y. Proof . (R,j) " (RO(RYxY),y) = (R Ci (X*XR) ,'6) (4.15 and 4.16). But since (RY,j|) and (XR,j 2 ) are extremal subobjects of X and Y respectively it follov/s that ((RYxY)A(X>'-Y) are isomorphic relations from X to Y. Consequently by the definition of image (4.1), (AR,a) and ( (RY/1 A)R,c:) are isomorphic as extremal subobjects of Y. 4.19. Corollary . Let (R,j) be a relation from X to Y. Then ((RY)R,a) and (XR,j,.) are isomorphic as extremal subobjects of Y.

PAGE 77

71 Proof. Let (X,l v ) play the role of (A, a) in 4.18. .A. 4.20. Coro llary. Let (R,j) be a relation from X to Y and let (B,b) be an extremal subobject of Y. Then BR"' 1 and (BOXR)R -1 are isomorphic as extremal subobjects of X. Proof. Immediate. 4.21. Proposition. Let (R,j) be a relation from X to Y. Then (RoR" 1 ^//) and (RoR _1 (RYxX) ,y) are isomorphic relations on X. Proof . Consider the following diagram. RYxX (RxX)H(X^R ! )Â» RoR To see. the diagram is commutative it need only be observed that (j 1 xl)(x i xl) = G 1 (jxl) . To show this note that (ji^UCi^l) = (Ji^xl) = (tt^IxI) an d u 1 G 1 (jxl) = if!Â©! (j x D = TTjjPj fr 1 (TfiJ x D, ^2 < ~l' ?; 3 >e i(j Xl ) = ^3 C j(J xl ) = P2 = ^aC-n"! j x l) Â• Thus, since (RoR" ,j#) is the intersection of all extremal subobjects through which < ~ 1 1 > ^f 3 >_ v factors , it follows that

PAGE 78

72 (Roir^j//) < (RYxX,j xl). Whence (RoR l ,j#) < (RoR^H (RY-X),y). 4.22. Theorem. Let (R,j) be a relation from X to Y. Then (RY,j ) and ((RoR OX, 3) are isomorphic as extremal subobjects of X. Proof. Consider the following products: (X*Y*X, tt ,tt tt ) , (Xx(YxX),P 1 ,p 2 ), ((XxY)xX,p 1 *,p 2 *) s (R^ 3 > Y = i-,^ (j *1) ^ = Tr 1 P 1 *(jxl)X 1 = T^jp.^. Thus TT 1 A = lr^jp^j = Ji^P-^. RoR Â«(RoR _1 )X Hj# '^Â•XxX * X Since <7,' i ,tt >y = j#x// and tt j ty = Bt , the following diagrai commutes . (RxX) H(XxR-l) tt// -J* (RoR -1 )X lf'l A l 1 RY *?Â— * X Jl But since Â£r* has the diagonal propertv (0.19) and it// is an epir-' mocphisra ana j is an extremal monomorphism then there exists a morphism I such that j-,5 = B and TjpjXj = Â£tt#. Thus ((RoR" 1 )X,3) <_ (RY,j ). Next it will be shown that the. following diagram is commutative.

PAGE 79

73 -*Â• RxX R -3*. (RxX) A(XxR _1 ) Â» jxl lxj* -> (XxY)xX Â•A -** Xx(YxX) w 1 1 (jxl) = Tr 1 P 1 *(jxl) = TT 1 jp 1 = r^j. ^ 2 1 (jxl) = Tr 2 P 1 *(jxl) ir 2 jp 1 = ir 2 j . TT 3 1 (jxl) = P 2 *(jxl) = P 2 <1 R ,TT 1 J> = T^lJ. ir 1 9 2 (lxj*) = P 1 (lxj-0

= tt 1 = ir x j Tr 2 2 (lXJ*) = TT J p 2 (lxj*)<7T 1 J J I> = 7r 1 J : ' : ^ 2 = l^j*! = T1 i <1T 2' 7T i > J = T 'ii TT 3 2 (lxj*) = Ti 2 p 2 (lxj*) T> = tT2J"T2 <7r lJ' T> = v 2i'' T = Consequently there exists a morphisra t* such that the above diagram commutes and such that y?* = Y5* = l_ Â— ._> xxyxx (RxX)H(XxR 1 )Â»RoR-1 V > (RoR X )X Â— fxx j *1 ->> X .Â«,> py "^

PAGE 80

74 Since (RY,j,) is the intersection of all extremal subobjects through which t^j factors, it follows that (RY.jj) = ((RoR -1 )X, 3) . Thus (RY,jj) = ((RoR -1 )X,8) which was to be proved. 4.23. Corollary. Leu (R,j) be a relation from X to Y. Then (XR,j 2 ) and ((R -1 oR)Y,8) are isomorphic as extremal subobjects of Y. Proof. (XR,j 2 ) = (R1 X,j 3 ) = (((R-^oCR1 )l )Y,l) = ((R^o^Y.P) (4.4, 1.11 and 4.22).

PAGE 81

SECTION 5. UNICES 5.1. Definition. If {(R.,i.): iel} is a family of relations from X to Y then let (^J^R-pj) be the intersection of all relations (i.e., extremal iel subobjects of X>-Y) "containing" each (R-,j.) (where containment is in the sense of "factoring through" as noted in Remark 0.12). (i*/R-:,j) shall iel be called the relation theoretic union of the family {(R. ,j. ): id}. 5.2. E xamp les. In the category Set the relation theoretic union is the usual set theoretic union together with the inclusion map. In the category Top the relation. theoretic union is the usual set theoretic union endowed with the subr.pace topology determined by the topology of X X Y together with the inclusion map. In the category Top ? the relation theoretic union is the closure of the set theoretic union together with the inclusion map. In the categories Grp and Ab the relation theoretic union is the subgroup generated by the set theoretic union of the relations. 5.3. Propos ition . Let {(Ri ; ji): iel} be a family of relations from X to Y, let (LJRi,k) denote the usual categorical union of subobjects, let iel (o.j) be the epi-extremal mono factorization of k and let the codomain of o (domain of j) be denoted R. Then R and KJJ RÂ± are isomorphic relaiel tions from X to Y. Pjroof. Since (i*/Ri,j) is the intersection of all extremal subobjects iel containing each (Ri,ji) and each (RÂ±,ji) < (L/ R iÂ»k) and. . . . iel (vJR i} k) ^_ (Rjj) and since j is an extremal monomorphism then iel

PAGE 82

76 (V*jRi.J) 1 (R,j). iel Since (U R >Â•<Â•) is the intersection of all subobjects which "coniel tain" each (R Â± ,J i ) then (VjR.k) <_ (V*JR ,j). Since j is an extremal iel iel iuonomorphism and (R,j) is the intersection of all extremal subobjects which "contain" (IjK.k) then (R,j) < (\*)K ,j). Thus iel iel (R,j) = (I^JR ,j).. iel 5.4. Remark . Notice that by the definition of relation theoretic union, if (R,Â»j,)) ( R 2 >j ;) )j an d (S,k) are relations from X to Y and if k) and ( W (S)k) ' then ( R !^ R 2 Â»^ 1 (S ' k) (cf ' 1 ' 19) " 5.5. Proposition . Let (R ,j ), (Rg.jg), (S ,k ) and (S 2> k 2 ) be relations from X to Y. If (R^j^ < (R 2> J 2 ) and (S^k^ <_ (S^k,,) then CRjU/Sj.j) < (R 2 ^S 2 ,k). Proof, (R.,^) < (R 2 ,j 2 ) < (R 2 'o> s 2 > k > and (Sj.kj) 1 ( s 2 ' k 2 ) ( R 2 ^ ,S 2' k) whence (R.^JS^j) < (R^S^k) (5. A) 5.6. Rema rk. The following proposition can be strengthened with the ad-ditioiii'l hypothesis that the category^ has finite coproducts (5.34); however, it is included here because it is of interest in its own right 5.7. P roposition . Let (R 3 j) be a relation from X to Y and let (S,k) and (T,m) be relations from Y to Z. Then ((RoS) [*J (RoT) , g) < (Ro(S \*)T) , g ) , Proof . Consider the following commutative diagrams. S .**' *Y>Z

PAGE 83

77 RxZÂ»(RxZ)nXx(SV*/T) >5> ,\ X*(St*jT) V>l x x a -> (XxY)xZ & Xx(YxZ) RxZ Â»RxZrjXxS Â»XxS Â»J x lj l x xk -*> (XxY)xZ Xx(YxZ) -*> XxYxZ RxZ >V rxzAxxt >y XxT Â»jxl l x x m *Â• (XxY)xZ ^3 * Xx(YxZ) -* XxYxZ RxZ AXx(Sli,'T) >* Yl -v XxYxZ * XxZ "^ Ro(SV*)T) RxZAXxS *-*Â• xxyxz
Â— *"XxZ RoS

PAGE 84

78 RxZflXxT VÂ»*XxYxZ RoT <7T l' Tr 3 > ~>XxZ RoS RoT >XxZ Since (S,k) < (S^T,a) and (T,m) <_ (Sl^T.a) it readily follows that ((RxZ)A(XxS), Y? ) Â± ( (R*Z) C\ (Xx (S \^J T) ) , Yl ) and that ((RxZ) A(XxT),y 3 ) _<_ ((RxZ)0(Xx(S(*;T)) j y 1 ). Thus there exist morphisms c.i and Â£ such that YjCj = Y 2 anc * Y 1^2 = Y 3" ^ encc Y 1 C 1 = <7T iÂ»^3 > Y 2 and Yl^2 = T3 Â• But (RoS,g 2 ) i s the intersection of all extremal subobjects through which Y2 ^ actors and since ^tF, ,tt 3 >y = <^ -i Â» tt 3 > Yi Â£ Â»i T l5ii we have (RoS,g 2 ) <_ (Ro (S [*J T) , 6, ) Â• And since (IloT.B,) is the intersection of all extremal subobjects through which y 3 factors and since < ^ 1 ^ 3 > V 3 = '~i ! ^3 > ' r i^2 = ^l" l 1^2 it: follows Lhat (RoT.3 3 ) <_ (RoC-S^T),^). Whence ((RoS) V*>(RoT) ,g) < (Ro(S V*/T) ,3 2 ) (5.4). 5.8. ^^opo_sition. Let (T,m) be a relation from Y to Z and let (R,j) and (S,k) be relations from X to Y. Then ((RoTH*KSoT),3) < ((R^S)oT,3). Proof. Analogous to the proof of 5.7.

PAGE 85

79 5.9. Lemma . Let {(Rj^ji): iel} be a family of relations from X to Y. Then ((^Ri)" 1 ^*) = (^("i) _1 . k )iel iel Proof . Consider the following commutative diagram. iel *< a.* iel -1 Since (R." 1 ,!.*) is the intersection of all extremal subobiects l ' J i through which j . factors it follows that Thus iel (l^CRi -1 )^) < (((^Rj.)" 1 ^*) (5.4) Now (R i ,j i ) <_ (K*J(R Â± 1 ),1 k) since iel :ir ,ir,>" i kX.*T. = < tt .-, , tt , > -1 j ,*t . = i.. Thus 21 ii 2' 1 J ii J i iel ' iel from v;hich it follows that iel ' iel whence iel iel

PAGE 86

80 5.10. Corollary. Let ((R.pj.^): iel} be a family of symmetric relations on X. Then {Â£J R^ is a symmetric relation on X. iel Proof . It is clear that for each iel, (R i ,J i ) = (Ri~ 1 ,j i *) (1.13). Thus (l*jRi,j) = (^(Rr 1 )^) < ((Vi/Ri)1 ^*) (5.9). id iel iel 5.11. Pro positio n. If (R,j) is a reflexive relation on X and (S,k) is any relation on X, then (Ri^jS,m) is reflexive on X. Proof. Since (R,i) is reflexive, (A i ) < (R,j). Thus A A (A x ,i x ) < (R,j) < (R(*JS,m) hence (R^IS,m) is reflexive on X. 5.12. D efinitio n. Let (R,j) be a relation on X. Consider the relation (Rl*>R -1 ,j#). *XxX Let (t#,x) be the epi-extremal mono factorization of 'i,j#. The domain of x (codomain of x#) shall be denoted by X R[*)R l Â»Â» T# X R^" J# R" -> X>
PAGE 87

81 5.13. Examples . In the category Set , X p = (R(*Jr 1 )X = XRU RX. That is X_ = {xeX: there exists yeX such that (x,y)eR or (y,x)?R}. In the category Top -. , X R is the same set as in Set endowed with the subspace topology determined by the topology of X. 5.14. Pro position . Let (R,j) be a relation on X and let (RX,j^) and (R^Xjjo) be the images of tTjJ and tTj j * respectively. Then (RX^R _1 X,a) < (X R ,x) = ((Rl*^R -1 )X,x). Proof. Consider the following commutative diagram. R Â— > RX R Rl*JR-l Â» =*x Since: j = TTjjiUj, = x t #^r = 3i T i and (^.jj) is the intersection of all extremal subobjects through which irjj factors then (RX.j,) < (X R ,x). Similarly, it can be shown that (R _1 X,J 3 ) <_ (X R ,x), whence (RXj^J R _1 X,a) <_ (X R ,x) (5. A). 5.15. Proposition . If (R,j) is a relation on X then (Rl*jR~ ! ,j#) is symmetric on X and (X R ,x) = (X(R \*J R _1 ) , j 2 ) . Proof. ((R^IT 1 )" 1 ,://*) = (^"^(R" 1 )" 1 ^) = (R _1 (^R,J^) (5.9 and 1.11). Thus (Rvi'R _1 ,j//) is symmetric so that (x R ,x) ((RVSJR-^X.x) = (x(R^R _1 ),j 2 ) (4.6). 5.16. Proposition. Let (R,j) be a relation en X and let (A v ,i ) be the a r x R diagonal of X^X^ Then (A x ,(x x x) i x ) = (& x C\ (\*\) ,?) where R X

PAGE 88

82 (AA(X *X ),p) is the intersection of the diagonal (A ,i ) of X*X with A K K XX Proof . Consider the following commutative diagram. TTj* IT o * 7^X R 51X Observe that ^(x x x)i X = Wi* 1 * = X^ 2 *i X = ^ 2 (x x x)i X Â• Thus, K K R R since i x is the equalizer of TTj and tt 2 , there exists a unique morphism Â• so that i x ^ = (x*x)i x R Thus, since (A x H Xr x Xr) p) is the intersection of (A x ,i x ) and (Xn>
PAGE 89

83 quasi reflexive if and only if (A , (x x x)i ) Â± (R.j)That is, (R,j) is X R X R quasi-reflexive provided that there exists a morphism X such that the following diagram commutes. 5.18. Proposition. If (R,j) is a reflexive relation on X, then (R,j) is quasi -reflexive on X. Proof . If (R,j) is reflexive then (R(*jR -1 ,j#) is reflexive (5.11); hence, tt ^ j // is a retraction (1.24). Thus Tr,j// is an epimorphism; so that if (x#,x) is the epi-extremal mono factorization of ttjJ//, ^jj// = X T # so that x is an epimorphism as v;ell as an extremal monomorphism. Hence >; is an isomorphism (0.17). Thus (X R ,x) = (X,lÂ„) whence (A ,i ) = (A ,i ). K R Thus, since (A ,i ) <^ (R.j), (R,j) is quasi-reflexive. 5.19. Pro position . Let (R,j) he a relation on X. Then (R,j) Â£_ (A ,i ) if A A. and only if (R,j) < (A ,(x x x)i v )Â• R R Proof . Suppose that (R,j) < (A ,i ). Then there exists a morphism a such A A that j = ij^ci. Thus tt-^j = ff^i^a = i^^X 01 = ^2^ ' wnence ^Â• i <1T 2Â» Tr l > -3 = ^2^ = ^iJ ~ 7I 2 <1 ' Z'^^J ' Consequently/ the unique epi-extremal mono factorization of j is (1 ,j), and (R,j) = (R _1 ,j*) (3. 12) . But, since (R,j) = (R -1 ,j*), then (R^j R" 1 , j#) = (R,j), and (X ,x) '((R(*>'R _1 )X,x) = (RX,j.). Thus, since (RX,j.) = (X . X ) = (XR,j.) it follows that (R.j) 1 (X, xX ,x*X) (3.4). i X R,

PAGE 90

84 However, it has been shown that (A Y , (x x x)i Y ) and (i v H (Xt^X,,) ,p) A R A R A K. K are isomorphic relations on X (5.16). So that since (R,j) < (A ,i ) and ""XX (R,j) _< ((X R xX R ),x> (XxXxX,^ 1 ,u 2 ,Tf 3 ) > (XxX j tt 1 ,-t 2 ), ((XxX)xX,fr 1 ,fi z ), (Xx(XxX) , if x ,t 2 ) , (RxX,^,^) and (X x A x , Pl *,p 2 *). R Consider the following commutative diagram. 3*1, RxX >*-> (XxX) xX RxXfiXxA., V>V R ->XxX **Â»Â• RoA... *R Next it will be shown that y = <7r iÂ»' ir 3 > Y'

PAGE 91

85 Ti 1 <71 1 ,TT 2 >Y = ^Y = ^1 < ' ,i l , TT 3 > YTr 2 < : TT 1 /Tf 2 >Y = v 2 y = ^ 2 2( 1 X x (x x x)ix > X 2 = ^1*2 ( 1 X x (x x X>ix > A 2 = R R * (X x x)iv P 9 *A = X^!*iv P ? % = X^o*t. P,*A 2 = 1 A R 2 z R R ^V^ D P 2* X 2 = M* x X)i x P 2 * X 2 "-= u 2^2(3 X x (x x X)i Xp ) A 2 = tt 3 2 (1 x( x x x )i )\ 2 u 3 y = TT 2 Y. R Thus by the definition of product Y = YNov; consider Y Â• It will be shown that <7Ti,tt 2 >y = jPl^lT ': < ^i>^2 >Y = ^i Y = ^i e i (J xl )^i Â• Thus ,T i <: ^i Â»^2 > 'Y = 1r l^l(j x l)^l = MJPlM 7: 2 Y = if 2 Y = ^2 Q i (j x l )^i= TTofl i (j x 1)a 1 = 7T 2 jp i^j . Hence, by the definition of product, y = jpjXj. Since (RoA v ,ct) is the intersection of all extremal subobjects X R through which <^i > ^3 >Y factors and Y <^\ , ^2 > Y = JPl^l ^t follows that (RoA v ,a) < (R,j). X R For the reverse inequality consider (R^*/R ,j#). Since (Rt*/R -1 ,j# ) is symmetric, ((Rl*/R -1 )X,Ji#) and (X(Rl*J R" 1 ) . J2#) are isomorphic extremal subobjects of X (5.15). V R X X X Â— xCrv^'r" 1 ) * "siJ V R Thus if (Â£#,x#) is the epi-extremal mono factorization of tt _jj#, there exists an isomorphism Â£, such that x ,Â£ > = 5#Â« So in particular, tt 2 jA r tt 2 j = x#5#X R X?C#A R .

