i
Chapter I
PRELIMINARIES
This chapter contains some basic concepts and results on semigroups
and topological vector spaces which are frequently needed in the sequel. It
also serves to fix notational and terminological conventions followed in this
work.
Elementary notions of set theory and of general topology as in Chapters
0, 1 and 3 of [24] are assumed to be known; so also are simple properties of
algebraic systems (such as groups, rings and vector spaces) and of Banach
spaces and Hilbert spaces as contained in Chapters 8, 9 and 10 of [38].
These are not explicitly stated here, and are often used without comment
in the sequel.
1.1. Semigroups
In this section, we present certain basic definitions and results concerning
semlgroups.
Definition 1.1. A nonempty set S, with a binary operation (x, y) 1+ xy
from S x S to S is said to be a semigroup if x(yz) = (xy)z for all x, y, z E S.
If A, B are subsets of a semigroup S, we write AB to denote the set
{xy : x E A, y E B}. Usual notational simplifications are used in the case
of singletons, such as writing Ab for A{b}.
A semigroup S may not contain an element e with xe = ex= x for all
x E 5; but if such an element exists in 5, it is easily seen to be unique.
1
1.1. Semigroupll 2
Definition 1.2. A semigroup 8 is said to be a monoid, if there exists an
element 1 E 8 with xl = Ix = x for all x E 8. The element 1, which is
necessarily unique, is called the the identity of 8.
It is easy to adjoin an identity to a semigroup which is not a monoid.
Let 5 be a semigroup and let 1 be an object not in 5 (for example, we can
take 1 = 5, since by the axiom of regularity of set theory we have 5 i 5).
Extend the binary operation on 8 to 5 U {1} by xl = 1x = x for all x E 5
and 1 1 = 1. Then it is easily seen that 5 U {I} is a monoid. We denote
this monoid by 8+ and define
5 {5, ,
1 _
 5+
if 5 is a monoid;
if 5 is not a monoid .
Thus in any case, 8 1 is a monoid containing S.
Definition of subsemigroups is analogous to that of subsystems of other
algebraic systems.
Definition 1.3. A nonempty set T of a semigroup 5 is said to be a sub
semigroup of SifT itself is a semigroup under the binary operation induced
by that on 5.
Since the induced binary operation on T is necessarily associative, it
follows that T is a subsemigroup of S if and only if xy E T for all x, YET.
By analogy with ring theory, we define ideals of a semigroup as follows:
Definition 1.4. A nonempty subset A of a semigroup 5 is said to be a left
ideal if 5 A ~ A, a right ideal if A5 ~ A and a (two sided) ideal if it is
both a left ideal and a right ideal.
1.1. Semigroupl! 3
If a is an element of a semigroup S then it is easily seen that Sla is the
smallest left ideal, aS I is the smallest right ideal and SlaSI the smallest
ideal containing a. These are called respectively, the principal left ideal, the
principal right ideal and the principal ideal generated by a.
AB usual, homomorphisms are defined to be structure preserving maps.
Definition 1.5. A map ~ from a semigroup S to a semigroup T is said
to be a homomorphism if ~(xy) = ~(x)~(y) for all x,y E S. An injective
homomorphism is called all isomorphism. If there exists an isomorpllism
from S onto T, then we say that Sand T are isomorphic and write S ~ T.
AB in other algebraic systems, there is a close connexion between ho
momorphic images and quotient structures. We define a quotient of a semi
group as follows:
Definition 1.6. An equivalence relation p all a semigroup S is said to be
a congruence if for a, b, c, dES, apb and cpd imply acpbd.
If p is a congruence on a semigroup S and a E S, then the congruence
class {x E 8 : xpa} containing a is denoted by p(a) and the set of a.ll
congruence classes in S by S / p. It is easily seen that
p(a)p(b) =p(ab) .
is a well defined, associative operation on S/ p. The ~emigroup S/ p with
the above operation is called the quotient se.migroup of S modulo p [8].
It is well known that congruences on a group are induced by normal
subgroups and vice versa. Since this fact is sometimes made use of in our
later work, we make an explicit statement ([20], 1.8.2; [21], 1.5.l(i)).
1.1. Semigroupll 4
Proposition 1.7. Let G be a group with identity e.
(1) If N is a normal subgroup of G then
PN = { (x, y) E G XG : xyl EN}
is a congruence on G and
PN(X) = Nx = xN for all x E G.
Also
PN = {(x, y) E G X G : x1y EN}.
(2) If p is a congruence on G, then N = p( e) is a normal subgroup of G
and p = PN. •
If P is a congruence on G, then G/ P can be easily seen to be a group
with p(X)l = p(x 1) for all x E G. If N = p(e), then we denote the
quotient group Glp by GIN (d. [21],1.5.4).
It can be shown that any homomorphic image of a semigroup can be
realized as a quotient of S. For this we require the following
Definition 1.8. If X and Yare sets and 4> a map from X to Y, then the
kernel of 4>, denoted ker 4>, is the relation on X given by
ker4> = {(x,x') E X x X: 4>(x) = ¢(x')}.
The following result is proved in Theorem 1.5.3 of [20].
• i
'.
t· ....· .
