SEMIGROUP THEORY

VICKY G
1. The Basic Concept
Denition: A semigroup is a pair (S, ) where S is a non-empty set and is an associative
binary operation on S. [i.e. is a function SS S with (a, b) ab and for all a, b, c S
we have a (b c) = (a b) c].
We abbreviate (S, ) by S and often omit in a b and write ab. By induction
a
1
a
2
. . . a
n
is unambiguous. Thus we write a
n
for
aa . . . a
. .
n times
.
Index Laws: We have that for all n, m N = 1, 2, . . . the following index laws hold
a
n
a
m
= a
n+m
_
a
n
_
m
= a
nm
.
Denition: A monoid M is a semigroup with an identity, i.e. there exists 1 M such
that 1a = a = a1 for all a M.
Putting a
0
= 1 then the index laws hold for all n, m N
0
.
Note. The identity of a monoid is unique.
Denition: A group G is a monoid such that for all a G there exists a b G with
ab = 1 = ba.
Example 1.1. Groups are monoids and monoids are semigroups. Thus we have
Groups Monoids Semigroups.
The one element trivial group e with multiplication table
e
e e
is also called the trivial semigroup or trivial monoid.
Example 1.2. A ring is a semigroup under . A ring with identity is a monoid.
1
2 VICKY G
Example 1.3. N is a monoid under . N forms a semigroup under + and N
0
is a monoid
under + and .
Example 1.4. Let I, J be non-empty sets and set T = I J with the binary operation
(i, j)(k, ) = (i, ).
Then T is a semigroup called the rectangular band on I J. To verify that T is a semigroup
we need to check that the associativity law holds. So
_
(i, j)(k, )
_
(m, n) = (i, )(m, n) = (i, n),
(i, j)
_
(k, )(m, n)
_
= (i, j)(k, n) = (i, n),
for all (i, j), (k, ), (m, n) T and hence multiplication is associative.
I
J
i
j
k

(i, j)
(k, )
(i, )
Figure 1. The rectangular band.
Notice: (i, j)
2
= (i, j)(i, j) = (i, j), i.e. every element is an idempotent. This cant be
looked at from ring theory because any ring where every element is an idempotent means
that the ring is commutative. However, the rectangular band does not have to be commu-
tative.
1.1. Adjoining an Element
Let S be a semigroup, which is not a monoid. Find a symbol not in S, call it 1. We now
extend the denition of on S to S 1 by
SEMIGROUP THEORY 3
a b = ab if a, b S,
a 1 = a = 1 a for all a S,
1 1 = 1.
Then is associative (check this). Thus we have managed to extend multiplication in S
to S 1. For an arbitrary semigroup S the monoid S
1
is dened by
S
1
=
_
S if S is a monoid,
S 1 if S is not a monoid.
So, S
1
is S with a 1 adjoined if necessary.
Example 1.5. Let T be the rectangular band on ab, c. Then T
1
= 1, (a, b), (a, c),
which has multiplication table
1 (a, b) (a, c)
1 1 (a, b) (a, c)
(a, b) (a, b) (a, b) (a, c)
(a, c) (a, c) (a, b) (a, c)
Example 1.6 (The Bicyclic Semigroup / Monoid). If A Z, such that [A[ < , then
max A is the biggest element in A. We also use this notation to represent
maxa, b =
_
a if a b,
b if b a.
We note some further things about max:
maxa, 0 = a if a N
0
,
maxa, b = maxb, a,
maxa, a = a,
max
_
a, maxb, c
_
= maxa, b, c = max
_
maxa, b, c
_
.
Thus we have that (Z, max) is a semigroup and (N
0
, max) is a monoid.
Note. The following identities hold for all a, b, c Z
()
_
a + maxb, c = maxa + b, a + c,
maxb, c = a + maxb a, c a.
Put B = N
0
N
0
. Now, on B we dene a binary operation by
(a, b)(c, d) = (a b + t, d c + t),
where t = maxb, c. We claim that B together with this operation forms a semi-
group/monoid called the Bicyclic Semigroup/Monoid.
4 VICKY G
Proof. With (a, b), (c, d) B and t = maxb, c we have t b 0 and t c 0. Thus we
have a b + t a and d c + t d. Therefore, in particular (a b + t, d c + t) B so
multiplication is closed. We have that (0, 0) B and for any (a, b) B we have
(0, 0)(a, b) = (0 0 + max0, a, b a + max0, a),
= (0 0 + a, b a + a),
= (a, b),
= (a, b)(0, 0).
Therefore (0, 0) is the identity of B. We need to check associativity of max. Let (a, b),
(c, d), (e, f) B. Then
_
(a, b)(c, d)
_
(e, f) =
_
a b + maxb, c, d c + maxb, c
_
(e, f),
=
_
a b d + c + maxd c + maxb, c, e
f e + maxd c + maxb, c, e
_
.
(a, b)
_
(c, d)(e, f)
_
= (a, b)
_
c d + maxd, e, f e + maxd, e
_
,
= (a b + maxb, c d + maxd, e
f e c + d + maxb, c d + maxd, e
_
.
Now we have to show that