PAGE 92

86 Let a be the isomorphism for which i Y a = <1 ,1 > (1.21). 41 x k' X R It next will be shown that the following equality holds: so that the following diagram is commutative. 01 < J, 7T 2J >
= ff x (l x <(x*x)i x ) R K Pl* = TTij. u 2 2 (l x >: (x x x)i x ) = it^ 2 (l x x( x x x )i x ) a5C#A R > tt : (X x x)i x P 2 * = R XPii v p 2 * = XPii Y Â°C5//A n = X R K "R Tf 3 2 (l X( X X X )Â± x ) = Tl 2 ff 2 (l x x( X X X U )<7T 1 J,05C#X R > = R ** TT 2 ( X x X )i x P 2 * R XP2 i XR P2*^lJ'Â°^#V xP2 <1 x R > 1 x R > ^ #x : x i Y 5?#x p = xC5#XÂ„ = tt 2 j. *X R ""R ir 1 e 1 = Tr 1 fr 1 = TTjj " 2 S 1 TT 2 ff l < J' 1T 2^ > = 1! 2-i Tf 3 1 = Tr 2 = fr 2 j.

PAGE 93

87 Thus by the definition of pi-oduct: < -"]J, TT 2JÂ» Tr 2J > = 1 < JÂ» T7 2J > = 2( 1 x x(x> = 0i(j x D < l u ,T'2i > (1-3D. Hence by the definition of intersection there exists a unique morphism Z such that yZ = <~n\ j ,^23 > 7T 2J > Â• Thus Yl = < Tfl,iJ3> = < TT 1 j , TT 2 j > = j Â• Hence j = agS; and consequently (R,j) ^_ (RoA^ ,a). R 5.21. Corol lary . Let (R,j) be a relation on X. Then R _1 , AÂ„ oR -1 and R oA are isomorphic relations on X. X R Proof . Recall that since (R VÂ£J R" 1 ,j#) is symmetric (5.15), (X R -i ,x*) and ((R {*) R -1 )X, j, //) are isomorphic as extremal subobjects of X (5.15 and 5.9). Also (X ,x) and ( (Rli!^ R" 1 )X, j #) are isomorphic as extremal subobjects of X (5.15) whence, (Xgl*XÂ£i,X* x X*) = (\*\,X*x) and hence (A _j , (x* x X*) \l ) and R "R (A Â»(x x x)iv )Â• But ^y tne theorem X R ' X R Consequently, (R l ,3*) = (A v OR" 1 , 3*) 5 (R _1 oA v ,a*). A R _1 A R _1 (R ! ,j*) = (A y oR,a#) E (R ^Ay ,ai7) . A R A R 5.22. Definiti on. Let R be a relation from X to Y. Then R is said to be di functional if and only if Ro(R _1 oR) <_ R and (RoR _1 )oR < R. The term difunctional relation is due to Riguet Â£22 J. 5.23. Proposi t ion . Let R be a relation from X to Y. Then R is difunctional if and only if R is difunctional. Proof . If (R,j) is difunctional then since (Ro (R~ 1 oR) , k, ) <_ (R,j) we have ((R~ 1 oR)oR~ 1 ,k 1 #) = ((R~ 1 oR) -1 oR~ 1 ,1c ) 5 ((Ro(R~ 1 oR))" 1 > k^*)<.(BT 1 ,j)

PAGE 94

S8 This fellows from 1.38, 1.11, and 1.12. Also since ( (RoR~ * )oR,k 2 ) <_ (R,j) it follows thai (R-MRoR -1 ),^//) = (R'ioCRoR -1 ) -1 ,^) = (((RoR" 1 )oR)~ 1 ) k 2 *) l(R _1 ,j*). Thus (R _1 ,j*) is di functional. If (R _1 ,j*) is difunctional than since ((R _1 ) _1 ,j#) = (R,j) (1.11) and since, applying the above to (R _1 ,j*), ( (R" 1 )~ 1 , j //) is difunctional it follows that (R,j) is difunctional. 5.24. Proposition. If R is a relation on X then R is a quasi-equivalence (3.2) if and only if R is quasi-reflexive and difunctional. Proof. If (R,j) is a quasi-equivalence then (R,j) is symmetric hence (R(^R _1 ,j#) E (R,j). Thus (\,x) = (RX.jj) s (XR,j 2 ) (5.15 and 4.6) from which it fellows that (A , (x^x)^ ) = (\rÂ» (J2 x J2) jt y R ) R R Since (R,j) is a quasi-equivalence then (R,^) is an equivalence relation on XR (3.10) so (A ,i ) < (R,i>). Hence XK XK (A ^2 x ^2^vp^ Â— ^>J) sc that (R,j) is quasi-reflexive on X. Since XK AK. (R,j) is symmetric and transitive then (Ro(R~ 1 oR),k 1 ) Â£ (Ro(RoR),k 1 ) < (RoR.j') <_ (R,j). Similarly, ((RoR~ 1 )oR,k 2 ) <_ ((RoR)oR,k 2 ) <_ (RoR,j') Â£ (R,j). Hence (R,j) is difunctional. Conversely if (R,j) is quasi-reflexive and difunctional then (A Y ,(x x x)i ) 1 (R,j) so that (A >(x*x)i Y ) 1 (R _1 ,j*> (1.16 and 1.12). A R A R A R A R Thus (RcR,j') < (Ro(A v oR),k) < (Re (R _1 oR) ,k, ) < (R,j) and X R t. (R'i.j*) < (A o(R -1 oA v ),k) < (Ro(R" 1 oR),k 1 ) < (R,j) (5.20 and 1.30). _ X R X R _ Thus (R,j) is both transitive and symmetric hence a quasi-equivalence. 5.25. Proposition. Let R be a relation on X, Then R is an equivalence relation _f and only if R is reflexive and difunctional.

PAGE 95

89 Froof . If (R,j) is an equivalence relation on X then (R.j) is reflexive and a quasi-equivalence on X. Thus (R,j) is difunctional (5.18 and 5.24) Conversely if (R,j) is reflexive and difunctional then (R,j) is quasi-reflexive and difunctional (5.18) hence (R,j) is a quasi-equivalence (5.24). Since (R,j) is also reflexive it must be an equivalence relation. 5.26. Theorem . Let (R,j) be a relation from X to Y and (RX,^) and (XR,j 2 ) be the usual images (3.4). Then R, RoA^.Â„ , and A R Â„oR are isomorphic relations from X to Y (cf. 5.20 and 1.31). Proof. Consider the following commutative diagrams. A x^~ , I I Â• ^-RX A >*> RX -*XxX t [Ji x Jl * -*RXxRX ~ Pi P 2 Pi* p 2 * Xx Ji ^RX Observe that Pi (j !>
PAGE 96

90 Thus: (A RX oR,k 2 ) <_ (A^RJ') e (R,j) (1.30 and 1.31). To see that (R,j) Â£ (A Rx oR,k ) consider the following commutative diagram. ^Â•XxX^Y *i> XxY ixxy Recall that (A RY ,i KY ) = (RX, <1 RY , 1 PY >) (1.21) hence tl J RX'"RX RX'^RX' nere exiscs an isomorphism a such that ^ryjIr-;^ = ^px Consider also the products (A^xYjp ^ ,o 2 ) , (X*R,ff j ,tt 2 ) , and ((XxX)xY,u 1 ,u 2 ). It will next be shown that the following diagram is commutative. Iwi i RX xl Y (XxX) xY ^l^>^l3,^23> = 02
->XxXxY \ <7r iJ'V X Xx R Â»-* X.x(XxY) l x-i X J v. 1 01 ( ( j i x j i ) i f0( x l y ) P i if i ( ( j i x j 1 ) Irxx l y ) PlOl*Jl)iRXP"l =: JiPi* i RXfV^i^ 2 -i > = JlPl^iRX^l = JiPl* <1 RXÂ» 1 RX >T l Ji T i = M.1-

PAGE 97

91 TT 2 l((Ji x .ii)i RX xl Y ) = P2 Tf l((Jl x 3l) i R X xl Y ) = P2(Jl X Jl) i R,XPl = J]P2* i RX OT l = 3lP2 , ' c<1 RX ! 1 RX >T l J1 T 1 = ^lJT f 30l((Jl x Jl) i RX xl Y) = ^2((Jl x Jl) i RX xl Y ) = l Y P 2 = TT 2 j. ^ 1 ? (l x Xj) = ff 1 (l x Xj)< 1 r 1 j,l R > = TT 1 = TTjj. lf 2 9 ? (l x ' = ir 1 TT 2 (l x XJ)<1T 1 j,l R > = TTlJTf2 <7T l3> 1 R > = *l 3 Â• Tf 3 2 (l x Xj) = 7T 2 Tf 2 (l x Xj) = TT 2 j tt 2 <^ \ j , 1 R > = 7T, j . if 1 2 = ff 1 = ir^i. TT 2 e 2 <7T 1 j , j> = TT 1 Tf 2 <7i 1 j,j> = TTjj. Tf 3 2 <7T 1 j,j> = T7 2 TT 2 = 7I 2 j. Consider the following commutative diagram. (Ji x Ji)i-RX xl Y A RX xY ># > t^ x X) *Y Aj^xYHXxR >?V -#Â• XxXxY 1 ^Xx(XxY) * Wfc A RX cP X*Y By the definition of intersection, there exists a unique morphism Â£ from R to (A / Â•

PAGE 98

92 Thus Y? = < ifi,Tf 3 >< 7T 1 j,T 1 j,^2J > = <1l iJÂ» 7T 2J > = 3Hence k 2 T = j; that is, (R,j) <_ (A Rx oR,k 2 ). Hence (R,j) e (A RX oR,k 2 ). Similarly it can he shown that (R,j) = (RoA^.kj). 5.27. Theorem. Let (R,j) be a relation from X to Y. Then ^RX'^VV 1 ^) 1 (RoR _1 ,j#) and (A XR , (J 2 x J 2 )1 x r) 1 (R^oR.j'). Proof. Consider the following products: (XxX,pi ,p 2 ) , (BXxRX.pj *,p 2 *) , (XxY.ttj ,ir 2 ), (RxX,p 15 p 2 ), (XxR~ 1 ,p 1 ,p 2 ) J ((XxY)xX,ff 1 ,ff 2 ) , (Xx(YxX) ,ti j * ,tt 2 *) and (XxYxX,tt, ,tt 2 ,tt 3 ) . Also consider the following commutative diagrams. R HT l RX Â»-> XxY Jl -* X R >* Next, the following diagram will be shown to be commutative, jxl v ^r rxx v* <3 R^1J > V (XxY)xX -"''lJ' 1 "V> RoR

PAGE 99

93 To see this, it need only be observed that 0l(jxl x ) = 02(l x xj 5 -)<^r 1 j,T>. Tf 1 G 1 (jXl x ) = Tr l Tf 1 (jxl x ) =TT 1 JP 1 <1 R ,1T 1 J> = TTjj ' t 2 l <: J xl X )<1 R' 1r lJ > = w 2 ff l ^J xl X^ <1 RÂ» ir lJ > = 7T 2JPl <1 R' 7T lJ > = v li ^3 G l ( J xl X )<1 R' TT lJ > = ft 2 ( J xl x :><1 R' 7T lJ > = ^2 <1 R' 7r lJ > = lr lJTT,e 2 (l x x j") TT ? Q 2 (l X >< j") < T 1 J,T> 7T i* A ^ 1 x x J" A '^ <7T iJ' T> = Pl
= ^lJTr 1 ir 2 *(l x xj*)< 7 r 1 j,T> = TTj j*P 2 T> = *i\$*i = 7T l <7T 2' Tr ] > J = ^?J ' ^3 2^ 1 X X J*^ = ^2 7T 2' : ^ 1 X X J' ! '^ = ^2 <7I 2' TI 1 > J = '"'l J Â• Thus the diagram is commutative and 1 (jxl x ) 7T 1 J> = 9 2 (1 X XJ*) = . Hence by the definition of intersection there exists a morphism I such that yZ = . Clearly YÂ£ = <'^ 1 .i Â»' ,r iJ > thus = j#x#Z. It next will be shown that if a is that isomorphism for which = ijivO then (Â° T i > (j i x j i)1ry) is the epi-extremal mono factorization of R Â— ->' X X X 9 (j i x j l)i-RX RX Â»-VsA RX P2 r -Jl x Ji) i RX CT l = JlP2" i R>v OT l = 3iP 2 * <1 RX' 1 RX >T l = J 1 " L 1 = ir lJ

PAGE 100

94 Thus the diagram above commutes and (ox , , (j , xj , )i ) is the epi-extremal mono factorization of (0.18). Since (A_ v , (j , xj , )i ) Â± s the intersection of all extremal subobjects through which factors and since = j#x//E it follows that ^ A RX' d lX ^ ^"^RX^ Â— ^ RoR ~ 1 >i J '^ which was to be proved. The proof that (A XR) (J2 x J2)i XR ) 1 (R _1 oR,j*) is similar. 5.28. Theo rem. If (R,j) is a relation from X to Y then R is difunctional if and only if (RoR _1 )oR = R = Rc(R _1 oR). Proof . If R is difunctional then (R,j) = (RoA^kj) < (RoCR^oR),!^) <_ (R,j) (5.26, 5.27 and 5.22). Similarly. (R,j) = (A^oR,^) Â£ ((RoR^oR.k.,) <_ (R,j). The converse is immediate from the definition (5.22). 5.29. Remark . Let c be a locally small quasi-complete category having (finite) coproducts. It is noted in passing that if R 1 has arbitrary products; i.e., is complete, then Â£j i s also (finitely) cocompletc \9 \. Recall that the unique epi-extremal mono factorization of a morphism is obtained by taking the intersection of all extrema l subobj ects of the codcmain of the morphism through which the morphism factors (0.21). Also recall that if the intersection of all subobjects of the codomain of the morphism through which the morphism factors is taken, then the unique extremal epi-mono factorization is obtained (0.21)Finally recall that if {(A., a.): id} is a family of subobjects of a (^-object X, then the subobject (\J\.,s.) is obtained by taking the intersection of all subiel X objects of X which "contain" each (A., a.).

PAGE 101

95 Now consider the coproduct ( XL A^ ,y^) of the (finite) family iel {(A-,a-): id}. By the definition of coproduct there exists a unique mo rphism y such that pu= afor each Iel. Let A^ be the "inclusion" of (A^,a^) into (i_JA^,a). Again by the definition of coproduct there iel exists a unique morphism A such that Ay= A^ for each iel. Thus the following diagram commutes. -*-X A-: Note that a is a monomorphism. It will be shown that A is an extremal epimorphism. To see this, it will be shown that (U Aj,a) is the iel intersection of all subobjects of X through which y factors. To that end let (Z,g) be any subobject of X through which y factors; i.e., there is a morphism h such that y = gh. Then a^ = yyj_ = ghy^ hence each (A-^,a^) factors through (Z,g). Thus by the definition of union there exists a unique morphism Â£, such that gÂ£ = a. But this is precisely what is required of the intersection of all subobjects of X through which y factors . Now suppose that {(A^,a^): iel} is a (finite) family of relations from X to Y; i.e., each (A 3 -,a^) is an extremal subobject of X*Y. Consider (JLL &Â±,^i) and ( U Aiifl)Again, let y be th
PAGE 102

96 X\i. = X . for each iel, where \. is the inclusion of (A.,a.) into ( (J A .,Â£). Let (r,p) be the epi-extremal mono factorization of a. Recalls I ling Proposition 5.3 it follows that the domain of p (codomain of t) is ^A.. Thus the following diagram commutes, iel ^ XX. A. > XxY iel x u v * ^ A i iel t iel 5.30. Theorem . Let ^ have (finite) coproducts, let {(A^a^: iel} be a (finite) family of subobjects of C, and let f be a fe -morphism from C to D, As above, let (l^A^p) be the extremal mono part of the factoriiel zation of the unique morphism p from JJL.A^ to C with the property that iel Hjij = a^ for each iel. Let (f\E-,e) be the intersection of all extremal jeJ subobjects of D thiough which each fa^ factors. Let (Im(A^) ,p^) and (lw( [Â£) A.. ) .p) denote the extremal mono parts of the epi-extremal mono iel factorizations of fa^ and fp respectively. Finally let ( \*) Im(A^) ,p) iel be the intersection of all extremal subobjects of D through which each pfactors. Then (Im^A^.p) E (flE.,e) = ((*J Im(k L ) ,0) . iel jeJ iel Proof . If (E. ,e.) is an extremal subobject of D through which each fa^ factors, then since (Im(AjO ,0-^) is the intersection of all extremal subobjects through which fa^ factors it follows that (Im(A n .),P,0 < (E.,e-). Thus (Im(A n . ) ,o) < (HE-,e) for each iel. Hence 1 *Â• Â— J j x Â— . T .1 (i v ^Im(A.),p) < (C\E,,e). iel iej J

PAGE 103

97 However, since each fa Â± factors through p and since p is an extremal monomorphism it follows that (f\Ej,e) <_ ( \*J Im(A i ) ,|S ) . jeJ iel Consider the following commutative diagram. ">Â• ^Im(A Â± ) iel iel > D Note that (it, 6) is the epi-extremal mono f actcrization of fu so that (Im((^A i ),p) is the intersection of all extremal subobjects iel through which fjj factors. Let ty be that unique morphism such that ^Uj_ = ^-Â±^Â± fÂ° r each iel. Now pyy i fAjJi = Pi^i = Â£a Â± = fuy Â± . Thus by the definition of coprc-duct it follows that fy = p>. Hence fy factors through (l*J Im(A-|_) ,P) iel whence (Im( \*)A Â± ) , p) Â£ (l*J lm(.AÂ±) ,0) Â• iel iel Now fa-j^ = fuui = fpxpi = pfiy-jj hence fa i factors through p, whence (Hs^e) Â£ (Im( V*/ A Â± ) ,(5) . jeJ " iel Thus: (^ImCAijJ) E (f\E J5 e) E (Im(l&J A Â± ) ,j5) . iel jeJ iel

PAGE 104

93 5.31. Definition . A category is said to be (finitely) union distr ibut ive if the following properties hold: (i) if X and Y are any fa -objects and {(A^a^: iel} is a (finite) family of extremal subobjects of Y, then Xx((*jAjO and ^'J (XxA i ) are iel iel isomorphic relations from X to Y; (ii) if X is any fe -object, {(A-j^a^: iel} is a (finite) family of extremal subobjects of X, and (B,b) is an extremal subobject of X, then BA((^)A i ) and [*} (BftA-^) are isomorphic as extremal subobjects iel iel of X. 5.32. Remark . It can be shown in any quasi-complete category that (5Â» (XxAp S v ) are isomorphic relations iel veV (i,v)sI*V from X to Z. In particular if (R,j) is a relation from X to Y and (S,k) is a relation from Y to Z, then Ro(tj/S v ) and ^/(R s v) are isomorphic vev VÂ£ V relations fr:-, X to Z and i\^J?.j)cS and (*/ (R-joS) are isomorphic relaiel iel tions from X to Z.