"tP1)..;: GJ~b44Jf
/1
. .,," 1.1. Semigroups 5
n'
Theorem 1.9. If p is a congruence on a semigroup 5, then the map 7r from
5 to SI p defined by 7r(x) = p(x) is a surjective homomorphism. Conversely,
if 5 and Tare semigroups and 4> from 5 to T is a homomorphism, then p =
ker 4> is a congruence on 5, and 51 p is isomorphic with the subsemigroup
4>(8) ofT under the map l/J on 81 p defined by 1f;(p(x)) = 4>(x). • .
If p is a congruence on a semigroup S, then the homomorphism 7r
defined above is called the canonical projection of 5 onto Sip. We also note
the following result (see Theorem 1.5.6 of [20]).
Proposition 1.10. If p and a are congruences on a semigroup 5 such
that p ~ a, then p(x) H a(x) is a homomorphism of 51 ponto Bla. Also
alp = {(p(x),p(y)) E 51p X 51p: (x,y) E a} is a congruence on 51p and
(51 p)/(al p) ~ 51a under the map (al p)(p(x)) H a(x). •
We now describe certain equivalences on a semigroup, which are of con
siderable importance in understanding its structure. Let 8 be a semigroup.
,
We define
.c = {(a, b) E 8 X 5 : 5 I a = 8 I b}
R. = { (a, b) E S X5 : aS I = bS I }
J ={(a, b) E 5 x 5 : 51 a5 1 = 51 b5 1}
~=.cnR
D=.cvR
where .c V R denotes the smallest equivalence containing .c and R..
The following result is immediate (cf. [201, II.l.I).
Proposition 1.11. The relations .c, R, J,}f and D are equivalences on 5.
For a, b E 5, a.cb if and only if there exist x, y E 51 with a = xb and b = ya;
1.1. Semigroupll 6
aRb if and only if there exist x, yES 1 with a = bx and b = ay, and aJb
if and only if there exist x, y, U, v E Sl with a = xby and b = uav. Also
! u R. ~ f) ~ J. II
The description of f) can be simplified a.'5 follows (d. [20],11.1.3,1.5.15).
Proposition 1.12. In any semigroup S, f) =! 0 R. = R. o!. Consequently
for a, b E S the following are equivalent:
(1) af)b;
(2) there exists c E S with a!cRb;
(3) there exists c E S with aRc!b. •
The five equivalences defined above are collectively called Green's re
lations. For xES, the !class, Rclass, Jclass, }/class and f)class con
tabling x will be denoted respectively by Lx, Rx, Jx,Hx and Dx. Since £" R
and J are defined in terms of principal ideals, we can define partial orders
on Sf!, Sf R. and Sf J by
La ~ Lb ¢=:> Sia ~ SIb
Ra ~ Rb {:=;> aS l ~ bS l
Ja ~ Jb {:=;> Sl aSl ~ SlbS 1,
These relations can be given simple descriptions in the case of idem
patents.
Definition 1.13. An element e of a semigroup S is said to be idempotent
if e2 = e. The set of all idempotents in a subset T of S is denoted by E(T).
1.1. SemigrouplI 7
If e is an idempotent in a semigroup S and a E S with L4 ~ L e , then
by definition there exists xES 1 with a = xe so that ae = xe 2 = xe =
a. Conversely if a . :. . ae, then by definition we have La ~ Le. Similar
computations yield the following
Proposition 1.14. Let e be an idempotent in a semigroup S and let a E S.
Then La :5 Le if and only if ae = a and Ra :5 Re if and only if ea = a. In
particular, for e, I E E(S), ell if and only if el = e and Ie = I; eRI if
and only if el = f and Ie = ej and eJi I if and only if e = f. 
An important concept in the theory of semigroups is that of regularity.
Definition 1.15. An element x of a semigroup S is said to be regular if
there exists x' E S with xx' x = x. S is said to be a regular semigroup if all
elements of S are regular.
The following result shows that regularity is a property of Vclasses
rather than of individual elements of S ([20], II.3.l).
Proposition 1.16. If a is a regular element of a semigroup S, then every
element of Da is regular. _
Thus if D is a Vclass of a semigroup, then either all elements of Dare
regular or no element of D is regular. In the former case, D is Jaid to be
a regular Vclass. In particular, since an idempotent e is evidently regular,
De is a regular Vclass. Conversely, it can be shown that a regular Vclass
contains an idempotent. In fact we have the following result ([20}, 11.3.2).
Proposition 1.17. In a regular Vclass, each lclass and each Rclass
contains at least one idempotent. _
1.1. Semigroupll 8
An idea closely related to regularity is that of an inverse of an element.
Definition 1.18. Let x be an element of a semigroup S. If there exists
x' E S with xx'x = x and x'xx' = x' then x' is said to be a (generalized)
inverse of x.
It must be noted that there is considerable variation in literature re
garding the term "generalized inverse" (see for example, [34]). If an element
x has an inverse, then x is obviously regular. An easy computation estab
lishes the following converse.
Proposition 1.19. If x is a regular element of a semigroup Sand xx'x = x
for x' E S, then x" = x'xx' is an inverse of x. •
Note that an element of a semigroup can have more than one inverse.
The next result shows how inverses of an element are distributed within a
D class (d. [20], II.3.5).
Theorem 1.20. If a is a regular element of a semigroup S and a' an inverse
of a in S, then a'Da and e = a' a, 1 = aa' are idempotents with af,eRa' and
a' £1 Ra. Conversely, if a is an element of Sand e, 1 idempotents in S with
e£aRI, then a is regular and there exists a unique inverse a' of a in S with
eRa' £1 and a' a = e., aa' = f. •
It is illuminating to depict the content of the above result by means of
what Clifford and Preston call an "egg box picture" [8]. In such a. picture,
each DcIass of a semigroup is represented by a. square which is divided
into vertical rectangles representing ! classes and the horizontal rectangles
representing Rclasses. Each small square formed by the intersection of such
1.1. SemigrouplI 9
rectangles represents an )Iclass. Using such a scheme, the situation in the
above result is given by the diagram below.
a f
e a'
The result says that if a' is an inverse of a, then a, a' are in the same
large square and there exist idempotents aa', a' a in the other two corners,
and conversely, if a is in a corner determined by two idempotents e and J
then one can find an inverse a' of a in the fourth corner with a' a = e, aa' = f.
If e is an idempotent in a semigroup S and a E He, then by 1.14,
ae =ea = a and by 1.20, there exists a' E He with aa' =a'a =e. Thus He
is a subgroup of S. Again, if G is a subgroup of Sand e, the identity of G,
then e is an idempotent of S, and it can be easily seen that G ~ He. This
gives the following
Proposition 1.21. If e is an idempotent in a semigroup S, then He is a
subgroup of S with identity e. Also, any subgroup of S is contained in the
}Iclass of an idempotent in S. •
Another result concerning the Vclass structure of a semigroup is. the
CliffordMiller theorem ([8],2.17).
1.1. Semigroupe 10
Theorem 1.22. Let S be a semigroup and a, b E S. Then ab is in Ra n Lb
if and only if ~ n La contains an idempotent. •
The eggbox picture of the situation is given below.
a ab
e b
A useful partial binary operation on a semigroup S can be defined as
follows. For x, yES, we define
x *y  {X y , if xy E Rx n LlI
 undefined otherwise.
This is called the trace product of x and y. Thus the equation z = x*y means
z :::: xy and emphasizes the fact that z == xy E Rz n L.,j (or equivalently
that there exists an idempotent in Lx n RlI ).
Next we consider some concepts related to the idempotents of a semi
group, which are introduced in [31].
Definition 1.23. Let S be a semigroup. On the set E(S) of all idempo
tents in S, we define the relations wi and wr by
ew' f ¢::> ef:::: e and ew r f ¢::> f e == e
1.1. Semigroups 11
for e, f E E(S); also we define
w =wi nw r •
It is ea.sily seen that ew l f if and only if L e :::; L f and ew r f if and only
if Re ~ Rf and hence wr and wi are quasi orders (reflexive and transitive)
on E(S) and w is a partial order. For e, f E E(S), we define
M(e,J) = {h E E(S): hw'e and hwrf}.
On M( e, f) we define the relation < by
g < h ¢::::> egwreh and gfw'hf
for g,h E M(e,f). This defines a qua.si order on M(e,f). The set
S(e,f) = {h E M(e,f) : g < h for all g E M(e,f)}
is called the sandwich set of e and f.
The importance of these concepts is that an abstract (axiomatic) char
acterization of the set of idempotents of a regular semigroup can be given
in terms of these [31]. Since we will not be making use of this in our work,
we do not discuss the details.
Another important set associated with a pair of idempotents e, f of a
semigroup is defined by
Sl(e,f) = {h E lvf(e,J) : ehf = ef}
The following result is from [31], Theorem 1.1.
.'
1.1. Semigroups 12
Theorem 1.24. Let S be a semigroup and e, f E E(S). Then
81(e,J) = {h E M(e,J) : h is an inverse of ef}
and 81 (e, J) ~ 8 (e, J). Also ef is regular if and only if 8 1(e, J) is non empty
and in this case, Sl(e,J) = S(e,J). 
It is shown in Proposition 2.5 of [31] that if el,e2,ft,h E E(S) with
el,Ce2 and ftRh, then S(el,ft) = S(e2,h). It can be easily seen that we
also have Sl(el,ft) = Sl(e2,h) in this case. Now if x and yare regular
elements of S, then there exist e,f E E(S) with elx and fRy by 1.17; and
so we can unambiguously define
S(x,y) = S(e,J) and Sl(X,y) = Sl(e,J).
Also if elx and fRy, then we have eflxf Rxy so that efDxy, and
hence it follows from 1.16 that xy is regular if and only if ef is regular.
This together with 1.24 gives the first part of the following; the proof of the
remaining part can be found in (31], Theorem 1.2.
Theorem 1.25. Let S be a semigroup and x, y regular elements of S.
Then xy is a regular element of S if and only if Sl(X,y) is non empty. Also
if hE Sl(X,y) then xy = (xh) * (hy) and y'hx' = (y'h) * (hx') is an inverse
of xy for any choice of inverses x' and y' of x and y. 
In Proposition 1.1 of [32}, it is shown how a partial order can be defined
on any regular semigroup.