PAGE 105

99 Proof . From the conditions on fi it is easy to see that: (((^R i )xZ)n(Xx(^S v )),y) = (((V^R Â± )xZ)n(V*)(XxS v )),9) iel veV . iel veV (^(((URi)xZ)n(XxS v )),Y) = (^)((^(Ri x Z))r\(XxS v ),y) veV iel Â„ veV iel (^(^Â»((RiXZ)n(XxS v ))),T) E ( ^*J ((R i xZ)n(XxSv)),y). veV iel (i,v)elxv Hence, from the theorem (with (R-^xZ) C\ (XxS v ) assuming the role of A Â± ) it follows that ((t*^R i )o(l*JS v ),a) = ( \*J (R i oS v ),a). iel veV (i,v)eIxV 5.35. Corollary . If P has finite coproducts and is finitely union distributive and if (R,j) is a quasi-equivalence on X then (Rl*jA x Â»p) is an equivalence relation on X. Proof . Clearly (R\*l hy,p) is both reflexive and symmetric (5.10 and 5.9) Since each of (R,j) and (A x ,i x ) is transitive (2.4 and 2.2) it follows that (R(*M x )o(Rl*M x ),p#) = ((RoR)l*J(RoA x )(*;(A x oR)C*;(A x oA x ),3) Â£ (Rl*jR^R^A x ,p) e (RV*jA x ,p); (5.34 and 1.31). Thus (RV^jA x ,p) is transitive and, consequently, is an equivalence relation. 5.36. Corollary . If JÂ§ has (finite) coproducts and is (finitely) union distributive and if (R,j) is a relation from X to Y and {(A^,a i ): iel} is a (finite) family of extremal subobjects of X, then [Â£) (A^R) and iel (C*jA-)R are isomorphic as extremal subobjects of Y. id Proof . Since (R f\ ( ( [*) A Â± ) xY) , y) = ( V*J (R ACA^Y) ) , y) the result follows iel iel from the theorem. 5.37. Corollary . If Â£ has (finite) coproducts and is (finitely) union distributive and if (R,j) is a relation from X to Y and {(B^,b^): iel} is a (finite) family of extremal subobjects of Y, then V*^(RB.) and iel R(l*^B.^) are isomorphic as extremal subobjects of X. iel

PAGE 106

100 Proof . immediate. 5.38. C orollary . If fÂ£ has (finite) coproducts and is (finitely) union distributive and if {(R.Â£Â»Ji): iel} is a (finite) family of relations from X to Y and if (A, a) Ls an extremal subobject of X then A(\j[jRj) and iel \*J (AR-j^) are isomorphic as extremal subobjects of Y. iel Proof. This result follows immediately from the theorem since ((l^Ri)n(AxY),Y) = (lol(Rin(AxY)), Y ). iel iel on 5.39. Cor ollary. If (J has (finite) coproducts and is (finitely) uni distributive and if ((R^j-^): iel} is a (finite) family of relations from X to Y and if (B,b) is an extremal subobject of X then (tÂ£jRj)3 and iel [Â£j (R-^B) are isomorphic as extremal subobjects of X. iel Proof. Immediate. 5.40. Remark . Without the extra conditions on Â£ ; i.e., only assuming that P is locally small and quasi-complete; it is possible to prove that [*J(A Â± R) < (l^A i )R and that l^J (AR Â± ) ^Ad*^). iel iel iel iel 5.41. R emar k. Recall that if g is a Â£ -morphism from X to Y where jg is locally small and quasi-complete then the intersection of all subobjects of Y through which g factors, (f| Ej,e), yields the extremal jeJ epi-mono factorization of g; i.e., there exists an extremal epimorphism h such that e = eh. Let C\ Zi be denoted Slm(X) and be called the subj jeJ image of g. (Recall that the image comes from the epi-extremal mono factorization of g.)

PAGE 107

101 f Slm(X) >-> Y -& Im(X) Now let {(A^,a-j^: iel} be a (finite) family of subobjects of a Â£ -object C and let f be a fe -morphism from C to D. Then there exists a unique morphism u from the coproduct (_ULA-Â£,Uj_) to C such that iel yuj = a^ for each iel. Let {(E-,e-): jej} be the family of all subobjects of D through which each fai factors. Let (0,6) and (0^,6^) be the extremal epi-mono factorization of u and fa^ respectively. Recall that ( [*} Slm(A^) , I) is the intersection of all subobjects of D through which iel each 5j factors. Let (o*,6*) be the extremal epi-mono factorization of fo. 5.42. Theore m. If fÂ£ has (fiiiite) coproducts then, using the notation above, (^ISImCA^), O E-: and SIm(LJA n -) are isomorphic as subobjects iel jeJ iel of D. Proof. Let if be the unique morphism from JJ.A^ to [^J Slm(A^) such that iel iel ^i ~ >l 'i a Â± ^ or cacn i^I where y^ is the "inclusion" of Slm(Aj) into USIm(A Â± ). iel Thus (as is easily seen) the following diagram commutes. U 5Im(A Â± ) iel "\ >Â» Sim ((J A ; ) ' iel

PAGE 108

102 Since fa i factors through I for each id, it follows that (O^.e) < (U SIm(A i ),5:). jej iel Nov; if (Â£. ,e.) is a subobject of D through which each fa. factors then since (Slm^)^) is the intersectio of all subobjects of D through which fa Â± factors it follows that (SIm(A. ) , 6 . ) <_ (E.,e.) thus (USIm(A.),Z) < (E e .) for each jej. Hence ( \J Sim (A. ) , E) Â£(f\E.,e). iel J J iei X jej 3 Since each fa-j factors through 6* then (PiE:,e) < (SIra( U A^) , 6*) jej J iel Since Z'-p = fy and &*(o*o) = fy is the extremal epi-mono factorization of fu, it follows that (SIm( U A-i_),6*) <_ ( U SIm(A-;) , E) . Thus iel iel OJ SlmCAi) ,E) = (AE js e) E (SIm( (J A Â± ) , 6*) . iel jeJ iel 5.43. Remark. Theorems 5.30 and 5.42 show that the (sub) image of a union is the union of the (sub) images; hence the epi-extremal mono factorization and the extremal epimono factorization properties respect unions in a proper manner.

PAGE 109

SECTION 6. RECTANGULAR RELATIONS 6.0. S

PAGE 110

104 Since there exists a morphism A for which 4>Â„A = a and a is a monomorphism, A must be a monomorphism. But 4> Y = Â„X<}). = ^yl^Thus it follows that A<}> A = 1. (6.0. i) so that A is a retraction. Hence A is an isomorphism (0.4 ) whence (A, a) = (\$,<|>y). 6.3. Proposition . Let (R,j) be a relation from X to Y and let Z be any K -object. Then Ro\$ and \$ are isomorphic relations from X to Z; and, \$oR and 4> are isomorphic relations from Z to Y. Proof. From 6.0.ii, (XxÂ«J>, l x ><<{)y xZ ) and (.^Axx(YxZ)^ are isomorphic as extremal subobjects of X X (Y>,a) = (Â£, v Â„) . Similarly it can be shown that \$oR and A X Z \$ are isomorphic relations from Z to Y. 6.4. Corollar y. If X is any ^ -object then (\$,4>xx:<) is a quasi-equivalence on X.

PAGE 111

105 Proof . By the proposition (\$>4>v x y ) = (^^''f'xxx^ hence transitivity is obtained. It is also clear that the following diagram commutes. *X*X X*X Since \$ is an extremal monomorphism and 1, is an epimorphism X*>A Â» it follows from the uniqueness of the epi-extremal mono factorization that (\$,0 V vv) (^ _1 >4>vÂ„v*) 5 hence symmetry is obtained. A X A AXA 6.5. Definition . Let (R,j) be a relation from X to Y. Then (R,j) is said to be rectangular if and only if there exist extremal subcbjects (A, a) and (B,b) of X and Y respectively such that (R,j) and (A*B,axb) are isomorphic relations from X to Y. 6.6. Remark . Since (\$x\$ )( f x^) and (y y) are isomorphic as extremal subcbjects of X*Y (6. 0. i and 6 .0. ii) it f ollows that (^^xxY^ ^ s a rec ~ tangular relation. 6.7. P roposition . Let (R,j) be a rectangular relation from X to Y and let (RY,j,) and (XR.j ) be the usual images of tt j and tt j respectively. Then R and RY*XR are isomorphic relations from X to Y. Proof. Since (R,j) is rectangular there exist extremal subcbjects (A, a) and (B,b) of X and Y respectively such that (R,j) = (A*B,axb). hence there exists an isomorphism a such that the following diagram commutes.

PAGE 112

106 -*XxY ->X * AxB -y> A Pi Since Pj is an epimorphism (6.0.iii),' pjO must he an epimorphism, and by the uniqueness of the epi-extremal mono factorization it follows that (A, a) I (RY.jj). Similarly (B,b) = (XR,.j ? ). Thus (AxB,axb) = (R,j) = (RYxXR,j 1 xj 2 ). 6.8. Corollary . Let (R,j) be a relation from X to Y. Then (R,j) is rectangular if and only if (R -1 ,j*) is rectangular. Proof . If (R,j) is rectangular then (R,j) = (RYxXR, j , xj 2 ) (6.7). It is immediate that the following diagram commutes. jl x j2 < ft2Â» TT l > RYxXR Vt -*Â» XxY V5>Â— Â— Â— Â— -**YxX *^ J2 x .3i XRxRY Again, by the uniqueness of the epi-extremal mono factorization property it follows that XRxRY and R -1 are isomorphic relations; hence (R ,j*) is rectangular. If (R -1 ,j*) is rectangular then by the above , ( (R _1 )"" 1 , j#) is

PAGE 113

107 rectangular; but ((R^)" 1 ^/) = (R.J) (1.13). Thus (R,j) is rectangular, 6.9. Proposition . Let (R,j) be a rectangular relation from X to Y and let (C,c) be an extremal subobject of X. Then ' (XR,j 2 ) if (CPlRY.Y) t (*,*x) , ( Y ) if (CrtRY.y) = (*,+ X ). (CR,k) Proof . Since (XR,j 2 ) 1 (Y.ly) it is clear that (XRrtY.y) = (XR,j 2 )Hence it follows that (Rn(CxY),Y ) E ((RYxXR)A(CxY), Yl ) = ( (RYA C) x(XR AY) , T? ) = ((RYrtC)*XR,3) (0. 8 ). Thus there exists an isomorphism a such that the following diagram commutes. (RYAC)xXR Â» -> Y Let (C,I) be the epi-extremal mono factorization of tt 2 p. Since to is an epimorphism and k is an extremal monomorphisia it follows by the uniqueness of the epi-extremal mono factorization that (Z,Z) = (CR,k) Â• But if ((RYnc)xXR,B) t (*,xxY> then the followln S diagram commutes. (RYAC)xXR Â»8 XxY I XR > v J2

PAGE 114

108 Since P 2 is an epimorphism (6.0. iii) it follows that (XR,J 2 ) = (Z,Z) E (CR,k). If ((RYr\C)xXR,g) = (^.^xxy) then there exists an isomorphism X such that the following diagram commutes. \$ Â» *> y Thus by the uniqueness of the epi-extremal mono factorization it follows that (Â£,Â„) if (XRrtA.v) = (*,* y ) Proof . The proof is analogous to that of Proposition 6.9. 6.11. P roposition . Let (R,j) be a rectangular relation from X to Y and let (S,k) be a relation from Y to Z. Then RoS <_ RYx(XR)S. Proof . It is easy to see that the following objects are isomorphic as extremal subobjects of X*Y*Z: ( (RYxXR)xZ)A(X*S) , ((RYxXR) *Z) A(RYxS) , and RYx((XRxZ)A S) . Thus there exists an isomorphism o such that the following diagram commutes. RYx((XRxZ)riS) V*~ -V> ((RYxXR)xZ)A(XxS) " Â— *" XxZ (RYxXR)oS

PAGE 115

109 Recall that (R,j) = (RY*XR, j ^j 2 ) (6.7); hence (RYxXR)oS and RoS ire isomorphic relations. Consider the following commutative diagrams. (XRxZ)nS >VP2 -* YxZ -*Z '(XR)S T ' Â©2(Jl xl Y> RYx(YxZ) > XxYxZ l RY x6j RYx (XR) S >XxZ Let (x,p) be the epi-extremal mono factorization of 0 2 (j 1 xi YxZ )(i RY x6 1 ) = ya. Thus since (P,p) is the intersection of all extremal subobjects through which 7a factors and since o is an isomorphism it follows that ((RoS), a) H ((RYxXR)oS,5) = (P,p) < (RY*(XR)S, j^Oj) . 6.12. Pro position . Let (R,j) be a relation from X to Y and let (S,k) be a rectangular relation from Y to Z. Then RoS <_R(SZ)xYS. Proof. Since (S,k) = (SZxYS .k^k^) (6.7) the result follows from argume nts analogous to those in the proof of Proposition 6.11 6.13. Lemma . Let (R^j^ = (AxB^axb^ be a rectangular relation from X to Y and let (R 2 ,j 2 ) E (R 2 x C,b 2 *c) be a rectangular relation from Y to Z,

PAGE 116

110 Then (R l0 R 2 ,j') E ((Ax3 1 )o(B 2 xC),j#) = |(AxC,ax c ) if (Bif\B 2 ,b) t (\$, \( XXZ ) if (Bif\B 2) b) = (\$, Y ) And in either case (RioR2,j') is rectangular. Proo f. It is straightforward (but tedious) to show that: ((AxB 1 )xZ)r>(Xx(B 2 xC)) = (Ax(BixZ))n(Xx(B 2 xC)) = (Ax (BixZ) (\ (Ax (B 2 xZ) ) = Ax((B 1 xZ)f\(B 2 xC)) E Ax((B 1 AB 2 )xC) E Ax(B } A B 2 )xC. If (BjHBpjb) Â£ ( ((AxBi)xZ)Pi (Xx(B2 x C)) * *> Ax(Bir\B2)xC "* => XxYxZ XxJ T# (AxB 1 )c(B 2 xC) It is easy to prove that is an epimorphisw since Ax(BjAB 2 )xC and (AxC)x (B 1 As 2 ) are isomorphic in a canonical way and 6.0.iii holds. Thus since a is an isomorphism it follows by the uniqueness of the epi-extremal mono factorization that Â«AxB 1 )o(B 2 xC),j#) e (AxC,ax c ). Hence (R 1 oR 2 ,j') = (AxC,ax c ) ; so it is rectangular. If (BiHB^b) = (*, VvVv Â„) are isomorphic as extremal subobjects hence by the uniqueness of the epi-extremal mono factorization it follows that ((A*Bi )o(B 2 *C) , j#) = ( Â„) and is rectangular (6.6).

PAGE 117

Hi 6.14. R emar k. As has been noted in Section 1 (1.37) the composition of relations is not necessarily associative. The examples 1.34 1.36 are in the category Top which satisfies the conditions of 6.0. The next theorem shows that the composition of rectangular relations jls associative. Hence, in particular, in Top the composition of rectangular relations is associative. 6.15. Theorem . Let (R,j) be a rectangular relation from X to Y, let (S,k) be a rectangular relation from Y to Z and let (T,m) be a rectangular relation from Z to W. Then Ro (SoT) and (RoS)oT are isomorphic relations from X to W. Proof . Since each of (R,j), (S,k) and (T,m) is rectangular there exist extremal subobjects of X, Y, Z and W such that (R,j) = (Ai *A 2 ,ai*a 2 ) , (S,k) = (B 1 xB 2 ,b 1 xb 2 ) and (T,m) E (Cj xC 2 ,c\ xc 2 ) Â• Then: f(B 1 xC 2 ,b 1 xc 2 ) if (B 2 r\C!,b) f (*,\$ ) (SoT,k#) E I V(\$,^ yxW ) if (B 2 r\C l5 b) E (*,o z ), and '(A 1 xB 2 ,a 1 xb 2 ) if (A 2 flB 1) a) t (\$, Y ) ,(*, XxZ ) if (A 2 AB 1>a ) E (4,* Y ). Thus there are two cases: 1) if (SoT,k#) = (B^xC2 ,bi x c 2 ) then as above it follows from 6.13 that t(AixC 2 , ai xc 2 ) if (A 2 ^B 1 ,a) i (\$,<(>Â„) ( *'*XxW> if C A 2ABi,a) = (*,* Y ). If (A 2 AB!,a) t (3>.^) Y ) then (RoS,j#) E (A 1 xB 2 ,a i xb 2 ) hence ((RoS)oT,g) e (A 1 xC 2 ,a 1 xc 2 ) since (B 2 ACi,b) t (.Â§,\$%)Â• If (A 2 HB l5 a) E (0,J) V ) then (RoS.i'O e (f Yv . 7 ) hence ((RoS)oT,B) 2 0.
PAGE 118

112 2) If (SoT,k#) E (*,Thus in any case (Ro(SoT),a) = ((RoS)oT.p). 6.16. Proposition . Let {(Ri,ji): iel} be a family of rectangular relations from X to Y. Then (ARj_,j) is a rectangular relation from X to Y. iel Proof. Each (Ri,Ji) is isomorphic to (Aj[xBi ,ai>'ai ) are isomorphic relations on X. a-,xa 2 < ^ 2 >"l > A^A 2 W ' > XxX Â» v*XxX

^Â»*. ^^ a 2 Xa l ** A 2 xA 2 "* Thus, since (R,j) is symmetric and (AixA 2 ,aixa2) f. (RÂ»j)> it fol-

PAGE 119

113 lows that (A 2 xAi,a 2 xa 1 ) = ( (A 1 xA 2 )" 1 , ( ai x a2 )*) Â± (R -1 ,j*) = (R,j) (1-13 and 1.12). 6.18. D efinition . Let (R,j) be a relation from X to Y and let (Ai,ai) and (A 2 ,a 2 ) be extremal subobjects of X and Y respectively. Then (Ai*A 2 ,a i x a 2 ) is said to be a maximal rectangle in R if and only if (Ai>:A 2 ,aixa 2 ) <_ (R,j) and if (Bjjbi) and (B 2 ,b 2 ) are extremal subobjects of X and Y respectively such that (AixA 2 ,aixa 2 ) <_ (BixB 2 ,b \b 2 ) <_ (R,j), then (B 1 >'B2,b 1 xb 2 ) = (A 1 xA 2 ,a i x a 2 ) . 6.19. Prop osi tion . Let Â£ be finitely union distributive, let (R,j) be a difunctional relation from X to Y (5.22), and let (Ri,ki) and (R 2 ,k 2 ) be maximal rectangles in R such that (Rj,ki) t (R 2 Â»k 2 ) . Then (RiYHRjY.n) = U, \$ x ) and (XR 1 nXR 2 ,X) = ( Y ) Â• Hence, in particular, (R 1 nR 2 ,y) e (Â§, cj) XxY ). P roof . If (Rj.kj) H ( \$, <Â£ XxY ) then (Ri.kj) Â£ (R 2 ,k 2 ) since the following diagram commutes. XxY Thus, since (Rj,kj) is a maximal rectangle, it follows that (Ri,kj) E (Ps. 2 ,k 2 ) contradicting the hypothesis. Hence (Rj,ki) t (\$>x*y)' Similarly (R 2 ,k 2 ) f (*, X x Y ) . Now, (Rj.ki) E (R 1 YxXR 1 ,y 1 xy 2 ) and (R 2 ,k 2 ) E (R 2 YXXR 2 , \\*W (6.7) 3y an argument similar to that used in the proof of Preposition 6.17 it follows that (R 2 _1 ,k 2 *) = (XR 2 xR 2 Y, A^A^ , and