1.1. Semigroup" 13
Theorem 1.20. Let S be a regular semigroup. Then the relation ~ on S
defined by
x ~ y ¢=} Rx ~ Ry and there exists e E E(R x ) with x = ey.
is a partial order on S whose restriction to E(8) is the relation w. •
This partial order is called the natural partial order on S. In the sequel,
whenever we write x ~ y for x, y in a regular semigroup, the symbol ~
stands for this order. .From Proposition 1.2 of [32] we have the following
alternate characterizations of the natural partial order.
Proposition 1.27. Let x, y be elements of a regular semigroup S. Then
the following are equivalent:
(1) x ~ y;
(2) Lx ~ L y and there exists e E E(L x) with x = ye;
(3) For each f E E{Ry) there exists e E E(Rz:) with ew f and x == ey;
(4) For each f E E(L y) there exists e E E(L x ) with ewf and x == yeo •
Another important property of the natural partial order is given in
Theorem 1.6 of [32].
Theorem 1.28. Let x, y be elemen ts of a regular semigroup S. Then there
exist tt, v E S with tt ~ x, V ~ Y and xy = u * v. •
In fact it suffices to choose h E Sl(X,y), define u = xh, v = hy and use
. 1.25.
We now consider some stronger forms of regularity. We begin with the
following from [3ul.
1.1. Semigroupll 14
Definition 1.29. An element x of a semigroup S is said to be completely
regular if there exists x' E S with xx' x = x and xx' = x' x. A semigroup S
is said to be completely regular if all elements of S are completely regular.
The result below, which is a slightly modified version of Proposition
IV.1.2 of [36], gives other characterizations of complete regularity.
Proposition 1.30. The following are equivalent for an element x of a
semigroup S:
(1) x is completely regular;
(2) tllere exists an inverse x' of:l; with xx' = x'x;
(3) t}lere exists e E E(S) with e)l Xi
(4) there exists an inverse x' ofx with x')lx. •
Another type of regularity can be defined for elements of a monoid.
Definition 1.31. Let S be a monoid. An element u of S is said to be a
right unit if there exists u' E S with uu' = 1 and a left unit jf there exists
u' E 8 with = 1. u is said to be a. unit if there exists u' E S with
tt' u
uu' = u'u = 1. An element x of S is said to be unit regular if there exists
a unit u in S with xux =x. S is said to be a unit regular semigroup if all
elements of S are unit regular.
The following characterizations of unit regularity are easily proved (cf.
[19], 7.3.4).
Proposition 1.32. Let S be a monoid. Then the following are equivalent
for xES:
(1) x is unit regular;
1.1. Semigroup!l 15
(2) there exists a unit u in 8 and e E E(8) with x = uej
(3) there exists a unit u in 8 and e E E(8) with x = eu. •
We next consider some classes of semigroups, which are used in the
sequel.
If a semigroup S contains an element 0 such that xO = Ox = 0 for all
XES, then we say that 0 is a zero element of S and that S is a semigroup
with zero. It can be easily shown that there can be at most one zero element
in a semigroup.
Definition 1.33. A semigroup 8 is said to. be simple if it contains no
ideals other than itself. A semigroup 8 with zero element 0 is said to be
Osimple if 8 2 =I 0 and S, {OJ are the only ideals in 8.
Recall that the idempotents of a semigroup S are partially ordered
with respect to the relation w defined by ewf iff ef = fe = e (see 1.23).
An idempotent e of 8 is said to be primitive if e i 0 and if e is a minimal
element with respect to w.
Definition 1.34. A semigroup S is said to be completely simple (com
pletely Osimple) if S is simple (Dsimple) and contains a primitive idempo
tent.
To describe another class of semigroups, we first consider the following
concepts from section 2.6 of [8].
Let I be an ideal of a semigroup S. Then the relation p defined on S
by (x, y) E p iff either x = y or x, y E I is easily seen to be a congruence on
S. The quotient semigroup S/ p is denoted by 8/ I and is called the Rees
factor semigroup of 8 modulo I . Now for a E 8, we denote by J (a) the
1.1. Semigroups 16
principal ideal SlaSI generated bya, and by Ja , the Jclass containing a.
Let I(a) = {x E J(a) : x ~ Ja }. If I(a) is nonempty, then it can be shown
to be an ideal in S and hence an ideal in J(a). We define P(a) by
P(a) = {J(a)/I(a), if I(a) ~s nonempty;
J (a), otherwise.
Then each of the semigroups P( a) is called a principal factor of S. It is
shown in Lemma 2.39 of [8] that if P is a principal factor of a semigroup
S, then P satisfies one and only one of the following conditions:
(1) p2 = OJ
(2) P is asimple;
(3) P is simple.
Definition 1.35. A semigrollp S is said to be semisimple if every principal
factor of S is osimple or simple. S is said to be completely semisimple if
every principal factor is completely simple or completely osimple.
A completely semisimple semigroup can be shown to be regular [8].
Further we have the following result. from [17].
Proposition 1.36. A. regular semigroup S is completely simple if and only
if no pair of distinct Drelated idempotents are comparable under w. •
Fina.lIy we have the following
Definition 1.37. A semigroup S is said to be bisimple if xVy for every
x,y E S.
1.2. Topological semigroups 17
1.2. Topological semigroups
A few results on topological semigroups, which will be used in the
sequel are presented here.