PAGE 120

114 (Rj-^kx*) E (XRxxRiY.ya^l)Suppose that (XR 1 nXR 2 ,X) f (\$,'<+>). Then (RloR2 -1 ,j) = (RiYxR 2 Y,y 1 xXi) (6.13) and hence ( (Rj oR 2 _1 )oR 2 ,ct) and (RjYxXR 2 ,uixX 2 ) are isomorphic relations (6.13). Similarly (R 2 oR 1 ~ 1 ,k) = (R 2 YxR 1 Y,A 1 x Ml ) (6.13) and hence ((R 2 oR 1 1 )cR 1 ,3) = (R 2 YxXR 1 ,X 1 xy 2 ) (6.13). Since (Rj.ki) < (R,j) and (R 2 ,k 2 ) Â£ (R,j), (Rx" 1 ,^*) < (R _1 ,j*) and (R 2 1 ,k 2 *) <_ (R _1 ,j*) (1.12) and hence ((R 1 oR 2 _1 )oR 2 ,a) <_ ((RoR -1 )oR,j') and ( (R 2 oRj -1 )oRj , 6) Â£ ( (RoR -1 )oR, j ' ) (1.30). Hence (RlV*/R2VÂ£M(RioR 2 1 )oR 2 )V*/((R 2 oRr 1 )oR 1 ),I 1 ) 1 (RV*J ( (RoR" 1 )oR) ,Z 2 ) (5.5). Since (R,j) is difuncticnal (5.22), (B.[*) ((RoR _1 )oR) ,Z 2 ) <_ (R,j) (5.5). Thus since g, is finitely union distributive, it follows that: ((R 1 Y^R 2 Y)x(XR 1 ^;XR 2 ),C.) E ((RxYxXRx) [*) (R 2 YxXR 2 ) [*J (RiYxXR 2 ) {*J (R 2 YxXRx) ,E) = (Rl\0 , R 2 ^((RloR 2 " 1 )oR 2 )^; ((R 2 oR i 1 )oRi) J I) 1 (R,j). Let K = (R 1 Y(*/R 2 Y)x(XR 1 ^XR 2 ). Since (Ri.ki) Â£ (K,0 and (R 2 ,k 2 ) <_ (K,5) and (K,0 is rectangular, by the definition of maximal rectangle, it follows that (Ri,ki) = (K,Â£) E (R 2 ,k 2 ), contradicting the hypothesis. Thus (XR 1 fiXR 2 ,A) E (\$,{>Â„). Now suppose that (RiYf\R 2 Y,y) t (?',*). Then (Rl -1 oR 2 ,j) E (XRixXR 2 ,y 2 xX 2 ) and ((R 1 o(Ri~ 1 oR 2 ) ,o) E (RjYxXR 2 ,ui^X 2 ) (6.13). Similarly (R 2 _1 oR 1 ,k) E (XR 2 xXRi,XiXy 2 ) and (R 2 o(R 2 _1 oRi),B) E (R 2 YxXRi,A :
PAGE 121

115 and since (Ri,ki) <_ (K,5) and (R2,k2> Â£ (K>5) it follows that (Rl,ki) = (K,5) = (R2,k2) contradicting the hypothesis. Consequently, (R^Hr^u) = (4,* x ). The above implies that (RiHR2 5Y) (\$,vv) since ' XX Y ((R 1 YxXR 1 )n(R2YxXR 2 ),Y) -= ( (Ri Y C\ R 2 Y) x (XRjA XR 2 ) ,y*A) = (Sx^^x^) h (*,<(. XxY ) (0.8 and 6.6). 6.20. Proposition . If g has (finite) coproducts and is (finitely) union distributive and if { (A^xB^,a-j_xbi.) : ^e ~^^ ^ s a (finite) family of rectangular relations from X to Y such that (A-jf^A,a) = (O.^y) and (B Â± f\B.,b) = (*,<|> Y ) for i ^ j, i,jel then (R,j) = ( \*) (k Â± *B Â± ) , j ) is a iel difunctional relation from X to Y. Proof . First consider (R -1 ,j ;,; ). Since ((A i xB i )~ 1 , (a i xb i )*) E (B i xA 1 ,b i xa i ) it follows that (R~ ! ,j*) = ((^(AiXBi))1 ^*) = <.\*) (B Â± xA Â± ) ,i) (5.8). Thus iel iel (RoR _1 ,a) 3 ({*) (A i xB i )oV*;(B j xA i ),a). But iel iel (^(A.xB.)oV*;(B i xA i ),a) = ( [*) ( (A^B.) o (B, xA, ) ) ,g) (5.34). From this iel iel (i,j)etxl J J and the fact that (B i AB.,b) = (\$,\$Â„) for i ^ j, i,jel it follows that (RoR-^.a) = ({*) (A i xA i ),a) (6.13). iel Similarly ( (RoR~ 1 )oR,k 1 ) = ( lÂ£> (A^A^ o l*J (A.xB . ) , y) iel jel J J (^((A i xA i )o(A,xB,)),Y) (5.34). (i.jjetxl 1 J J But since (A^AA,^) E (?,<> x ) for i 4 j, i,jel it follows that (^((A,x/\ ;i )o(A i xB i )),Y) E ((*! (A Â± x?> Â± ),3) (R,j). Thus it has been (i,j)Â£ixi J ^ iel shown that ((RoR _1 )oR,k^ ) = (R,j). Similarly it can be shown that: (Ro(R" 1 eR) ,k 2 ) = (R,j). Hence (R,j) is difunctional. 6.21. Def init ions . Let X be any g -object and let X be a relation on X. Then (R ; j) is a square in X V X if and only if there exists an extremal

PAGE 122

116 subobject (A,a) of X such that R and AxA are isomorphic relations on X. If (S,k) is a relation on X and (R,j) is a square in X*X such that (R>j) Â£ (S,k) then (R,j) is said to be a maximal s quare in (S,k) if and only if for any square (T,m) in XxX for which (B,j) <_ (T,m) <_ (S,k) holds, it follows that (T,m) = (R,j). 6.22. Proposition . Let g be finitely union distributive and let (S,k) be a quasi-equivalence on X. Then (R,j) is a maximal rectangle in (S,k) if and only if (R,j) is a maximal square in (S,k). Proof. Assume (R,j) is a maximal square in (S,k). Suppose (R,j) is not a maximal rectangle in (S,k) then there exist extremal subobjects (B,h) and (C,c) of X such that (R,j) Â£ (B>:C,bx c ) <_ (S,k) and (R,j) f (Bxc,bx c ). Since (S,k) is symmetric and (BxC,b*c) < (S,k), (CxB,cxb) Â£ (S -1 ,k*) = (S,k) (1.12 and 1.13). Since (S,k) is transitive it follows that ( (BxC) o (CxB), a) = (BxB.bxb) <_ (SoS -1 ,a) = (SoS,k#) <_ (S,k) and ((C*B)o(BxC),a) = (CxC,cx c ) < (S -1 oS,a) E (SoS.k//) <_ (S,k) (6.13 and 1.30). Since fc is finitely union distributive it follows that (B^C)x(B^C) = ((BV*JC)xB)\*;((Bi*;C)xC) (BxB) (.*/ (CxB) {*) (BxC) \*J (CxC) . Since each of BxB, CxB, BxC, and C V C is contained in S it follows that (R,j) < (BxC,bx c ) <_ ( (B \*JC) * (B [*) C) . B) < (S.k). Since (R,j) t (BxC,bx c ), (R,j) f ((B'c.JC)x(B^C),6) contradicting the maximality of (R, j ) .. Conversely if (R.j) is a maximal rectangle in (S,k) then (R,j) = (BxC,bxc.) for some paj.t of extremal subobjects (B,b) and (,C,c)

PAGE 123

117 of X. Repeating the above it follows that (R,j) = (BxC,bxc) and (BxC,bx c ) <_ ((B^*JC)x(BlÂ£jC),6) <_ (S,k). Since ((BV*/C)x(B \*) C) , 3) is a rectangular relation then by the raaximality of (R,j) it follows that (R,j) = ((BlÂ£f C)x(Bl*J C),B) and hence is a square. Thus (R,j) is a maximal square in (S,k). 6.23. Example . Consider the following symmetric relation in Top 1 . Let X = Â£.0,1 j with the usual topology, let S = [0,i\ x [%, 3/4*1 U [^,3/4*1 x [p,lj and let k be the inclusion map taking S into XxX. It is clear that (.0,1 J x ^,3/4j together with its inclusion map is a maximal rectangle in (S,k) that is not a maximal square. It is also clear that |*S,3/4J x \^,3/4j together with its inclusion map is a maximal square in (S,k) that is not a maximal rectangle. Note this shows that even in a symmetric relation it may be the case that both maximal squares and maximal rectangles exist and are distinct . Also note that the above example is valid in Top ? and in CpT 2 Â• By neglecting the topology and considering the underlying set, the example is valid in Set. 6.24. Proposition . If g is finitely union distributive, (R,j) is a quasi-equivalence on X, and (Rj.kj) and (R2,k 2 ) are maximal squares in (R,j) such that (R^kj) t (R 2 ,k 2 ) then CR 1 nR 2 ,k) = <-*>\$x*x)' Proof. Both of (R ,k ) and (R^k,) are maximal rectangles (6.23). (R,j) is difunctional (5.24) hence the result follows immediately from 6.19. 6.25. Prop o sition . Let fa have (finite) coproducts and be (finitely) union distributive and let {(A-^,a^) : iel} be a (finite) family of

PAGE 124

118 extremal subobjects of X with the property that (Aif\Aj,a) = (,<{>x) for i i J. ifjel. Then (.[*) (.A^k^) ,a) is a quasi-equivalence on X. id Proof. Since (A^A^a^ai) = ((AjXAi)"" 1 , (a-j^a-j^*) (3.3) it follows that ((^(AiXAi))" 1 ^*) 5 (\Â£J (AiXAi)" 1 ^) E 0*) (A^Ai) ,a) (5.8). Hence iel iel iel (Uy (Aj_xAj_),a) is symmetric. iel Next observe that: (V*j(A 1 xA Â± )oV*;(A i xA 1 ) 1 a#) = ( \*J ( (A i xA i )o (AjX Aj ) ) , P) (5.34). Hence, iel iel (i,j)elxl since (Aif\Aj,a) = (4>,x) for i / j, it follows that ((CJ ((Ai> X ) if i J j, i,jel. Such a family is said to be a ( finite ) partiti on of X if {\*}k Â±i a.) = (X,l x ). iel 6.27. Theorem. Let Â£ have (finite) coproducts and be (finitely) union distributive and let {(A-^.a-^): iel} be a (finite) partition of X. Then ( KÂ£J (A^xA^) ,a) is an equivalence relation on X. iel Proof. Id view of 6.25 it is immediate that ( {*) (A^A-j ) Â» a ) is a quasiiel equivalence on X. To see that ( \*J (Ai x Ai) ,a) is reflexive it suffices iei to show that iT^a is an epimorphism (3. 9 ). Since (AixAi,aixai) is rectangular it follows that ((AixA i )X,j 1 ) E (Ai,ai) E (X(AixAi),j 2 ) (6.7). Also since (AixAi,aixai) <_ ( [*) (AixAi) ,a) it follows that for each iel, iel ((A i xA i )X,j 1 ) < ((^(AixAi))X >3j ) (4.10); i.e., for each iel, iel

PAGE 125

119 (Ai,ai) < ((^(AixAi))X,ai). Thus (^Ai,a) <_ (( \*J (AixAi))X,ai) (5.1), iel :L,f:I iei But ((C*i(A i xA i ))X ) a 1 ) Â£ (X,l x ) and <'\^A i>a ) E (X,1 X ). Thus iel Lei (X,l x ) = (l*jA i5 a) = ((^(A i xA i ))X I. That is, there exists an isomoriel iel phisra o such that the following di; commutes. iel -VXXX X Thus irja a t\. But ti is an epimorphism and o is an isomorphism.' hence irja is an epimorphism, as was to be proved.

PAGE 126

BIBLIOGRAPHY 1 A. R. Bednarek and A. D. Wallace, "Equivalences on Machine State Spaces", Matematicky casopi s 17 (1967) #1, 1-7. 2 , "A RelationTheoretic Result with Applications in Topological Algebra", Math . Systems Theo ry 1 (1963) #3, 217 224. 3 P.M. Cohn, Universal Algebra , Harper and Row, New York, 1965. 4 P. Freyd, Abel] a n Categories , Harper and Row, New York, 1964, 5 H, Herri ich, "Constant Maps in Categories", Preprint. 6 , "Topologische Reflexionen und Coref lexionen" , Lectur e Notes in Mathematics , Vol. 78, Springer-Verlag, Berlin, 1968. 7 H. Herrlich and G. E. Strecker, "Coref lective Subcategories I Â— Generalities" (to appear in Trans . Amer . Math . Soc. ) . 8 __, "Coref lective Subcategories in General Topology" (to appear in Rozprcwy Matematyczne ) . 9 , Category Theory , Allyn and Bacon, Boston (to appear) . 10 S. M. Howie and J. R. Isbell, "Epimorphisns and Dominions II", Journal of Al gebra 6 (1967), 7 21. 11 J. R. Isbell, "Subobjects, Adequacy, Completeness, and Categories of Algebras", Rozprowy Matematyczne XXXVI (1964), 1 32. 12 , "Epimorphisms and Dominions". In "Proceedings of the Conference on Categorical Algebra, La Jolla, 1965", Lange and Springer, Berlin, 1966, 232 246. 13 J. Lambek, "Goursat's Theorem and the Zassenhaus Lemma", Canad. J. Math. 10 (1968), 45 56. 14 , "Goursat's Theorem and Homological Algebra", Canad. Math. Bui . 7 Oct, (1964) #4 , 597 607. 15 .> "Completions of Categories'", Lect ure No tes in MaJ^emaj^Lcjs, Vol. 24, Springer-Verlag, Berlin, 1966. 120

PAGE 127

121 16 , Lectures in Rings and Modules , Ginn Blaisdell, Waltham, Mass., 1966. 17 C. E. Linderholm, "A Group Epimorphism is Surjective", A mer . Math. M onthly 77 (1970) #2, 176 177. 18 SÂ« MacLane, "An Algebra of Additive Relations", Proc. Nat. Acad. Sci. 47 July (1961) #7, 1043 1051. 19 , "Categorical Algebra", Bui . Amer . Math. Soc . 71 Jan. (1965) #1, 40 106. 20 S. MacLane and G. Birkhoff, Algebra , The Macmillan Company, New York, 1967. 21 B. Mitchell, Theory of Categories , Academic Press, New York, 1965. 22 J. Riguet, "Relations Binaires, Fermetures, Correspondences de Galois", Bui . Soc. Math. France 76 (1948), 114 155. 23 , "Quelques Proprietes des Relations Difonctionelles", C. R. Acad . Sci. Paris 230 (1950), 1999 2000.

PAGE 128

BIOGRAPHICAL SKETCH Temple Harold Fay was born August 12, 1940, at Washington, D.C. He was graduated from Swarapscott High School, Swampscott, Massachusetts, in June, 1959. In June, 1963, he received the Bachelor of Science degree from Guilford College at Greensboro, North Carolina. In August, 1964, he received the Master of Arts degree from Wake Forest University. After spending the academic year of 1965 66 as a graduate teaching assistant at the University of South Carolina, he returned to an instructorship at Wake Forest University. In September of 1966 he enrolled in the Graduate School of the University of Florida, embarking upon a program leading toward the degree of Doctor of Philosophy. From September, 1970, until the present time he has been an Assistant Professor of Mathematics at Hendrix College, Conway, Arkansas. 122

PAGE 129

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. George E. Strecker, Chairman Assistant Professor of Mathematics I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. V iX\\w n ^ o* -
PAGE 130

I certify that I have read this 'Study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. >xÂ£K-V -Rerrait N. Sigmon Assistant Professor of Mathematics I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Leo Polopolus Professor of Agricultural Economics This dissertation was submitted to the Dean of the College of Arts and Sciences and to the Graduate Council, and v;as accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. March, 1971 kiaWiIt Dean, College of Arts and Sciences Dean, Graduate School

Citation

## Material Information

Title:
Relation theory in categories
Creator:
Fay, Temple Harold, 1940-
Place of Publication:
Gainesville, Fla.
Publisher:
University of Florida
Publication Date:
1971
Language:
English
Physical Description:
vi, 122 leaves. : illus. ; 28 cm.

## Subjects

Subjects / Keywords:
Algebra ( jstor )
Equivalence relation ( jstor )
Factorization ( jstor )
Isomorphism ( jstor )
Mathematical congruence ( jstor )
Mathematics ( jstor )
Morphisms ( jstor )
Rectangles ( jstor )
Topology ( jstor )
Universal algebra ( jstor )
Categories (Mathematics) ( lcsh )
Dissertations, Academic -- Mathematics -- UF
Mathematics thesis Ph. D
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

## Notes

Thesis:
Thesis -- University of Florida.
Bibliography:
Bibliography: leaves 120-121.
Also available on World Wide Web
General Note:
Manuscript copy.
General Note:
Vita.

## Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
022444242 ( AlephBibNum )
13731322 ( OCLC )
ACZ7747 ( NOTIS )

Full Text

Relation Theory in Categories

By

TEMPLE HAROLD FAY

A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF
THE UNIVERSITY OF iLOir -. IN PARTIAL
FULFILLMENT OS TIH REQUIREMENTS FOR THE DEGREE' OF
DOCTOR OF PHILOSOPHY

UNIVERSITY 0: FL..IDA

To Dr. George E. Strecker, without whose tactful prodding,

suggestions this work would never have been completed.

Abstract.................................... .................... ... iv

Introduction................................. ...................... 1

Section 0. Preliminaries...................................... 6

Section 1. Generalities ................ ......................... 18

Section 2. Categorical Congruences............................. ]

Section 3. Categorical Equivalence Relations and
Quasi-Equivalence Relations........................ 8A

Section 4. Images............................................ 63

Section 5. Unions............................................ 75

Section 6. Rectangular Relations .............................. 103

Bibliography......................................................... 120

Biographical Sketch ...................................... ......... 122

Abstract of Dissertation Presented to the
Graduate Council of the University of Florida in Partial Fulfillment
of the Requirements for the Degree of Doctor of Philosophy

RELATION THEORY IN CATEGORIES

By

Temple Harold Fay

March, 1971

Chairman: Dr. George E. Streckcr
Major Department: Mathematics

The purpose of this dissertation has been to systematically

generalize relation theory to a category theoretic context. A quite

general relation theory has emerged which is applicable not only to

concrete categories other than the category of sets and functions, but

also to abstract categories whose objects need have no elements at all.

This categorical approach has provided the opportunity to comprehend

classical relation theory from a new vantage point, thus hopefully

leading to an eventual better understanding of the subject.

A relation from an object X to an object Y is a pair (R,j) where

j is an extreimal monomorphism having domain R and codomain XxY. By

choosing j to be an extremal monomorphism, relations in the category

of sets are the usual subsets of the Cartesian product, relations in

the category of groups are subgroups of the group theoretic product,

and relations in the category of topological spaces are s.bspaccs of

the topological product. This latter fact would not be the case if

relations would Le defined to be merely subobjects of the categorical

product.

Section 0 notes results which are purely categorical in nature

and vhicl.h vill be used extensively throughout the sequel. Particular

emphasis is given to the epi-extremal mono factorization property and

necessary and sufficient conditions for the existence of this factori-

zation and equivalent forms of the property.