Definition 2.1. A semigroup S with a Hausdorff topology is said to be a
topological semigroup if the binary operation (x, y) xy from S x S to S
H
is continuous with respect to this topology on S and the product topology
on S x S. A group G with a Hausdorff topology is said to be a topological
group if it is a topological semigroup and the map x H x 1 from G to G is
continuous.
If T is a subsemigroup of a topological semigroup S, then it is easily
seen that T is a topological semigroup with respect to the relative topology
induced by the topology on S. However, if p is a congruence on a topological
semigroup S, then Sip may not be a topological semigroup with respect to
the quotient topology [6].
Definition 2.2. A congruence p on a topological semigroup S is said to
be a topological congruence if Sip is a topological semigroup with respect
to the quotient topology. p is said to be a topologica.l group congruence if
Sip with the quotient topology is a topological group.
In the following, if S is a topological semigroup and p is a congruence
on S, then Sip is always assumed to have the quotient topology.
If p is a topological congruence on a topological semigroup S, then the
canonical projection 71' : S ~ Sip is a continuous homomorphism; in fact it
is a quotient map. (Recall that a surjection J from atopological space X
to a topological space Y is said to be a quotient map if the topology on Y
1.2. Topological eemigroupe 18
is the largest topology for which f is continuous [44]). Before stating the
analogue of the second part of 1.9, we make the following
Definition 2.3. Two topological semigroups S, T are said to be topologi
cally isomorphic if there exists a semigroup isomorphism of S onto T, which
is also a homeomorphism..
The following result is proved in [6]' 1.49.
Theorem 2.4. Let 8, T be topological semigroups and let </> : 8 t T be a
surjective homomorphism, which is also a quotient map. Then p = ker </> is
a topological congruence on Sand 81 p is topologically isomorphic with T
under the map 'Ij; defined on 81p by 'Ij; (p(x)) = </>(x). _
We can also extend 1.10 to topological semigroups([G], 1.50).
Proposition 2.5. Let p and a be topological congruences on a topolog
ical semigroup 5 with p ~ a. Then p(x) H a(x) is a continuous 110
momorphis111 of 51 ponto 51a. Also pia is a topological congruence on
Sip and (Slp)/(pla) is topologically isomorphic witil 81a under tile ma.p
(alp) (p(x)) H a(x). _
The following situation is sometimes encountered in the sequel. Let p
and a be topological group congruences on a topological srmigroup 5 with
p ~ a. If e E E(5), then p(e) is the identity of the group 51 p. Also
(alp) (p(e)) = {p(x) : a(x) = a(e)} = {p(x): x E a(e)} = 1l' (aCe))
where IT : 81 p is the canonical projection. Hence by 1.7, 1l' (a(e)) is a
8 t
normal subgroup of 51 p and (51 p)/(al p) = (51 p)/7f (a(e)). So we have the
following
I.S. Topological vector IIpacell 19
Proposition 2.6. Let S be a topological semigroup and p, a topological
group congruences on S with p ~ a. If 9 = Sip and N = 11' (a(e)), where
e E E(S) and 11' : S ~ Sip is the canonical projection, then N is a normal
subgroup of 9 and S I a is topologically isomorplIic with giN under tIle
map a(x) 1+ p(x)N. •
1.3. Topological vector spaces
We consider only vector spaces over the field R of real numbers or the field
C of complex numbers. The symbol K is used to denote either R or C. Any
statement about vector spaces in which the field of scalars is not explicitly
mentioned is to be understood to apply to both the cases K = Rand K =
C.
Definition 3.1. A vector space X (over I{) endowed with a Hausdorff
topology (1\ being equipped with usual topology) is said to be a topological
vector space if the maps (x,y) H x+y from X x X to X and (A,x) 1+ AX
from KxX to X are continuous with respect to this topology on X and the
product topology on the respective domains.
Since many authors do not require the Hausdorff condition above, we
reiterate that all topological vector spaces considered by us are assumed to
be Hausdorff spaces.
W~ first note that subspaces and quotients by closed subspaces of a
topological vector space are well behaved. So is the product of a family of
topological vector spaces ([3], 11.6, H.8, 11.9).
Proposition 3.2. If Y is a subspace of a topological vector space X, then
Y with the relative topology is a topological vector space. If Y is a closed
1.3. Topological vector Ilpaces 20
subspace, tllen XI Y with the quotient topology is a topological vector space
and the canonical projection 11" : X ~ XIY defined by 1I"(x) == X + Y is a
continuous, open linear transformation. Also the product of any family of
topological vector spaces is a topological vector space with respect to the
product topology. •
In the sequel all subspaces of a. topological vector space a.re a.B8umed
to have the relative topology, all quotients the quotient topology and all
products the product topology.
Defore considering continuous linear transformations between topolog.
ical vector spaces, we introduce the following notations.
Definition 3.3. If X and Yare vector spaces and t : X ~
Y a linear
transformation, tlIen the range oft, denoted by R(t), is defined by
R (t) = { tx : x EX}
and the null space of t, denoted by N(t) is defined by
N (t) = { x EX: tx = 0 }.
It is easily seen that R(t) is a subspace of Y and N(t) is a subspace
of X. Also if X and Yare topological vector spaces and t is continuous,
then N(t) = t1(O) is closed, since Y is Hausdorff.. The following result is
proved in [25], 5.8.
Proposition 3.4. Let X, Y be topological vector spaces and t : X Y ~
a continuous linear transformation. Then <p : XI N(t) ~ R(t) defined by
<p (x + N(t)) == t(x) is a continuous, bijective linear transformation. _
I.S. Topological vector spaces 21
Note that ¢ defined above need not be a homeomorphism. We now
consider those linear maps for which the induced maps are also open.
Definition 3.5. Let X and Y be topological vector spaces. A continuous
linear map t : X  ? Y is said to be a topological homomorphism if for eadl
open set G of X, the image t(G) is an open subset of R(t). An injective
topologicallIomoIDorphisID is said to be a topological isomorphism. If there
exists a topological isomorphism from X onto Y, then X and Yare said
to be topologically isomorphic and we write X ~ Y.
It is not difficult to see that the map ¢ in 3.4 is a topological isomor
phism precisely when t is a topological homomorphism ([39], III. 1.2).
Proposition 3.6. Let X, Y be topological vector spaces. A linear map
t :X ? Y is a topological homomorphism if and only if the induced map
x + N(t) H tx of XjN(t) onto R(t) is a topological isomorphism. _
We will be mostly interested in continuous linear transformations from
a topological vector space to itself.
Definition 3.7. Let X be a topological vector space. A continuous linear
transformation from X to itself is called an operator on X.
Idempotent operators playa major role in our work. We first consider
the algebraic case([38], Sec 44).
Proposition 3.8. If X is a vector space and e : X ? X an idempotent
linear map, then so is 1 e. Also N(1 e) = R(e) and R(l e) =N(e)._
There is a one to one correspondence between idempotent linear maps
from X to itself and certain pairs of subspaces of X.
1.3. Topological vector Ilpacell 22
Definition 3.9. Let X be a vector space and A, B, subspaces of X. If
A + B = X and An B = {O}, then A and B are said to be algebraic
complements of each other in X and we write X = A+B.
The following result is proved in [38], Sec 44, Theorem B.
Prop08ition 3.10. Let X be a vector space. If e : X ~ X is an idempotent
linear map, then X = N(e)+R(e). Conversely, if A and Bare subspaces
of X with X = A+B, then there exists a unique idempotent linear map
e : X ~ X with R(e) = A and N(e) = B. •
Now let X be a topological vector space and X = A+B. Then the
idempotent e with R(e) = A and N(e) = B is not necessarily continuous.
For this we will have to strengthen the requirements on A and B. Note
that the map (x, y) H X +Y from A x B to X is a continuous bijection, but
not necessarily open.
Definition 3.11. Let X be a topological vector space and A, B, subspaces
of X. If X = A+B and if the map (x, y) H X + Y from A x B to X is a
homeomorphism then A and B are said to topological complements of each
=
other and we write X A $ B.
Now we can state the topological analogue of 3.10 ([3], 20.3, 20.7, 20.9).
Prop08ition 3.12. Let X be a topological vector space and A, B, sub
spaces of X with X = A+B. Let e be the idempotent linear map with
R(e) = A and N(e) = B. Then the following are equivalent:
(1) X = AEB B;.
(2) e is continuous;
1.3. Topological vector Ilpacell 23
(3) 1 e is continuous;
(4) the map ~ : A + X/B defined by ~(x) = x +B is a topological
isomorphism onto X/ B;
(5) the map tf; : B + X/A defined by tf;(x) = x +A is a topological
isomorphism onto X/A.
In particular, if X = A EB B, then A and B are closed.
Using 3.6 and the equivalence of (2) and (4) in 3.12 we can prove the

following
Proposition 3.13. An idempotent operator on a topological vector space
is a topological homomorphism. _
It should be noted that though every subspace of a (topological) vector
space has an algebraic complement, a closed subspace of even a Banach
space may fail to have a topological complement (see [26], 31.2(5) for an
example}. However, certain types of subspaces in any topological vector
space do possess such complements. Recall that by the codimension of a
subspace Y of a vector space we mean the dimension of X/ Y. The following
result is from [3], 23.7 .
Proposition 3.14. Let X be a topological vector space and let A be a
closed subspace of finite codimension in X. If B is a subspace of X with
X = A+B, then X = AEB B. _
Next we note some results on spaces of finite dimension. The following
is proved in [3], 23.6.
1.3. Topological vector spaces 24
Proposition 3.15. Let X be a topological vector space and let A be a
subspace of finite dimension. Then A + Y is closed for any closed subspace
Y of X. In particular, A is closed. _
From 23.9 of [3] we have the following result on a linear transformation
whose domain is of finite dimension.
Proposition 3.16. Let X and Y be topological vector spaces, where X is
of finite dimension. Then any linear map t : X + Y is continuous. _
Note that the finite dimensionality of the range does not imply conti
nuity. We consider such maps in the next result ([39]' III.1.3)
Proposition 3.17. Let X, Y be topological vector spaces and let t be a
linear map from X to Y with R(t) of finite dimension. Then the following
are equivalent:
(1) t is continuous;
(2) N(t) is closed!
(3) t is a topological homomorphism.