In Section 1, the basic machinery for categorical relation theory

is developed. For example, such notions as inverse relation, reflexive

relation, symmetric relation, and composition of relations are defined

and several important results are obtained.

Section 2 deals with a categorical definition of a congruence

relation. Several algebraic results of Lambek and Cohn are generalized.

Equivalence relations and quasi-equivalence relations (symmetric,

transitive relations) are studied in Section 3. A quasi-equivalence on

an object X is shown to be an equivalence relation on a subobject of X.

If R is a set theoretic relation from the set X to the set Y and

A is a subset of X then AR = {ycY: there exists acA such that (a,y)eR}.

This definition is generalized in Section 4 and results similar to those

obtained by Riguet are demonstrated.

If {(Ri,ji): iel} is a (finite) family of relations from X to Y

then the relation theoretic union (URi,j) of the family is obtained by
iel
taking the intersection of all relations from X to Y which "contain"

each Ri. If the category being investigated is assumed to have (finite)

coproducts then the union of the family considered as subobjects and the

relation theoretic union of the family considered as extremal sabobjects

turn out to be given by the unique extremal epi-mono and unique epi-

extremal mono factorizations of the canonical morphism from the coproduct

of the family to XxY.

The notion of a (finite) union distributive category is introduced.

Roughly speaking, this property guarantees that unions commutee" with

products and intersections.

Section 5 deals with unions and the importance of the concept of

difunctional relation is brought out.

A well known result in set theoretic relation theory is that a

partition determines an equivalence relation. In order to obtain this

result in its generalized form the existence of an initial object which

behaves similarly to the initial object in the category of sets (namely

the empty set) is postulated and disjointness becomes a useful categor-

ical notien. Also the notion of difunctional relations was crucial in

obtaining the above result.

Section 6 deals with rectangular relations and the above result

INTRODUCTION

The purpose of this work has been an attempt to systematically

generalize relation theory to a category theoretic context. In doing

so, several goals have been realized. Firstly, a quite general rela-

tion theory has emerged which is applicable.not only to concrete cate-

gories other than the category of sets and functions, but also to

abstract categories whose objects need have no elements at all. Second-

ly, taking a categorical approach has provided the opportunity to

comprehend classical relation theory from a new vantage point, thus

hopefully leading to an eventual better understanding cf the subject.

Hany relation theoretic results have been rather straightforward

to prove in an "element free" setting, once the appropriate machinery

has been constructed to handle them. On the other hand it has been

surprising to see that some results which are easy to prove in the set

theoretic context are much more difficult to show categorically.

For example, it is easy to prove that if R is a set theoretic

relation from X to Y such that RY = X then RoR"' = {(x,z): there exists

yeY such that (x,y)eR and (y,z)cR-]} is reflexive. This result can be

generalized to categories but is no longer easy to Drove and the result

gains some significance.

Another easy result in set theoretic relat-con theory is that if

AX and Av are the diagonals on X and Y respectively then AxcF. = R = Roby

This result is also generalized to category es bIt "i -or.crhic as rela-

tions" replaces "equality" and the result is nc longer easy to prove.

Whenever one is generalizing properties care must be taken to be

certain that the generalized definitions are really generalizations of

the notions be-ig considered and that the proper generalization of the

definition is obtained. This seems to be particularly important in

category theory. Care has been taken when selecting the basic notion

of a relation from an object X to an object Y to be an extremal sub-

object of the categorical product XxY; i.e. a pair (R,j) where j is an

extremal monomorphism having domain R and codomain XxY. By doing so

relations in the category of sets are the usual subsets of the carte-

sian product, relations in the category of groups are subgroups of the

group theoretic product, and relations in the category of topological

spaces are subspaces of the topological product. This latter fact

would not be the case ii relations would be defined to be merely sub-

objects of the categorical product. Much care has also been taken with

the definition of composition of relations (1.26). Using this defini-

tion many nice results have been obtained; however, in general, the

composition of relations is not associative (1.36). This, at first

glance, seems to be pathological and casts doubt on the suitability of

the definition of composition of relations. However, the wealth of

other important results obtained belies this doubt (see 1.37). Also,

some further atonement is yielded by trre fact that for rectangular

relations composition is asscciarive (6.15).

Cohn [31 and Lambek i3J define a congruence ii an algebraic

setting to be a sLbalgebra of the cartesian product -vhich is "coipat-

ible' with the algebraic operations and which is set theoretically an

equivalence relalion. in this work, a generalized noLion of congruence

is given which is equivalent to the above in algebraic categories and

the result that a (categorical) congruence is a (categorical) equi-

valence relation is obtained.

It was found that categorical unions were very difficult to work

with. However, by assuming the category being studied had (finite)

coproducts as well as being locally small and quasi-complete the notion

of union became somewhat easier to handle.

For instance, if {(R.,j ): icl} is a (finite) family of relations

from X to Y then the union (CJR.,j) of the family, considered as sub-
iT 1
objects of XxY is not necessarily a relation from X to Y,since j is not

necessarily an extremal monomorphism. The relation theoretic union of

the family is obtained by taking the unique epi-extremal mono factoriza-

tion of j (5. 3) or equivalently by taking the intersection of all rela-

tions from X to Y which "contain" each R.. If the category being inves-

tigated is assumed to have (finite) coproducts in addition to being

locally small and quasi-complete then the union of the family considered

as subobjects and the relation theoretic union of the family considered

as extremal subobjects turn out to be given by the unique extremal epi-

mono and unique epi-extremal mono factorizations of the canonical mor-

phism from the coproduct of the family to XxY (5.29). It is also snown

that when the category has (finite) coproducts both factorizations

respect unions (5.30 and 5.42).

Unions are still difficult to handle even with the assumption of

(finite) coproduets mentioned above; hence, the notion of a (finite)

union distributive category is introduced (5.31). Roughly speaking,

this property guarantees that unions "commuce" .ith products and inter-

scutions and thus unions become "easy" tj handle. Examples of union

distributive categories show that such categories tend to be more of a

topological nature rather than of an algebraic nature.

The set theoretic notion of difunctional relation is due to

Riguet [221 and its importance has been ncted by Lambek 13] and

HacLane 8 A set theoretic relation R is difunctional if and only
-1
if RoR oR C R. The categorical definition in view of the fact that

associativity cannot be assumed reads: R is difunctional if and only if
-1 -1
(RoR )oR < R and Ro(R oR) < R where "<" is the usual order on sub-

objects. It is easy to prove, again by choosing elements, that if a
-1
set theoretic relation R is difunctional then R = RoR oR. However,

the similar result in the categorical setting is much harder to obtain
-1
and is rephrased: if R is difunctional then R (RoR )oR and
_]
R E Ro(R oR) where "-" means isomorphic as extremal subobjects (5.28).

A well known result in set theoretic relation theory is that a

partition determines an equivalence relation. In order to obtain this

to the category being studied. In particular, the existence of an ini-

tial object which behaves similarly to the initial object in the cate-

gory of sets (namely the empty set) had to be postulated and disjoint-

ness became a useful categorical notion. Again, examples of such cate-

gories are non-algebraic. Also the notion of difunctional relations was

crucial in obtaining the above result (-6.20).

The excellent reference paper by Riguet 22J has been used as a

guide for the results of set theoretic relation theory. Indeed, most

all of the results contained herein are generalizations of results in

.22 The papers by Lambek 3 14 MacLane 8 and Bednark

aid Wallace provided motivation for many of the generali-
,1 1- 1,

nations.

The basis for the categorical notions has been taken from the

papers of Herrlich and Strecker 7 ], S8 1 Isbell 11 ], 12 ,

and the forthcoming text by Herrlich and Strecker [9 J (which has

greatly influenced this work). For most of the basic categorical

notions the reader is referred to the texts by Mitchell [211 Freyd

14 and Herrlich and Strecker 9 9.
The work here is begun with a preliminary Section 0 which notes

(often without proof) results which are purely categorical in nature and

which will be used extensively throughout the sequel. Particular empha-

sis is given to the epi-extremal mono factorization property and neces-

sary and sufficient conditions for the existence of this factorization

and equivalent forms of the property. However, it is not intended that

the preliminary section give a complete category-theoretical background.

It is expected that the reader be familiar with the basic categorical

notions,

SECTION 0. PRELIMINARIES

0.0. Remark. It is assumed that the reader is familiar with the basic

notions of category theory and hence such basic notions as epimorphism,

monomorphism, retraction, section, equalizer, regular monomorphism,

coequalizer, regular epimorphism, subobject, and limits shall not be de-

fined. The reader is referred to Mitchell [21 and Herrlich and Streck-

er (9] for such notions. All of the following results are proved in

detail in Herrlich and Strecker (9] Since Theorem 0.21 is vital to

this work the proof is sketched here.

0.1. Notation. The category whose class of objects is the class of all

sets and whose morphism class is the class of all functions shall be

denoted by Set.

The category whose class of objects is the class of all groups

and whose morphism class is the class of all group homomorphisms shall

be denoted by Grp.

The category whose class of objects is the class of all tcpological

spaces and whose morphism class is the class of all continuous functions

shall be denoted by Topi.

In a manner similar to that described above, one obtains the fol-

lowing categories:

FSet finite sets and functions;

FGp finite groups and grcup homomorphifsms;

Ab Abelian groups and group homomorphisms;

SGp

SG1

Rng

Top2

CT 2

0.2. Proposition.

be m -morphisms.

1) If f and

2) If f and

3) If gf is

4) If gf is

5) If gf is

f g
Let : be a category and let X-----Y and Y-----Z

g are monomorphisms then gf is a monomorphism.

g are epimorphisms then gf is an epimorphism.

a monomorphism then f is a monomorphism.

an epimorphism then g is an epimorphism.

an isomorphism then g is a retraction and f is a

section.

0.3. Remark. In general, an equalizer is a limit of a certain diagram.

It is an object together with a morphism whose domain is the object. A

regular monomorphism is a morphism for which there exists a diagram so

that the domain of the morphism together with the morphism is the equal-

izer of the diagram.

It is observed in Herrlich and Strecker ,9 J that certain func-

tors preserve regular monomorphisms while not preserving equalizers,

hence one reason for the above distinction between equalizers and regu--

lar mcnomorphisms.

In this paper, since we shall not deal with functors, no distinc-

- semigroups and semigroup homomorphisms;

- semigroups with identity and semigroup homomorphisms which

preserve the identity;

- rings and ring homomorphisms;

- rings with identity and ring homomorphisms which preserve

the identity;

- Hausdorff spaces and continuous functions;

- compact Hausdorff spaces and continuous functions.

tion shall be made between equalizers and regular monomorphisms; i.e.,

between the pair (object and morphism) and the morphism alone. Both will

be called equalizers.

f
0.4. Proposition. Let P be a category and let X----- Y be a -

morphism. Then the following are equivalent:

1) f is an isomorphism,

2) f is a monomorphism and a retraction,

3) f is an epimorphism and a section,

4) f is a monomorphism and a regular epimorphism,

5) f is an epimorphism and a regular monomorphism.

0.5. Definition. Let {Ai: iIl} be a family of -objects then the pro-

duct ~tA.,Tr.) of the family is a -obj-ect T"Ai together with pro-
i 1
is 1 iel
section morphisms Ti:TTAi ----- Ai with the property that if P is
iel
any -object for which there exist m -morphisms p.: P ----- 'Ai for

each icl, then there exists a unique morphism X: P ---> Ai such that
iel
ilX = i for each iel.

The dual notion is that of the coproduct (JJ.Aii.).
iel

0.6. Definition. Let {(Ai,ai): iEl} be a family of subobjects of a --

object X. Then the intersection ( ~A,,a) of- the family is a --object
iel
1 Ai together with a morphism a: Oi1A ----^X where for each i there
icl iEl
is a morphism Xi: X f A ---- Ai such cthat ai.. = a with the property
ic!2
that if P i.' any object for which there exist: --morphisms p: P ----" X

and pi: P ---- '-A1 such that a1ig. p for each isi then there exists a

unique norphism ,: P --1-" A. such that aX = p.
It i
It follows that a is a mnonom.orphism.

0.7. Remark. The above two definitions are mentioned because of the

fundamental role they play in the sequel. They are special limits and

are perhaps the most important limits in the categories that will be

considered in tnis work.

The following theorem is a special case of a more general theorem

dealing with the commutation of limits which can be found in Herrlich

and Strecker  A variation of the theorem will be proved in

Section 1 (1.5).

0.8. Theorem. Let {(Ai,a.): icI} and {(Bi,b.): icl} be families of sub--

objects of 0 -objects X and Y respectively. Then if P has finite

products and arbitrary intersections then (( A.i)x( Bi) and ^(-(A.xB.)
iI 1 iEI il
are canonically isomorphic.

0.9. Notation. Let {Xi: icI} be a family of -objects and suppose
f.
{Z ---- -- Xi: icl} is a family of 0 -morphisms. Then by the defini-

tion of product there exists"a unique morphism h from Z to TT Xi such
i.E
that irih fi for each iI. This morphism h shall be denoted by
iET
Let A and B be 0 -objects and suppose that a: A ---- X and

b: B ----->Y are -morphisms. If P1 ard P2 are the projection mor-

phisms from AxB to A and B respectively then apl: AXB ---- X and bP2

AxB ----' Y, hence by the definition of product there exists a unique

morphism g from AxB to XxY such that g = a and g = b. Ths mor-
1 l and 7..g = b 2P This mor-
phism g shall be denoted by axb and shall be called the product of a

and b.

Let f be a ? -morphisrm from X to Y. If f is a moncmorphism then

the following notation shall be used:

X ---- ---

If f is an epimorphism then the following notation shall be used:
f.
X ---- Y

If f is an equalizer then the following notation shall be used:
f
X ----------- Y

If f is an isomorphism then the following notation shall be used:
f
X 1----------v Y

a b c d
0.10. Proposition. Let A -----X, B ---- Y, X----- Z, and Y --- W

be ;' -morphisms. Then (cxd) (axb) = caxdb.

a b
0.11. Proposition. Let A ---- X and B --->Y be monomorphisms (respec-

tively, sections, isomorphisms) then axb is a monomorphism (section,

isomorphism).

0.12. Remark. A partial order may be defined on the subobjects of an

object in in the following way:

If X is a P -object and (A,a) and (B,b) are subobjects of X; i.e.,

a and b are monomorphisms with codomain X and domains A and B respec-

tively, then (A,a) < (B,b) if and only if there exists a morphism c from

A to B such that be = a.

b
B 1-- --X-

I- a
A 7

By an abuse of language, if (A,a) < (d,b) then (B:b) is said to

contain (A,a) and the morphism c is sometimes called the inclusion of

(A,a) into (B,b). It is easy to see that if (Aa) < (B,b) an'

(B,b) < (A,a) then the morphism c is an isomorphisnm. In this case,

(A,a) and (B,b) are said to be isomorphic as subobjects of X. This is

a stronger condition than A and B just being isomorphic objects in the

category The following notation shall be used to denote the case

where (A,a) and (B,u) are isomorphic as subobjects of X:

(A,a) E (B,b).

Sometimes it is written (inaccurately) that A < B or that A and

B are isomorphic as subobjects of X. When this is done, the morphisms

a and b should be clear from the context.

It is immediate that (A,a) E (B,b) if and only if (A,a) < (B,b)

and (B,b) < (A,a). Thus the relation "<" on subobjects is easily seen

to be a partial order up to isomorphism as subobjects.

0.13. Definition. Let f from X to Y be a -morphism. f is an extremal

imonomorphism if and only if f is a monomorphism and whenever f = gh and

h is an epimorphism then h is an isomorphism.

If f is an extremal monomorphism the following notation shall be

used:
f
X -- >Y

The dual notion is that of an extremal epimorphism and is denoted:
f
X f--- Y

If f is an extremal monomorphism f: X -----tY, then (X,f) is

called an extremal subcbject of Y.

0.j4. Remark. The definition of extremal monomornhism is due to Isbell

]i] The concept of extremal monomorphism is important since it

yields what shall be called the "irage" of a morphism (see 0.18),

0.15. Examples. In the categories Set, Grp, Ab and FGp, extremal mono-

mcrphis:.s are precisely the onomorphisms (i.e., one-to-one morphisms).

In the categories Top and CpT extremal monomorphisms are precisely

the cmbeddings. In the category Top2 they are the closed embeddings.

f
0.16. Proposition. If X.-- ,Y is a
0.16. Proposition. If X is a -morphism such that f = gh

and f is an extremal monomorphism then h is an extremal monomorphism.

f
0.17. Proposition. If X -----Y is a -morphism then the following

are equivalent:

1) f is an isomorphism,

2) f is an epimorphism and an extremal monomorphism,

3) f is a monomorphism and an extremal epirorphism (c.f. 0.3).

0.18. Definition. A category is said to have the unique epi-extremal

mono factorization property if for any --morphism X -----Y, there

exist an epimorphism h and an extremal moncmorphism g with f = gh such

that whenever f = g'h' where g' is an extremal monomorphism and h' is an

epimorphism then there exists an isomorphism o such that the following

diagram commutes.

f

'NN

If has the unique epi--extremal rnono factorization property and

if f = gh where h is an epimorphism and g is an extremal monomorphism,

then the pair (h,g) shall be used to designate the epi-extremal mono

factorization of f. The extremai subobject (Z,g) of Y is called the

inage of X under f. Sometimes (Z,g) is referred to as the image of f.

The notion of the unique extremal epi-mono factorization property

is defined dually.

If has the unique extremal epi-mono factorization property and

f = gh where g is a monomorphism and h is an extremal epimorphism then

the pair (h,g) shall be used to designate the extremal cpi-mono factori-

zation of f. The subobject (Z,g) of Y is called the subimage of X under

f. Sometimes (Z,g) is referred to as the subimage of f.

0.19. Definition. A category is said to have the diagonalizing

property if whenever gh = ab such that h is an epimorphism and a is an

extremal monomorphism, then there exists a (necessarily unique) morphism

s such that ( h = b and a g = g.

h
X --. Y

b \$ --

W z ----- ------>a Z
a

0.20. Theorem. Let be a locally small category having equalizers and

intersections. Then the following are equivalent:

1) Y has the unique epi-extremal mono factorization property,

2) has the diagonalizing property,

3) the intersection of extremal monoaiorphisms is an extramsa mono-

morphism and the composite of extremal monomorphisms is an extremral

monomorphi sm,

4) if has pullbacks and if (P,a,3) is the pullback of f av' g

where f = go and f is an extremal monomorphism then a is an extremal

uionomorphism.

5) if 9 has (finite) products then the (finite) product of

extremal monomorphisms is an extremal monomorphism.

0.21. Theorem. If is locally small and has equalizers and inter-

sections then f2 has both the unique epi-extremal mono factorization

property and the unique extremal epi-mono factorization property.

Proof. (sketch). First we will show the existence of the unique extremal

epi-mono factorization property. If f from X to Y is any -morphism

then let ((E.,e) be the intersection of the family {(Ej,e.): jCJ} of
jcj 3 -l -
all subobjects of Y through which f factors. Then it follows that e is

a monomorphism and that f factors through e; i.e., there exists a mor-

phism h such that f = eh. Now, to see that h is an epimorphism suppose

a and B are -morphisms such that ah = Bh. Let (E,k) be the equalizer

of a and Z. It follows from the definition of equalizer that there exists

a morphism g such that kg h.

f
X --- -------- Y

g h e
E >'>^------------ r\ E Z' Z
k jcJ J

Thus it follows that f factors through ek and since ek is a mono-

morphism then there exists a morphism ): i% E. ---- E such that ekX = e.
jJe
From this it follows that k is an iscmcrphism whence a = i and so h is

an epimorphism.