Most of our work will be about operators on some special types of

topological vector spaces. Recall that a subset A of a vector space X is said
to be convex, if x,y E A implies AX + (1  A)Y E A, for all scalars A with
o ~ ). ~ 1. Also by a neighborhood of a point x in a topological space, we
mean a set containing an open set containing x.
Definition 3.18. A topological vector space X is said to be a locally
convex space if for every X E X andevery neighborhood U of x, there
exists a convex neighborhood V of x with V ~ U.
1.3. Topological vector lIpaces 25
We note first that this property is preserved by subspaces, quotients
and products ([3], 33.4, 33.5).
Proposition 3.19. Every subspace and every quotient by a closed sub
space of a locally convex space is locally convex. Also the product of any
,family of locally convex spaces is locally convex. _
In the study of (topological) vector spaces, (continuous) linear maps of
the space to the underlying field play an important role.
Definition 3.20. Let X be a vector space. A linear map 4> : X +K 18
called a linear functional on X. The set of all linear functionals on X is
called the algebraic dual of X and is denoted by X'. If X is a topological
vector space, then the set of all continuous linear functionals on X is called
the topological dual of X and is denoted by X·.
One important feature of a locally convex space is that it has suf
ficiently many continuous linear funetionals to distinguish points. More
precisely we have the following result from Theorem 3.5 of [37]. Note that
if Y is a subset of a topological space, then the closure of Y is denoted by
Y.
Theorem 3.21. Let Y be a subspace of a locally convex space X and
Xo EX. If Xo f/. Y, then there exists ¢ E X· with ¢(xo) = 1 and ¢(y) = 0
for all y E Y. •
We also note the following version of the HahnBanach theorem ([37],
3.6).
1.3. Topological vector IIpacell 26
Theorem 3.22. IfX is a locally convex space and Y a subspace of X, then
every continuous linear functional on Y may be extended to a continuous
linear functional on X. •
A consequence of the above theorem which we often use is given below
([3], 34.10).
Proposition 3.23. If X is a locally convex space then every finite dimen
sional subspace of X ha.s a topological complement in X. •
We define below two more types of topological vector spaces. Note that
a metric d on a vector space X is said to be invariant if d(x + z, y + z) =
d(x, y) for all x, y, z EX.
Definition 3.24. Let X be a topological vector space with topology T.
X is said to be metrizable if there exists a metric on X which induces the
topology T. X is said to be an Fspace if T is induced by a complete,
invariant metric. If X is a locally convex F space then X is called a Frechet
space.
It should be pointed out that there is considerable variation in litera
ture regarding the names Fspace and Frechet space. In the above definition
we follow the terminology of [37].
Another class of locally convex spaces can be described as follows. Let
X be a vector space and Y, a subspace of X' which separates points of X in
the sense that for each x E X with x i 0 there eXIsts 4; E Y with 4;( x) i O.
Then among all the topologies T on X for which (X, T) is a locally. convex
space with (X, T)· = Y, there exists a strongest one ([28], Theorem 5).
1.3. Topological vector spaces 27
This topology is denoted by r(X, Y). Now if X is a locally convex space
then X* separates points of X and so r(X,X*) exists.
Definition 3.25. A locally convex space X with topology T is said to be
a Mackey space if T = r( X, X*).
Many of the locally convex spaces used in applications are Mackey
spaces. For example, every metrizable space is a Mackey space ([391, IV.3A).
If X is a locally convex space then r(X, X·) is the strongest locally
convex topology on X with respect to which X has dual X'. Now we
describe the smallest such topology. First note that X is a set and 1 a
nonempty family of maps f :X + Yf where each Yf is a topological space,
then among all topologies on X for which each f 1 is continuous, there
E
is a smallest ([44]' 8.9). This is called the weak topology on X induced by
1.
Definition 3.26. If X is a locally convex space then the weak topology
on X induced by X· is called the weak topology on X.
The following is a special case of Theorem 3.10 (see also 3.11) of [37].
Proposition 3.27. IE X is a locally convex space, then X with the weak
topology is a locally convex space whose topological dual is also X·. •
From 17.13 of [25], we have the following facts on the weak topology
on subspaces, quotients and products.
Proposition 3.28. Let X be a locally convex space and Y a subspace of
X:
1.3. Topological vector spaces 28
(1) the weak topology on Y is the relativization of the weak topology on
Xj
(2) ifY is closed, then the weak topology on X/Y is the quotient topology
derived from the weak topology on X.
Also the weak topology on the product of a family of locally convex spaces
is the product of the weak topologies on the factors. _
To distinguish topological notions pertaining to the weak topology from
those relating to the given topology, we use the adjective 'weakly'(or 'weak')
when referring to the former. Thus for example, a linear map between
locally convex spaces X and Y is said to be weakly continuous if it is
continuous with respect to the weak topologies on X and Y.
The following result is a special case of [251, 17.1.
Proposition 3.29. Let X be a locally convex space and Y a subspace of
X. Then Y is weakly closed if and only if it is closed. _
The next result relates continuity and weak continuity of linear maps
([39], IV.7.4, Corollary to 7.5 and 7.6 ).
Proposition 3.30. Let X' and Y be locally convex spaces and t : X Y
+
a linear map. If t is continuous, then it is weakly continuous and if t is a
topological homomorphism, then it is a weak topological homomorphism.
If X is a Mackey space and t weakly continuous, then it is continuous and if
X and R(t) are Mackey spaces and t is a weak topological homomorphism,
then it is a topological homomorphism. _
An important topology on the dual X· of a topological vector space
can be defined as follows.
1.3. Topological vector spaces 29
Definition 3.31. Let X be a topological vector space and for each x in
X, let ~% be defined on X* by ~%(t/J) = t/J(x) for all t/J E X*. Then the weak
topology on X· induced by the family {~% : x EX} of maps is called the
weak* topology on X·.
One interesting feature of this topology is that it is always locally
.convex ([37], 3.10, 3.14).
Proposition 3.32. If X is a topological vector space, then X* with the
weak* topology is a locally convex space. _
We next see that if X and Yare locally convex spaces then to each
weakly continuous linear map from X to Y there exists a weak* continuous
linear map from y* to X* ([28], 21.5).
Proposition 3.33. Let X, Y be locally convex spaces and let t be a weakly
continuous linear map from X to Y. Then t* defined on Y* by t* (~) = #
for all ~ E y* is a weak* continuous linear map from Y· to X·. . _
The map t· defined in the above result is called (topological) adjoint
of t. The next result gives further relations between t and t* ([25], 21.4,
21.6 ).
Proposition 3.34. Let X, Y be locally convex spaces and let t : X  Y
be a weakly continuous linear map. Then t is a weak topological homomor
phism if and only if R(t*) is weak* closed and t* is a weak* topological
homomorphism if and only if R(t) is closed. _
I.S. Topological vector spaces 30
Finally, we mention some results on the set of all (closed) subspaces of
a (topological) vector space. It is eaBily seen that the set of all subspaces
of a vector space is a poset under set inclusion and in fact forms a lattice
with A + B aB the join and An B as the meet of two subspaces A and B
([261, 7.10(1)). We denote this lattice by l(X). An important property of
this lattice is the following
Proposition 3.35. l(X) is a modular lattice for any vector space X; that
IS,
A~C => A + (B n C) = (A + B) n C
for subspaces A, Band C of X. •
If X is a topological vector space, then the set of all closed subspaces
of X does not form a sublattice of .c(X), since A + B may not be closed for
closed subspaces A and B of X. However if we define A vB = A + B, ( the
closure of A and B) and A 1\ B = An B, then this set becomes a lattice
with respect to these operations. We denote this lattice by I(X). Also
the lattice of all weak* closed subspaces of X· with similar operations is
denoted by I(X·). It can be shown that if X is a locally convex space, then
I(X) and I(X') are dually isomorphic. To make this explicit we make use
of the following
Definition 3.36. Let X be a topological vector space. If Y ~ X, then
the annihilator of Y, denoted by y.l, is defined by
y.l = { 4J E X' : 4J(y) = 0 for all y E Y }.
If y ~ X', then the preannihilator of Y, denoted by .1 Y, is defined by
.lY={xEX:4J(x)=O [orall 4JEY}.
I.:L Topological vector spaces 31
The result below is obtained by using 3.29 above with 20.3(3) of [26]
(see also [3],'38.14). Note that a map ~ from a lattice Ll to a lattice L2 is
said to be a dual isomorphism if ~ is a bijection and x ~ y in L 1 if and
only if ~(x) ~ ~(y) in L2.
Proposition 3.37. If X is a locally convex space, then map A H AJ.. is a
dual isomorphism of the lattice I(X) onto the lattice I(X*). •
Though the lattice I(X) is not modular, certain pairs of subspaces do
satisfy a local version of modularity.
Definition 3.38. An ordered pair (a, b) of elements of a lattice L is said
to be a modular pair, written m(a,b), if for x E L
x ~ b =;. x V (a t\ b) = (x va) t\ b
and (a, b) is said to be a dual modular pair, written m* (a, b), if for x E L
x ~ b =;. x t\ (a Vb) = (x t\ a) vb.
Note that in general m(a, b) does not imply m(b, a); nor does m*(a, b) imply
m*(b,.a). In other words m and m* do not define symmetric relations on L.
Modular and dual modular pairs in the lattice of weakly closed sub
space of a topological vector space are given in Theorem III.6 and Theorem
llI.7 of [27]. The condition for dual modularity is easily seen to hold in
'c(X). Again, if X is a locally convex space, then using 3.29 above, the
condition for modularity can also be carried over to 'c(X).
1.3. Topological vector spaces 32
Proposition 3.39. Let X be a topological vector space and A, B closed
subspaces of X. Then m·(A, B) in I(X) if and only if A + B is closed. In
the case when X is a locally convex space, m(A, B) in I(X) if and only if
the"map ~ : Ax B  A+B defined by ~(x, y) = x+y is a weak topological
homomorphism. _
From this we can easily deduce the first part of the following result;
the second part follows from 3.30.
Corollary 3.40. Let A, B be closed subspaces of a locally convex space
X. Then A and B are weak topological complements if and only if A and
B are algebraic complements with m(A, B) in I(X). In particular, if X is
a Mackey space, then A and B are topological complements if and only if
A and B are algebraic complements with m(A,B) in I(X). •