Next it will be shown that h is an extremal epimorphism. Suppose

L: = hlh2 where h1 is a monomorphism. Then eh! is a monomorphism through

which f factors. From this it follows, as above, that h1 is an isomor-

phism and hence h is an extremal epimorphism. Suppose f = g'h' where g'

is a monomorphism and h' is an extremal epimorphism. Then since g' is

a monomorphisim through which f factors there exists a morphism i from

( E. to che codomain of h' (domain of g') such that e = g'T. Since e
j J
and g' are monomorphisms, it follows that h' = Th and that T is a mono-

morphism. Since h' is an extremal epimorphism it follows that T is an

isomorphism. Thus has the unique extremal epi-mono factorizaticn

property.

Now suppose that ge = mA where e is an epimorphism and m is an

extremal monomorphism. It will be shown that there exists a morphism

a from the codomain of e to the domain of m such that Ge = h and mo A g.

Let (.( A.,a) be the intersection of the family {(A.,a.): isI}
iT
of all subobjects of the codomain of g (codomain of m) through which

g and m factor. This family is non-empty since both g and m factor

through the identity morphism on the codomain of g. It follows that both

g and m factor through a. Thus there exist morphisms al and a2 such that

the following diagram commutes,

e
X --------- Y

h C Ai

W Z

It will be shown next that 22 is an epimnorphism. Suppose a* and

are -mcoiphisims for which aca2 = hSa2. Let (L*,k"O ) be the equalizer

of 0* and B*. It follows from the definition of equalizer that there

exists a morphism b, such that k*b1 = a2, since o.*a2 = *a2. Since the

diagram conmrutes it follows' that a*al = B*ale. But e is an epimorphism

hence o*al = e1 s,) that by the definition of equalizer there exists a

morphism b2 such that k*b2 = a1. Thus it follows that m = ak*bi and

g = ak*b2 and so both m and g factor through ak* from which it follows

that k* is an isomorphism. Hence a* = B* and a 2 is an epimorphism. But

n is an extremal monomorphism and m = aa2 and a2 is an epimorphism. Thus

a is an isomorphism. Thus defining o = a]al it follows that the fol-
2 21
lowing diagram commutes and has the diagonalization property.

e
X -~ Y

w X, Z
111

Hence. has the unique epi-extremal rono factorization property

(0.20).

0.22. Theorem. Let ( be any category then the following are equivalent:

1) a is finitelyy) complete,

2) has (finite) products and (finite) intersections,

3) has (finite) products and equalizers,

4) has (finite) products and pullbacks.

0.23. Definition. A category __ is said .o be quasi--complet-, if has

finite products and arbitrary intersections.

0.24. Examples. The categories FSet and FGp are quasi-complete cate-

gories which are not complete. The categories Set, Top1, Top2, CpT2,

Grp, Ab, Ring, and SGp are quasi-complete.

0.25. Remarks. A quasi-complete category is finitely complete but is uot

necessarily complete as the examples FSet and FGp above show.

Also, a locally small, quasi-complete category has both the unique

extremal epi-mono factorization property and the unique epi-extremal

mono factorization property (0.20 and 0.21).

It can be shown that the unique epi-extremal mono factorization

of a morphism can be obtained by taking the intersection of all extremal

monomorphisms through which the morphism factors. It has been shown that

the unique extremal epi-mcno factorization property is obtained by

taking the intersection of all subobjects through which the morphism

factors (0.20). These characterizations shall be used frequently in the

sequel.

SECTION 1. GENERALITIES

1.0. Standing Hypothesis. Throughout the entire paper it will be assumed

that is a locally small, quasi-complete (finite products and arbitrary

intersections) category.

As noted in the preliminary section s enjoys the unique epi -- ex-

tremal mono factorization property.

1.1. Examples. Many well known categories are locally small, and quasi-

complete. Among such are the categories: Set, Top Top2, Grp, Ab, SGp,

SGp1, Rng, Rny, CpT_ and FGp.

1.2. Definition. Let X and Y be -objects. A relation R from X to Y is

an extremal subobject of XxY; i.e., a relation from X tc Y is a pair

(R,j) where R is a -object and j is an extremal monomorphism having

dcmain R and codcmain XxY. A relation from X to X is called a relation on

X.

1.3. Definition. Let (R,j) and (S,k) be relations from X to Y. Then (R,j)

and (S,k) are said to be isouorphic relations if and only if they are iso-

morphic as extremal subobjects cf XxY.

1.4. Examples. In the categories Set, aid Topi relations are subsets of

the Cartesian product together with the inclusion map.

In the categories Grp, and Ab relations are subgroups of the Car-

tesian product together vith the inclusion map.

19

In the categories Top2, and CpT2, relations are closed subspaces

of the Cartesian product together with the inclusion map.

1.5. Proposition. Let X and Y be r -objects and let (A,a) and (B,b) be

extremal subobjects of Y. Then Xx(ACIB) and (XxA)n (XxB) are isomorphic

relations from X to Y.

Proof. Consider the following commutative diagrams.

B
A/AB -------> B

A Y-------
a

-; XxB

1

i-xa

Sxb

SX xy
X> XXY

Consider also (Xx(A B), 1 xc = y1). Since extremal subobjects are closed

under intersections and products (0.20) yl and Y2 are extremal monomor-

phisms.

Since (lXxa)(1XxXA) = 1Xxc = Y1 and (lxxb) (1XXB) = 1Xxc = Y1 then

by the definition of intersection there exists a unique morphism o from

Xx(AAB) to (XxA)/-(XXB) so that y2o = Y1 and the following diagram com-

mutes. Thus:

XI

X xA f

(X xA) (xxb)
I -.

(Xx(AAB), Y) < ((XxA)n (XXB), y ).

1 Xa
X

] xx
X

Xx(A5 B) -- -- -- '- (XxA) (XxB) ------ -- --Xx

B X

XxB

Now let (Trl,T 2), (Tr1,T), (o],P 2) and (pl,p) be the projections of

XxY, Xx(AriB), XxA, and XxB respectively. Observe that:

1Y2 1(Xxa)X = pX l(Xx 2 = 1i 2

2Y2 = 2(] xa)X = aP2X1 = 2(I xbx)X2 = bP2X2
Thus by the definition of intersection there exists a unique morphism z

from (XxA)( (XxB) to AB such that cE = T2y2 and thus by the definition

of product there exists a unique morphism C from (XxA)f/(Xx B) to Xx(AflB)

such that < = ; i.e., Trc = TY2 and 729 = E. Now yS = (1Xxc)E

hence ITYi --- 1Xi = ~TY2 and 2Yl = cr = c = 2 = 2. Thus "y1E = Y2'

whence:

((XxA) r (XxB), y2) ( (Xx(AriB), y').

1 2
X ---------------- (XxA) .jXxB)

A rB XxA Y2 XxB

I/ 3 xa

Xx (, ) ....-Yx------ -
Y1

I. xb
X

1.6. Notation. Let X and Y be -objects and let (XxY,T1,T2T) and

(YxX,pl',2) be the indicated products of X and Y. Then there exists a

unique isomorphism from XxY to YxX, denoted by , such that the

following diagram commutes.

XY IY VI
xY -------- --------- Yx
3
11 1 '
xX
x )------ ---?x

Note that: = 1x and = 1Xx

1.7. Definition. Let (R,j) be a relation from X to Y and let (T,j*) be

the unique epi-extremal mono factorization of j (see 0.18). The

codomain of T (domain of j*) is denoted by R-1 and (R-1,j*) is called

the inverse relation of (R,j) or more simply, when there is little like-

lihood of confusion, the inverse of R.

R 3-- ^----^ XxY X y -^ 2_._ > yxX
R R "7xj*

1.8. Example. In the categories Set, Top1, Top2, Grp, Ab, and FGp,

<72,~1 >: XxY ------. YxX
is defined by (x,y) = (y,x); hence, if (R,j) is a relation from

X to Y then R-1 g {(y,x): (x,y)eR} with j* the inclusion map of R-1 into

YxX.

1.9. Proposition. If (R.j) is a relation from X to Y the R and R"- are

.?iso:crphic objectr, of .

Proof. Since <,2,' i> is an isomorphism and j is an extremal monomorphisr:

then <72, l >j is an extremal monomorphism. But <7r2,IT>j = j*T. Thus since

T is an epimorphism then from the definition of extremal monomorphism it

follows that T is an isomorphism.

1.10. Definition. If (R,j) is a relation from X to X then R is said to be

symmetric if and only if (R-1,j*) < (R,j).

1.11. Proposition. Let (R,j) be a relation from X to Y. Then the inverse

relation ((R-1)-1,j#) of (RP-,j*) and (R,j) are isomorphic relations.

Proof. Consider the following commutative diagram.

R -- --- X Y

< 7 2 < T2' 1>

R- 1 Y---- x----- X

`# I V

(R )- Xxy

Since the two inner squares commute the outer rectangle commutes. Both

of T and T# have been shown to be isomorphisms (1.9). And, as also has

been observed: <72,T > = 1Xxy (1.6). Consequently, T#T is an iso-

morphism and j = j#(T-r). Thus (R,j) ((R~1)-1,j#).

1.12. Proposition. Let (R,j) and (S,k) be relations from X to Y. Then

(S,k) < (Rj) if and only if (S-1,k*) i (R-',j*).

Proof. Consider the following conmiutative diagram.

j <7;2' 1>
R XxY -- YxX
AV-

k -U R1 k*

S-1

If (S,k) < (R,j) then there exists a morphism a: S -- R R such that
-1
ja = k. Define B = TT-1. Then j* = j*-Tt-1 = <2 >jaT1 k

= k*i-T = k*. Thus (S-l,k*) < (R-1,j*).

If (S-1,k*) < (R-,j*) then by the above, ((S-1)-1,k#) <

((R-1)-1,j#) thus (S,k) < (R,j) (1.11).

1.13. Corollary. If (R,j) is a symmetric relation on X then

(R,3) < (R-,j*) whence (R,j) E (R-I,j*).

Proof. Since (R,j) is symmetric (R-1,j*) < (R,j). Thus

(R,j) = ((R-1)-1,j#) < (R-Ij*) (1.11 and 1.12).

Consequently (R,j) (R-1,j*).

1.14. Definition. Recall that since 0 is quasi-complete it has equal-

izers, thus for each -object X let (AX,iX) denote the equalizer o.f

and Tr2 where Tr and i2 are the projections of XxX. Since iX is an

equalizer it is an extremal monomorphism. Hence (Ax,i ) is always a rela-

tion on X (called the diagonal of XxX).

A relation (R,j) on X is said to be reflexive on X provided that

(Axi ) (R,).

1.15. Example. In the categories Grp, Ab, Set, lop,, Top2, and CpT2, it

follows that AX {(x,x): xEX)} XxX with the inclusion map.

1.16. Proposition. For any -object X, <12,T'>i' = iX. Thus:

(Ax,ix) (Ax-l,i*).

Proof. Consider the following commutative diagram.

<1T2' >
XxX >>----------- ---.--- XxX
SX ---x
x .y- /-- i
1 X

TliX = T2iX = TliX = TT2<"'i2,71>iX.

Thus the epi-extremal mono factorization of'iX is (1. ,5-).

1.17. Corollary. Let (R,j) be a relation on X, then (R,j) is reflexive

on X if and only if (R1-,j*) is reflexive on X.

Proof. If (Ax,ix) < (R,j) then (AX,ix) (X-,ix*) <- (p-l,j*) (1.16 and

1.12).

Conversely if (AX,iX) < (R-1,j*) then

(Ax,ix) (A ,ix*) < ((1)-1 ,j#) E (R,j) (1.16, 1.12. and 1.11).

1.18. Proposition. Let (R,j) and (S,k) be relations from X to Y. Then

the relations (RcS)-1 and (R-1fS-1) are isomorphic relations.

Proof. According to the definitions of intersection and inverse relation

we have the following commutative diagrams.

SR .1

X I -" x -"*
RA S >^-------------- xY i- I s-1 ) S --- X-----. YxX

S 2 flk \4 k*
^\~~~ ^ ^ ^ :

9 <'[2 ,'IT1>
R S n s------>- XxY -------- YxX

(R A S)-11

Observe that =1 j*1iX and = k* T2. Thus by the definition

of intersection: (RAS,i) > (R-1A S-1, ). However since r"' is an

isomcrphism, (RkS,) E ((RAS)",p*); whence

((RAS)-1,I*) < (R-1A S-1, ).

To obtain the reverse inequality, note that by the definition of

intersection (R-IAN S-I,) < (R(AS,i) since jT-1I3 = and

kT-1XL = #. Thus (R-1(\S-1,) (R/S,).

Whence (R-1(1 S~ ,) < (RnS,P) H ((RflS)-1,i*). Consequently:

(R-1( S-1,) = ((Rn S)-1, *).

1.19 Remark. It is clear from the definition of intersection (0.6 ) that

if (R,j) < (S,k) and (R,j) < (T,m) then (R,j) < (S(\T,n).

1.20. Proposition. Let (R,j) be a relation on X. Then RAX, R-'(IAX, and

RAR-1n XA are isomorphic relations on X.

Proof. Consider the following commutative diagram.

.... ---+ R-~

X 4

21
/ j
R (R' A ---- --A R -------- XxX ---------2>- XxX
1 R- iX X

b

Note that since ix equalizes I1 and v2', iX = iX (1.16) and

also that < 2' -1 = ; i.e., '< 2,TT 1 > = 1XxX (1.6). Observe

that jT-rX4 -j* = -1 X3 = i X3. Consequently

JT-X4 = i X = 3. Thus by the definition of intersection:

(R-7()AXX3) <_ (R(c X,i X 2).

Also observe that j*rX = J = < 2' I>i 2. Whence

j'TXI = i X2 so that by the definition of intersection:

(R1AxiX2) < (R-1iAi x 3).

Thus:

(R AxiX 2 ) = (R-IfAxix x 3).

Clearly (RCR-ln Axi X 6) < (R(fAx,i 2). But

(Rt Ax,i X2) < (R-'nAxiX 3) and (RftAx X2) < (Rr.A ,iX 2). Thus:

(RA. ) (Aix) (RA,ix 2)/2(R-lI AAx,i X3) (Rn/R-! fAx,i X6)

Hence,

(R}lAx,iX2) (RfR-1('Axi X 6)

<1X, IX
1.21. Lemma. If X is a ( -object and X----- ->XxX is the unique

morphism h such that rr h = T2h = 1,, then (X,<1 ,>) and (AXi ) are

isomorphic relations on X.

Proof. Since r = Tr2<,1X > and iX is th3 equalizer of -r, and T2,

it follows that (X,<,1 X>) < (AXi ). Since l = l, <1XI > is

a section, hence an extremal monomorphism.

Clearly, z I<1 1 >ri = i1 liX = li and
x 1 -X 1 A 1
T2<1X .1 >71iX 1X iX = Tli = T2iX. Hence, by the definition of pro-
duct, l i = iX. Thus (A ,i ) < (X,<1,1 >).
X IX xA x A X x-

1.22. Example. In the categories Set, Top1, To p,, Grp, Ab, and Rng,

<]X,1 x>: X ------- XxX can be defined by <1i.1 >(x) = (x,x) eXxX for all
XtX.

1.23. Remark. It is also easy to see that up to isomorphism of extremal

subobjects (X,<]X,1X>) (and thus (Ax,'iX) also) is the equalizer of each

of the following sets of morphisms:

{i] '2 }, {T], <1x,1X> 2}, {<1X,1X>', 1XxX} {"2, 1,,xX and

{T] ,< ]X,1x>2,' 1XxX.

1.24. Proposition. If (R,j) is a reflexive relation on X then T1j and Tn2J

are retractions.

Proof. Since (X,) < (AXx) < (R,j) there exist morphisms a and

B such that ita = and jB = iX. Thus X = "1<1X,1I> = TliXa = i7ljXa.

Thus r1j is a retraction. Similarly T2j is a retraction.

1.25. Remark, Consider the following products: (XxY,p,,p2), (YxZ,1,12),

(XxYxZ,1if J2,3), ((XxY)xZ, ; 2) and (Xx(YxZ),fl*,T 2*). It is easy to

see there exist isomorphisms

01 = and 2 = <1,*' 2' 2*
( _1 02
(XxY)xZ 01 .-- XxYxZ ------- XX(YXZ)

such that 0101 = PS1, 21 P271' 301 = T2 and 1 02 -7"' 7, 202 P'2

and 1302 = 2 '.

1.26. Definition. Let (R,j) be a relation from X to Y and (S,k) be a re-

lation from Y to Z. Consider the following intersection.

lxj
RxZSRxzy--- ------ .-----<( Xxyx

2
Xx S' -^--------------'X>: (YxZ)
Ixk

Let <71i, 3> denote that unique morphism from XxYxZ to XxZ such that

o1 = 1 and 02< 1,3'> = 3 where uJ and o are the projections cf

XxZ to X and Z respectively.

Let (T',j') be the unique epi-extremal mono factorization of

<,1'"3>Y, and let the codomain of c' (domain of j') be denoted by RoS.

The relation (RoS,j') is called the composition of R and S.

1.27. Examples. In the categories Set, Grp, Ab, and Topi, the composition

of R and S is isomorphic to the set

{(x,y): there exists a ycY such that (x,y)cR and (y,z) ES}.

This is the usual set theoretic composition of relations (which incident-

ally is not the usual notation for the composition of functions when they

are considered as relations).

In the category Top2, the composition of R and S is the closure

of the above set.

1.28. Definition. If (R,j) is a relation on X then K is said to be

transitive if and only if (RoR,j') < (R,j).

A relation on an object X is said to be an equivalence relation

if and only if it is reflexive, symmetric, and transitive.

1.29. Examples. In the categories Set and Top1, transitive relations and

equivalence relations are the usual set theoretic transitive relations

and equivalence relations together with the inclusion maps.

In the category Top2, equivalence relations are closed set theoretic

equivalence relations.

In the categories Grp, and Ab, equivalence relations are subgroups

of the catesian product which are set theoretic equivalence relations.

1.30. Proposition, Let (R, ,j1) and (R2j 2) be relations from X to Y and

let (S ,ki) and (S2,k2) be relations from Y to Z and suppose

(?.,jl) < (K- ,2 ) and (S1,k1) < (S2,k2). Then (rl.OSl,j) < (R2oS2,k).

29

Proof. Since (R,jl) < (R2,J2) and (Sl,kl) < (S2,k2) it is immediate that

(RlxZ,jlxl) < (R2xZ,j2xl) and (XxS1,lxkl) < (XxS2,1lxk) whence

((RxZ) f (XxS1),Yl) < ((R2xZ) ( (XxS2),Y2). Consequently there exists a

morphism a such that the following diagram conmutes.

(R2xZ) CI

(RB1Z) n

j2xl
R2xZ >'.------- ------- (XxY)xZ

(XxS ) Y2
Xx

(XxS1) *'<
XxS2 1 .2 1xk2 2
XxS1 ------------- Xx(YxZ)
ixkI

Thus y 2a = Yl.

Since (RloS1,j) is the intersection of all extremal subobjects

through which Y1 factors (0.21) and since Y1 factors

through (R2oS2,k) it follows that (RloS,j) < (R2oS2,k) which was to be

proved.

1.31. Theorem. Let (R,j) be a relation from X to Y then RoAb, R, and

AxoR are isomorphic relations from X to Y.

Proof. First consider RoAy. From the definition of composition of rela-

tions the following commutative diagram is obtained.

jx!
RxY Y------------- --- (X>:Y)xY

(RxY) (Xx^y) >-- ---------.--- --------- xYxY
Y'

Xx -- -------------- ----* XX (YxY)
Ixi.

Recall that (A ,i ) is the equalizer of the projections pl and p2

from yxy to Y.

It will next be shown that y = y. Let 0i, and 2 be

the projections of XxA to X and A respectively, and let Il"* and IT 2 be

the projections of Xx(YxY) to X and YxY respectively. Then

P1<1' 2 >Y = itY = P 3< 3>Y.

P2<1'pY 2 2 2 202(xi ) 2 1= 2*2 Pliy 2 = 2iY 2 2

P2 2*(]xi )X2 = "32(1xiy )2 = 3= p<2 <1 3>y.

Hence <1, 2I >y y.

Let p1 and P2 be the projections of (Xxy)xY to XxY and Y respec-

tively and let p and p2 be the projections of RxY to R and Y respec-

tively.

Since ITjp1 = 1P1(jxl)X1 = 0 (jxl)x = iT y = ,y, and

Ti2JP I*I = T2P1J- (x)Il = T2rl(jxl)Al = T2Y = T2y, and

< l' I2>y = < 1',T3>y = j T-, then the following diagram commutes.

PI*
Rxy ---- ----- R

(Rxy) (XxAy) -< >.T> XXY

^T ROA,

Thus since (RoA ,j') is the intersection of all extremal subobjects

through which y = y factors (0.21) it follows that

(RcA j') < (R,j).
J
To see that (R,j) < (RoAy,ji ) consider the following commutative

diagrams.

I 1
XxY

--- :>-~ Xx'ixY

<1lW2>

XxY

ly

<1R' 2j
R -- RxY

1R 1*

Recall that (Ay,i ) = (Y,<1 ,1 >) (].21), thus there exists a

morphism o:Y ----->y such that i y = <1 ,1 >.

R 12------ XXA

x

---`c (yx)xy r--

It now will be shown that the following diagram commutes.

jxly
RxY ------- -- (XxY)xy
<1R)' 2J> E 01

1 O
R >------- Xxyxy

<,lj,wT2 -,
XxA >"----1- >-- Xx(Yxy)
Y 1 Xi
yX Y

1 1l(jXl) = Tll(jxl) = IlJP R'2j = 1JlR = r

0 (l(jxl) 2 * = T2j.
3 1 R' 2 22(x1)<12 R 2 2

l0l = lP1 = lj'

i201 = Tr2P = 72j.

<301 j> P=2 1 2 2 1 2j > (2)-

_12(1lXiy) = Tf1 ((Xi) = Pl =

T-2 2(lxiy)<1l j,o r2j> 1= plI2*(lxiY) = Pliy>2

PliyOTf2j = Pl<1( >IT2j = l = ]2j = T2j.

3 92(1xiY)<7lj,o'r2j> = P2 2*(1xiy)< lj,' m2j> = p2iYP2< 111,o-2j> "

P2i 2j = g2T2j 1I 2j = T2j.
Thus by the definition of intersection there exists a unique mor.-

phism C from R to (RxY)f1(XxA ) such that

y = e = 01 < > = 1(j ,) = 2(1xiy)<"7Tj:orT2j>. Thus

<1, 3 >Y3 = <13>0 . But since

'<1' l'>01 = 710 = iPl 22<7,T 1 3>l = 30l = 2j it follows from the

definition of product that < , 3>0 j.

Thus j = y = j'T'C whence (R,j) < (RoAy,j).

The proof that (Rj) E (A oR,j"") follows from analogous arguments.

1.32. Proposition. If (R,j) is reflexive and transitive

(RoR,j') E (R,j).

Proof. Since (R,j) is transitive then (RoR,j') < (R,j).

reflexive then (Ax,iX) < (R,j). Thus (R,j) < .(RoA X, )

(1.31 and 1.30). Hence (RoR,j') E (R,j).

on X then

Since (R,j) is

< (RoR,j') (R,j)

1.33. Remark. As has been remarked in (1.27), if (R,j) is a relation from

X to Y and (S,k) is a relation from Y to Z then in the categories Set, Ab,

Grp, and Top (RoS,j') may be taken to be the set

{(x,z): there exists a yeY such that (x,y)cR and (y,z)ES}

together with the inclusion map j^. Thus the categorical definition of

composition (1.26) yields in these special concrete categories the usual

set theoretic composition.

A similar remark can be made about the definition of the inverse

relation. That is, the categorical definition yields the usual set theo-

retic definition in the categories Set, Ab, Grp, and Top1 to only men-

tion a few. Indeed, the categorical definitions were obtained by analyz-

ing the situation in the set theoretic case.

However, in the category Top2 of Hausdorff spaces and continuous

maps the extremal :Tonomorphisms are the closed embeddings which leads to

the following consequences.

1.34. Example. If (R,j) is a relation from X to Y and (S,k) is a relation

from Y to Z for Top2-objeccs X,Y, and Z. Let T be the following set.

{(x,z): there exists a yeY such that (x,y)ER and (y,z)ES}

Thaen (RoS,j) (clT,j) where "cl" means closure with respect to the top-

ology of XxZ (c.f. 1.27).

Proof. Recall that the extremal monomorphisms are the closed embeddings,

thus RoS is a closed subset of XxZ. Clearly the following diagram

commutes.

Y
Full Text

PAGE 1

Relation Theory in Categories ay TEMPLE HAROLD FAY A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF THE UNIVERSITY OF FLORIDA J.N PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY 07 FLORIDA

PAGE 2

To Dr. George E. Strecker, without whose tactful proddings, infinite patience in proofreadings of handwritten drafts ana helpful suggestions this work would never have been completed .

PAGE 3

TABLE OF CONTENTS Abstract iv Introduction 1 Section 0. Preliminaries 6 Section 1. Generalities IS Section 2. Categorical Congruences I,] Section 3. Categorical Equivalence Relations and Quasi-Equivalence Relations AS Section 4 . Images 63 Section 5 . Unions 75 Section 6. Rectangular Relations 103 Bibliography 120 Biographical Sketch 122 in

PAGE 4

Abstract of Dissertation Presented to the Graduate Counci] of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy RELATION THEORY IN CATEGORIES By Temple Harold Fay March, 1971 Chairman: Dr. George E. Streckcr Major Department: Mathematics The purpose of this dissertation has been to systematically generalize relation theory to a category theoretic context. A quite general relation theory has emerged which is applicable not only to concrete categories other than the category of sets and functions, but also to abstract categories whose objects need have no elements at all. This categorical approach has provided the opportunity to comprehend classical relation theory from a new vantage point, thus hopefully leading to an eventual better understanding of the subject. A relation from an object X to an object Y is a pair (R,j) where j is an extremal monomorphism having domain R and codomain X X Y. By choosing j to be an extremal monomorphism, relations in the category of sets are the usual subsets of the Cartesian product, relations in the category of groups are subgroups of the group theoretic product, and relations in the category of topological spaces are s.ubspaces of the topological product. This latter fact would not be the case if relations would be defined to be merely subobjeccs of the categorical product . Section notes results which are purely categorical in nature and which will be usee 1 , extensively throughout, the sequel. Particular emphasis is give:-; to the epi-extremal mono factorization property and IV

PAGE 5

necessary and sufficient conditions for the existence of this factorization and equivalent forms of the property. In Section 1, the basic machinery for categorical relation theory is developed. For example, such notions as inverse relation, reflexive relation, symmetric relation, and composition or relations are defined and several important results are obtained. Section 2 deals with a categorical definition of a congruence relation. Several algebraic results of Lambek and Cohn are generalized. Equivalence relations and quasi-equivalence relations (symmetric, transitive relations) are studied in Section 3. A quasi-equivalence oi\ an object X is shown to be an equivalence relation on a subobject of X. If R is a set theoretic relation from the set X to the set Y and A is a subset of X then AR = {ycY: there exists aeA such that (a,y)e.R). This definition is generalized in Section 4 and results similar to those r obtained by Riguet are demonstrated. If {(R-jj^): iÂ£l) is a (finite) family of relations from X to Y then the relation theoretic union (^_J^i>J) Â°^ *-' he f ara ily ^ s obtained by iel taking the intersection of all relations from X to Y which "contain" each Rj. If the category being investigated is assumed to have (finite) coproducts then the union of the family considered as subobjeets and the relation theoretic union of the family considered as extremal subobjeets turn out to be given by the unique extremal epi-mono and unique epiextremal mono factorizations of the canonical morphism from the coproduct of the family to X>
PAGE 6

Section 5 deals with unions and the importance of the concept of difunctional relation is brought out. A well known result in set theoretic relation theory is that a partition determines an equivalence relation. In order to obtain this result in its generalized form the existence of an initial object which behaves similarly to the initial object in the category of sets (namely the empty set) is postulated and disjointness becomes a useful categorical notion. Also the notion of difunctional relations was crucial in obtaining the above result. Section 6 deals with rectangular relations and the above result about partitions is obtained. V3

PAGE 7

INTRODUCTION The purpose of this work has been an attempt to systematically generalize relation theory to a category theoretic context. In doing so, several goals have been realized. Firstly, a quite general relation theory has emerged which is applicable. not only to concrete categories other than the category of sets and functions, but also to abstract categories whose objects need have no elements at all. Secondly, taking a categorical approach has provided the. opportunity to comprehend classical relation theory from a new vantage point, thus hopefully leading to an eventual better understanding of the subject. Many relation theoretic results have been rather straightforward to prove in an "element free" setting, once the appropriate machinery has been constructed to handle them. On the other hand it has been surprising to see that some results which are easy to prove in the set theoretic context are much more difficult to show categorically. For example, it is easy to prove that if R is a set theoretic relation from X to Y such that RY = X then RoR" 1 = {(x,z): there exists ysY such that (x,y)eR and (y,z)eR *} is reflexive. This result can be generalized to categories but is no longer easy to prove and the result gains some significance. Another easy result in set theoretic relation theory is that if t y and A v are the diagonals on X and Y respectively then A cR R = RcA, This result is also generalized to categories but "isomorphic as rela-

PAGE 8

tions" replaces "equality" and the result is nc longer easy to prove. Whenever one is generalizing properties care must be taken to be certain that the generalized definitions are really generalizations of the notions being considered and that the proper generalization of the definition is obtained. This seems to be particularly important in category theory. Care has been taken when selecting the basic notion of a relation from an object X to an object Y to be an extremal subobject of the categorical product X*Y; i.e. a pair (R,j) where j is an extremal monomorphism having domain R and codomain XxY . By doing so relations in the category of sets are the usual subsets of the cartesian product, relations in the category of groups are subgroups of the group theoretic product, and relations in the category of topological spaces are subspaces of the topological product. This latter fact would not be the case if relations would be defined to be merely subobjects of the categorical product. Much care has also been taken with the definition of composition of relations (1.26). Using this definition many nice results have been obtained; however, in general, the composition of relations is not associative (1.35). This, at first glance, seems to be pathological and casts doubt on the suitability of the definition of composition of relations. However, the wealth of other important results obtained belies this doubt (see 1.37). Also, some further atonement is yielded by the fact that for rectangular relations composition is_ associative (6.15). Cohn ^3j and Lambek \ 13 j define a congruence in an algebraic setting to be a sebalgeura of the cartesian product which is "compatible 1 ' with the algebraic operations and which is set theoretically an equivalence relation. In this work, a generalized notion of congruence

PAGE 9

is given which is equivalent to the above in algebraic categories and the result that a (categorical) congruence is a (categorical) equivalence relation is obtained. It was found that categorical unions were very difficult to work with. However, by assuming the category being studied had (finite) coproducts as well as being locally small and quasi-complete the notion of union became somewhat easier to handle. For instance, if {(R.,j ): iel} is a (finite) family of relations i i from X to Y then the union (tjR.J) of the family, considered as subiel 1 objects of X*Y is not necessarily a relation from X to Y, since j is not necessarily an extremal monomorphism. The relation theoretic union of the family is obtained by taking the unique epi-extremal mono factorization of j (5. 3 ) or equivalently by taking the intersection of all relations from X to Y which "contain" each R.. If the category being investigated is assumed to have (finite) coproducts in addition to being locally small and quasi-complete then the union of the family considered as subobjects and the relation theoretic union of the family considered as extremal subobjects turn out to be given by the unique extremal epimono and unique epi-extremal mono factorizations of the canonical morphism from the coproduct of the family to X X Y (5.29) > It is also shown that when the category has (finite) coproducts both factorizations respect unions (5.30 and 5-42). Unions are still difficult to handle even with the assumption of (finite) coproducts mentioned above; hence, the notion of a (finite) union distributive category is introduced (5.31). Roughly speaking, this property guarantees that unions "commute" with products, and intersection? and thus unions become "easy" tj handle. Examples of union

PAGE 10

distributive categories show that such categories tend to be more of a topological nature rather than of an algebraic nature. The set theoretic notion of difunctional relation is due to Riguet [_ 2 ? J and its importance has been nc ted by Lambek \13j and MacLane [18 J . A set theoretic relation R is difunctional if and only -1 if RoR oR C R. The categorical definition in view of the fact that associativity cannot be assumed reads: R is difunctional if and only if -I -1 (RoR )oR _< R and Ro (R oR) _< R where "<" is the usual order on subobjects. It is easy to prove, again by choosing elements, that if a set theoretic relation R is difunctional then R = RoR oR. However, the similar result in the categorical setting is much harder to obtain and is rephrased: if R is difunctional then R ~ (RoR )oR and _] R E Ro(R oR) -..'here " = " means isomorphic as extremal subobjects (5.28). A well known result in set theoretic relation theory is that a partition determines an equivalence relation. In order to obtain this result in its generalized form additional hypotheses had to be added to the category being studied. In particular, the existence of an initial object which behaves similarly to the initial object in the category of sets (namely the empty set) had to be postulated and disjointness became a useful categorical notion. Again, examples of such categories are non-algebraic. Also the notion of difunctional relations was crucial in obtaining the above result (-6.20). The excellent reference paper by Riguet j 22 J has been used as a guide for the results of set theoretic relation theory. Indeed, most all of the results contained herein are generalisations of results in j[22j . The papers by Lambek 113 J , I 14 J ? MacLane I ] 8 \ and Bednarck aiid Wallace s j j , / \ provided motivation for many of the general!-

PAGE 11

zations. The basis for the categorical notions has been taken from the papers of Herrlich and Strecker ]_ 7 J , [s ) , Isbell \_ll"]' L 12 J ' and the forthcoming text by Herrlich and Strecker [_9 J (which has greatly influenced this work) . For most of the basic categorical notions the reader is referred to the texts by Mitchell \1\ J , Freyd \k \ and Herrlich and Strecker \, 9 J Â• The work here is begun with a preliminary Section which notes (often without proof) results which are purely categorical in nature and which will be used extensively throughout the sequel. Particular emphasis is given to the epi-extremal mono factorization property and necessary and sufficient conditions for the existence of this factorization and equivalent forms of the property. However, it is not intended that the preliminary section give a complete category-theoretical background. It is expected that the reader be familiar with the basic categorical notions ,

PAGE 12

SECTION 0. PRELIMINARIES 0.0. Remark. It is assumed that the reader is familiar with the basic notions of category theory and hence such basic notions as epimorphism, monomorphism, retraction, section, equalizer, regular monomorphism, coequalizer, regular epimorphism, subobject. and limits shall not be defined. The reader is referred to Mitchell Â£21J and Herrlich and Strecker Â£9] for such notions. All of the following results are proved in detail in Herrlich and Strecker Â£\$3 . Since Theorem 0.21 is vital to this work the proof is sketched here. 0.1. Nota t ion . The category whose class of objects is the class of all sets and whose morphism class is the class of all functions shall be denoted by Set . The category whose class of objects is the class of all groups and whose morphism class is the class of all group homomorphisms shall be denoted by Grp. The category whose class of objects is the class of all topological spaces and whose morphism class is the class of all continuous functions shall be denoted by Top 1 . In a manner similar to that described above, one obtains the following categories: FSet^ finite sets and functions; FGp finite groups and group homomorphi sras ; Ab Abelian groups and group homomorphxsms ;

PAGE 13

SGp semigroups and semigroup homomorphisms ; SGp semigroups with identity and semigroup homomorphisms which preserve the identity; Rng rings and ring homomorphisms; Rng~ rings with identity and ring homomorphisms which preserve the identity; Top,. Hausdorff spaces and continuous functions; CpTÂ„ compact Hausdorff spaces and continuous functions. f Â§ 0.2. Proposition. Let p be a category and let X* Y and YÂ— ** Z be &!, -morphisms. 1) If f and g are monomorphisms then gf is a monomorphism. 2) If f and g are epimorphisms then gf is an epimorphism. 3) If gf is a monomorphism then f is a monomorphism. 4) If gf is an epimorphism then g is an epimorphism. 5) If gf is an isomorphism then g is a retraction and f is a section. 0.3. Remark . In general, an equalizer is a limit of a certain diagram. It is an object together with a morphism whose domain is the object. A regular monomorphism is a morphism for which there exists a diagram so that the domain of the morphism together with the morphism is the equal' izer of the diagram. It is observed in Herrlich and Strecker J, 9 j that certain functors preserve regular monomorphisms while not preserving equalizers, hence one reason for the above distinction between equalizers and regular monomorphisms. In this paper, since we shall not deal with functors, no distinc-

PAGE 14

t.ion shall be made between equalizers and regular monomcrphisir.s; i.e., between the pair (object and morphism) and the morphism alone. Both will be called equalizers. , f 0.4. Proposition. Let g be a category and let X Â—*Â» Y be a Â£morphism. Then the following are equivalent: 1) f is an isomorphism, 2) f is a monomorphism and a retraction, 3) f is an epimorphism and a section, 4) f is a monomorphism and a regular epimorphism, 5) f is an epimorphism and a regular monomorphism. 0.5. Definit ion . Let {A.: iel} be a family of Â£? -objects then the product (TTA.,i r .) of the family is a fe -object I i A together with proiel iel jection morphisms tt^ :TT*Aj_ Â— Â— pAj with the property that if P is iel any j*' -object for which there exist j? -morphisms p.: P Â— Â— Â— **A. for each iel, then there exists & unique morphism X: P Â— Â— *>Â»"f|~A. such that iel ir. A = p. for each iel. y i The dual notion is that of the coproduct (JLLa. ,u.). iel X 0.6. De finitio n. Let {(A., a.): isl} be a family of subobjects of a object X. Then the intersecti on ( O A... ,a) of the family is a ^-object iel Â« ^ A^ together with a morphism a: C\k. > X where fur each i there id iel is a morphism A.: f'\ A. Â• HÂ» A, . such that a-X. = a with the property id that if P is any object for which there exist g -morphisms p: PÂ— Â— *X and p,-: P Â— -~"V_A ,Â• such that a,-p,= p for each iel then there exists a unique morphism X: P "O'/jA; such that aA = p. iel It follows that a is a monomorohi sin .

PAGE 15

0.7. Remar k. The above two definitions are mentioned because of the fundamental role they play in the sequel. They are special limits and are perhaps the most important limits in the categories that will be considered in tnis work. The following theorem is a special case cf a more general theorem dealing with the commutation of limits which can be found in Herrlich and Strecker [9J . A variation of the theorem will be proved in Section 1 (1.5) . 0.8. T heore m. Let {(A., a.): iel} and {(B^b.): iel} be families of subobjects of C -objects X and Y respectively. Then if S has finite products and arbitrary intersections then (,C\ A.)x( f\ B. ) and (~\ (A. xB^) iel iel ' iel are canonically isomorphic. 0.9. Notation. Let {X.: iel} be a family of Â£ -objects and suppose |2 Â— v X.: iel} is a family of fe -morphisms. Then by the definition of product there exists a unique morphism h from Z to TT Xj such iel that if.h = f . for each iel. This morphism h shall be denoted by < f< > i^I Let A and B be g -objects and suppose that a: A rX and b: B > Y are g -morphisms. If P, and PÂ„ are the projection morphisms from A X B to A and B respectively then aP 1 : A X B Â— ^ X and bP 2 ^ A*B Â— Â— Â—> Y, hence by the definition of product there exists a unique morphism g from A X B to X X Y such that 7r 1 g = ao and TT g = bp2Â« This morphism g shall be denoted by axb and shall be called the prod uct of a and b . Let f be a h -morphism from X to Y. If f is a monomorphism then the following notation shall be used: x >Â„. __: _^ Y

PAGE 16

10 If f is an epimorphism then the following notation shall be used f X ~?oY If f is an equalizer then the following notation shall be used XÂ»>~ -*Â• i If f is an isomorphism then the following notation shall be used X"*-v.> Y be d Â— >Y, X Â— Â— * Z, and YÂ— Â— > W 0.10. P roposition . Let A Â— SÂ»X, B be Â£? -morphisms. Then (c*d) (a*b) = ca-db. a b 0.11. Proposition. Let A Â— Â— Â— > X and B > Y be monomorphisms (respectively, sections, isomorphisms) then a*b is a monomorphism (section, isomorphism) . 0.12. Remark. A partial order may be defined on the subobjects of an object in j^ in the following way: If X is a fe -object and (A, a) and (B,b) are subobjects of X; i.e., a and b are monomorphisms with codomain X and domains A and B respectively, then (A,a) <_ (B,b) if and only if there exists a morphism c from A to B such that be = a . B >4 I A .-v X By an abuse of language, if (A, a) < (B,b) then (K.b) is said to contain (A, a) and the morphism c is sometimes called the inclusion of (A, a) into (B,b). It is easy to see that if (A, a) <^ (B,b) an J (B,b) < (A, a) then the morphism c is an isomorphism. In this case,

PAGE 17

11 (A, a) and (B,b) are said to be isomorphic as_ subobjects of X. This is a stronger condition than A and B just being isomorphic objects in the category g . The following notation shall be used to denote the case where (A, a) and (B,u) are isomorphic as subobjects of X: (A, a) I (B,b). Sometimes it is written (inaccurately) that A < B or that A and B are isomorphic as subobjects of X. When this is done, the morphisms a and b should be clear from the context. It is immediate that (A, a) = (B,b) if and only if (A, a) <_ (B,b) and (B,b) <_ (a, a). Thus the relation "_<" on subobjects is easily seen to be a partial order up to isomorphism as subobjects. 0.13. Defini tio n. Let f from X to Y be a fc -morphism. f is an ext remal mo n o mo r p h i s m if and only if f is a monomorphism and whenever f = gh and h is an epimorphism then h is an isomorphism. If f is an extremal monomorphism the following notation shall be used: f X >>Â— >Y The dual notion is that of an extremal e pimorphism and is denoted: X Â— Â«\$*> Y If f is an extremal monomorphism f: X Â— -*-Y, then (X,f) is called an extre mal sub obj ect of Y. 0.14, Remark . The definition of extremal monomorphism is due to Isbell {^llj . The concept of extremal monomorphism is important since it yields what shall be called the "image" of a morphism (see 0.18),. 0.15. Example? . In the categories Set, Grp , Ab and _FGp_, extremal monoiticrpbi.jrr.s are precisely the monomorphisms (i.e., one-to-one morphisms).

PAGE 18

12 In the categories Top and CpT extremal monomorphisms are precisely the embeddings. In the category Top they are the closed embeddings. f . 0.16. Pro posit ion. If X Â— *-Y is a g -morphism such that f = gh and f is an extremal monomorphism then h is an extremal monomorphism. 0.17. Proposition. If X *-Y is a fe -morphism then the following are equivalent: 1) f is an isomorphism, 2) f is an epimorphism and an extremal monomorphism, 3) f is a monomorphism and an extremal epimorphism (c.f. 0.3). 0.18. Def ini tion. A category jÂ£> is said to have the unique epiex trema l mono facto rizatio n property if for any )Â° -morphism XÂ— sÂ»Y, there exist an epimorphism h and an extremal inonciuorphism g with f = gh such that whenever f = g'h' where g' is an extremal monomorphism and h ? is an epimorphism then there exists an isomorphism a such that the following diagram commutes. If Â£ has the unique epi-extremal mono factorization property and if f = gh where h is an epimorphism and g is an extremal monomorphism, then the pair (h,g) shall be used to designate the epi-extremal mono factorization of f. The extremal subobject (Z,e) of Y is called the

PAGE 19

13 image of X under f . Sometimes (Z,g) is referred to as the image of f . The notion of the uniqu e extremal epi-mono factorization proper ty is defined dually. If H has the unique extremal epi-mono factorization property and f = gh where g is a monomorphism and h is an extremal epimorphism then the pair (tug) shall be used to designate the extremal epi-mono factorization of f . The subobject (Z,g) of Y is called the sub image of X u nder f. Sometimes (Z,g) is referred to as the subimage of f . 0.19. Definition . A category g is said to have the diagonalizing pr operty if whenever gh = ab such that h is an epimorphism and a is an extremal monomorphism, then there exists a (necessarily unique) morphism t, such that E, h = b and a Â£ = g. I **Â• Y v W >Â•> > Z 0.20. Theorem. Let C be a locally small category having equalizers and intersections. Then the following are equivalent: 1) ig has the unique epi-extremal mono factorization property, 2) g has the diagonalizing property, 3) the intersection of extremal monouiorphisms is an extremal monomorphism and the composite of extremal monomorphisms is an extremal monomorphism, 4) if g 5 has pullbacks and if (P,q,3) is the puliback of f and g where fg = get and f is an extremal monomorphism then a is an extremal

PAGE 20

14 monomorphism. 5) if ^ has (finite) products then the (finite) product of extremal monomorphisms is an extremal monomorphism. 0.21. Theorem . If Y> is locally small and has equalizers and intersections then t* has both the unique epi-extremal mono factorization property and the unique extremal epi-meno factorization property. Pr oof . (sketch) . First we will show the existence of the unique extremal epi-mono factorization property. If f from X to Y is any fc> -morphism then let (OE.,e) be the intersection of the family {(E.,e.): ieJ} of all subobjects of Y through which f factors. Then it follows that e is a monomorphism and that f factors through e; i.e., there exists a morphism h such that f = eh. Now, to see that h is an epimorphism suppose a and 3 are \% -morphism? such that ah = gh. Let (E,k) be the equalizer of a and (3. It follows from the definition of equalizer that there exists a morphism g such that kg h. Â»* Y Thus it follows that f factors tHrough ek and sines ek is a monomorphism then there exists a morphism X: f\ E. * E such that ek\ = e From this it follows that k is rn isomorphism whence a = 3 and so h is an epimorphism. Next it will be shown that h is an extremal epimorphism. Suppose h = h 1 h ? where h., is a monomorphism. Then eh., is a monomorphism through

PAGE 21

15 which f factors. From this it follows, as above, that h is an isomorphism and hence h is an extremal epimorphism. Suppose f = g'h' where g' is a monomorphism and h 1 is an extremal epimorphism. Then since g' is a monomorphism vhrough which f factors there exists a morphism x from P\E. to che codomain of h' (domain of g') such that e = g'x. Since e and g' are monomorphisms, it follows that h' = xh and that x is a monomorphism. Since h' is an extremal epimorphism it follows that x is an isomorphism. Thus fe has the unique extremal epi-mono f actorizaticn property. Now suppose that ge = mE where e is an epimorphism and m is an extremal monomorphism. It will be shown that there exists a morphism o from the codomain of e to the domain of m such that oe = h and mo g. Let (f\k.,a) be the intersection of the family {(A., a.): isl} iel of all subobjects of the codomain of g (codomain of m) through which g and m factor. This family is non-empty since both g and m factorthrough the identity morphism on the codomain of g. It follows that both g and m factor through a. Thus there exist morphisms a, and a such that the following diagram commutes. m It will be shown next that Â£Â„ is an epimorphism. Suppose ot* and are L> -morphisms for which ot*aÂ„ 3*a.. Let (E*,k*) be the equalizer

PAGE 22

16 of a* and 3*. Jt follows from the definition of equalizer that there exists a morphism b^ such that k*!^ = a 2 since 0*3 = B*a_. Since the diagram commutes it follows' that a*a e = 3*a,e. But e is an epimorphism hence a*aj = Â£ v a so that by the definition of equalizer there exists a morphism b~ such that k*b~ = a-^. Thus it follows that m = ak*b-, and g = ak*b2 anc ^ so both m ant ^ S factor through ak* from which it follows that k* is an isomorphism. Hence a* = g* and a, } is an epimorphism. But in is an extremal monomorphism and m = aa and aÂ„ is an epimorphism. Thus a is an isomorphism. Thus defining a = a^a it follows that the following diagram commutes and ^ has the diagonal ization property. Hence K has the unique epi-extrema.l mono factorization property (0.20) 0.22. T heorem . Let K be any category then the following are equivalen 1) t^ is (finitely) complete, 2) Q^ has (finite) products and (finite) intersections, 3) fe> has (finite) products and equalizers, A) Jg has (finite) products and pullbacks. 0.23. Definition. A category 7^ is said to be quasi-complete if j^ has finite products and arbitrary intersections. t:

PAGE 23

17 0.24. Exa mples . The categories FSet and FGp are quasi-complete categories which are not complete. The categories Set, Top , Top ^ , CpT ? , Crp, Ah, Ring , and SGp are quasi-complete. 0.25. Re marks . A quasi-complete category is finitely complete but is not necessarily complete as the examples FSe t and FGp above show. Also, a locally small, quasi-complete category has both the unique extremal epi-mono factorization property and the unique epi-extremal mono factorization property (0.20 and 0.21). It can be shown that the unique epi-extremal mono factorization of a morphism can be obtained by taking the intersection of all extremal monomorphisms through which the morphism factors. It has been shown that the unique extremal epi-mcno factorization property is obtained by taking the intersection of all subobjects through which the morphism factors (0.20). These characterizations shall be used frequently in the sequel.

PAGE 24

SECTION 1. GENERALITIES .1.0. Sta nding Hypothesis . Througho ut the entir e paper it will be assumed that /Â» jls a_ lo cally small , quasic omplete (fi nite prod ucts and arbitrary intersections) c ategory . As noted in the preliminary section fe enjoys the unique epi -extremal mono factorization property. 1.1. Exampl e s. Many well known categories are locally small, and quasicomplete. Among such are the categories: Set , Top. , Top , Grp, Ab, SGp_, SGjp 1 , Rng, Rng 1 , CpT 2 , and FGp . 1.2. Defi nition. Let X and Y be g -objects. A relation R from X to Y is an extremal subobject of X>
PAGE 25

19 In the categories Top ,, , and CpT ? , relations are closed subspaces of the Cartesian product together with the inclusion map. 1.5. Proposition , Let X and Y be K -objects and let (A, a) and (B,b) be extremal subobjects of Y. Then Xx(AftB) and ^X^AjCi (X X B) are isomorphic relations from X to Y. Proof . Consider the following commutative diagrams. (X*A)f\ (XxB) Consider also (Xx(AAB), l y x c = Yi)Since extremal subobjects are closed un-ler intersections and products (0.20; y^ and Yo ar e extremal monomerphisms. Since (l x x a) (! x xA A ) = l x *c = z Yj and (l x xb) (l x x * B ) = l x x c = Yj then by the definition of intersection there exists a unique morphism o from Xx(AAB) to (XxA)/-\(XxB) so that Y 9 Â°" = Yj and the following diagram cc mutes. Thus: -jm(Xx(AAB), Yi ) 1 ((XxA)n(XxB), Y? ) '2'

PAGE 26

20 X*A Xx(A.^B) XxB Nov let (it , it ), (tt' it'), (o , p ) and (p' pp be the projections of XxY, Xx(AAB), XxA, and XxB respectively. Observe that: V-j " ^(V a)X l = Vl X l = *l (1 X Xh)X 2 'Vl X 2 V2 = 1r 2 (1 X XaU l = Sp 2 X l = ^2 (1 X Xb)X 2 = bp 2 X 2' Thus by the definition of intersection there exists a unique morphism Â£ from (XxA) A (XxB) to AAB such that r.E = ttÂ„yÂ„ and thus by the definition Â£ Â• 2 of product there exists a unique morphism Â£ from (XxA)A(Xx B) to Xx(AAft) such that Â£ = ; i.e., tt'^ = 1 yÂ« and ir^ = E. Now y,? = (lÂ„ x c)5 hence tTjY^ ri ^l^ = n l Y 2 and Tr 2 Y l ? = ClT 2 ? = cI = ^2' ThuS Y l^ = ''2' wheuc ((XxA)A (XxB), y) < (Xx(AriB), y) *it 2 (XxA)AXxB)

PAGE 27

21 1.6. Notation. Let X and Y be t -objects and let (X*Y,7T ,it ) and (YxX,p ,p ) be the indicated products of X and Y. Then there exists a unique isomorphism from X*Y to YxX, denoted by , such that the following diagram commutes. Xx Y V> ^ Â— ~ X Â»Â— 4> YxX I VtY Note that: 2 .T^xp^p^ = l y> XxY' 1.7. Definition. Let (R,j) be a relation from X to Y and let (t,j*) be the unique epi-extremal mono factorization of j (see 0.18). The codomain of x (domain of j*) is denoted by R1 and (R~ ,j*) is called the inverse relation of (R,j) or more simply, when there is little likelihood of confusion, the inverse of R. Â£/YxX 1.8. E xamp le. In the categories S3t , Topi , TopT , Grp , Ab, and FGp ti 1 >: X.xY Â— Â— Â— Â— Â— Â— -> YxX is defined by (x,y) = (y,x); hence, if (R,j) is a relation from X to Y then R _1 {(y,x): (x.y)Â£R} with j* tiie inclusion map of R -1 into YxX. 1.9. Jrx>p_osit_iop. Ii (R>j) is a relation from X to Y the R and R -i are iso^cr^hic obiects of f* .

PAGE 28

22 Proo f . Since <7i2.ifi> is an isomorphism and j is an extremal monomorphism then j is an extremal monomorphism. But 7r l > J = J* T * Thus since i is an epiruorphism then from the definition of extremal monomorphism it follows that t is an isomorphism. 1.10. Def inition. If (P M j) is a relation from X to X then R is said to be symm etric if and only if (R ,j*) <^ (R,j). 1.11. Proposition. Let (R,j) be a relation from X to Y. Then the inverse relation ((R~ )~ ,j#) of (R ,j*) and (R,j) are isomorphic relation?.. Pro of . Consider the following commutative diagram. R>> R-l >Â£ J# Â— *> XxY ->YxX

-*. XxY Since the two inner squares commute the outer rectangle commutes. Both of t and t# have been shown to be isomorphisms (1.9). And, as also has been observed: -po >Pi ><7T 2 j 71 ! > ~ ^-XxY (1*6) Â• Consequently, t;'/t is an isomorphism and j j#(t#t). Thus (R,j) I ( (R -1 ) _1 , j//) . 1.12. Proposition. Let (R,j) and (S,k) be relations from X to Y, Then (S,k) <_ (R,j) if and only if (S -1 s k*) < (R" 1 ,^*). Proof. Consider the following commutative diagram.

PAGE 29

23 R -*Â» Y*X If (S,k) < (R,j) then there exists a morphism a: S Â•* R such that jet = k. Define 3 = tot . Then j*B = j*xaT 1 = jo;t 1 = k*TT -1 = k*. Thus (S -1 ,k*) _< (R -1 ,j*). If (S -1 s k*) <_ (R -1 ,j*) then by the above, ((S -1 ) _1 ,k#) _^ ((R -1 )1 ,^) thus (S,k) < (R,j) (1.11). < TT 2 ,TT 1 >kf 1.13. Corollary. If (R,j) is a symmetric relation on X then (R,j) < (R _1 ,j*) whence (R,j) E (R -1 ,j*). Proof. Since (R,j) is symmetric (R -1 ,j*) Â± (R,j). Thus (R,j) = ((R" 1 )1 ^*) 1 (R _1 ,j*) (1.11 and 1.12). Consequently (R,j) = (R -1 ,j*). 1.14. Definition. Recall that since g* is quasi-complete it has equalizers, thus for each g -object X let (A, , i ) denote the equalizer of t-, and 7T where tt-, and ttÂ„ are the projections of X*X. Since i is an equalizer it is an extremal monomorphi sm. Hence (AÂ„.i v ) is always a relaA ' A tion on X (called the d iagonal of X*X) . A relation (R,j) on X is said to be re flexive on X provided chat (A x ,i y ) < (R,j). 1.15. Exa mple . In the categories Grp, Ab, Set, Top , To p ? , and CpT~ , it follows that A v z {(x,x): xeX} C XxX with the inclusion map. A

PAGE 30

24 1.16. Pro position . For any (X -object X, <1, 2> TI l > ix ~ *X' 1' nus; (A x ,i x ) E (A x -l,i x *). Proof. Consider the following commutative diagram. XxX Â» <7To j TTi > ^>Â» XXX TT 1 Â± x = rr 2 i x = TT X i x = 7t 2 i x . Thus the epi-extremal mono factorization of ' i is (i. ,:'). A Ay A 1.17. Corollary . Let (R,j) be a relation on X, then (R,j) is reflexive on X if and only if (R ,j*) is reflexive on X. Proof. If (A x ,i x ) 1 (R,j) then (A x ,i x ) = (A x _1 ,i x *) < (R _1 , j*) (1.16 and 1.12). Conversely if (A x ,i x ) <_ (R _1 , j*) then ( A X>%) E (V 1 '^")! ((R" 1 )" 1 ,:?/) = (R,j) (1.16, 1.12. and 1.11). 1.18. Propo sition . Let (R,j) and (S,k) be relations from X to Y. Then the relations (RHS)" 1 and (PC l r\S~ l ) are isomorphic relations. Pr oof . According to the definitions of intersection and inverse relation we have the following commutative diagrams. > XxY Nt-. s-r > YxX

PAGE 31

25 RAS >=?Â» X*Y Â»^(RAS)1 * 1