a b d + c + max
_
d c + maxb, c, e
_
=

a b + max
_
b, c d + maxd, e
_
,

f e + max
_
d c + maxb, c, e
_
=

f e c + d + max
_
b, c d + maxd, e
_
.
We can see that these equations are the same and so we only need to show
c d + max
_
d c + maxb, c, e
_
= max
_
b, c d + maxd, e
_
.
Now, we have from () that
max
_
maxb, c, c d + e
_
= max
_
b, c d + maxd, e
_
.
The RHS of this equation is
max
_
b, c d + maxd, e
_
= max
_
b, maxc d + d, c d + e
_
,
= max
_
b, maxc, c d + e
_
,
= maxb, c, c d + e,
= max
_
maxb, c, c d + e
_
.
Therefore multiplication is associative and hence B is a semigroup/monoid.
SEMIGROUP THEORY 5
Example 1.7. T
X
is a semigroup. See the examples class for proof.
1.2. Easy Facts / Denitions
A semigroup S is commutative if ab = ba for all a, b S. For example N with + is
commutative. B is not because
(0, 1)(1, 0) = (0 1 + 1, 0 1 + 1) = (0, 0),
(1, 0)(0, 1) = (1 0 + 0, 1 0 + 0) = (1, 1).
Thus we have (0, 1)(1, 0) ,= (1, 0)(0, 1). Notice that in B; (a, b)(b, c) = (a, c). It is important
to see that in general we have NO cancellation in Semigroups. Thus we have that
ac = bc , a = b,
ca = cb , a = b.
For example in the rectangular band on 1, 2 1, 2 we have
(1, 1)(1, 2) = (1, 2) = (1, 2)(1, 2)
but (1, 1) ,= (1, 2). A semigroup is cancellative if
ac = bc a = b,
ca = cb a = b,
are true for all a, b, c S. For example, a group is cancellative (indeed, any subsemigroup
of a group is cancellative). N
0
is a cancellative monoid, which is not a group.
1.3. Zeros
A zero 0 of a semigroup S is an element such that, for all a S,
0a = a = a0.
Adjoining a Zero
Let S be a semigroup, then pick a new symbol 0. Let S
0
= S 0; dene a binary
operation on S
0
by
a b = ab for all a S,
0 a = 0 = a 0 for all a S,
0 0 = 0.
6 VICKY G
Then is associative, so S
0
is a semigroup with zero. We say that S is a semigroup with
a zero adjoined.
2. Reminder of Familiar Ideas
Example 2.1 (Function). Let A = 1, 2, 3 and B = a, b, c then we dene a function
f : A B by f(1) = a, f(2) = a, f(3) = c. We can also describe this using two row
notation
f =
_
1 2 3
a b c
_
.
We write xf rather than f(x). Here 1f = a = 2f, 3f = c.
Binary Operations
Let A be a set. A binary operation on A is a function from AA A. We write a b
for (a, b).
Composition of Functions
Let f : A B and g : B C then we dene the composition of these functions,
f g : A C, to be
a(f g) = (af)g.
Now given another function h : C D then the composition of these functions is associa-
tive. For any a A we have
a
_
(f g) h
_
=
_
a(f g)
_
h,
=
_
(af)g
_
h,
= (af)(g h),
= a
_
f (g h)
_
.
Identity Function
For any set X we dene the identity function I
X
: X X such that for all x X
xI
X
= x.
Also, if f : X Y a function then
I
X
f = f = f I
Y
.
Let X be a set such that X ,= . Recall that o
X
= : X X [ is a bijection
is a group under , called the symmetric group on X. Usually write o
n
for o
n
where
n = 1, 2, . . . , n.
SEMIGROUP THEORY 7
Theorem 2.1 (Cayley). If G is a group then there exists a 1:1 homomorphism : G o
G
.
Denition: T
X
= [ is a function X X.
We write T
n
for T
n
. This has size n
n
because thats how many ways there are of choosing
a function on n elements. T
X
is a monoid with identity I
X
called the full transformation
monoid / semigroup on X.
Claim. Clearly o
X
T
X
and if [X[ 2 then o
X
,= T
X
.
Proof. For any n X let c
n
T
X
be the constant map on n, i.e. yc
x
= x for all y X. Note
that Imc
n
= n and if [X[ 2 then c
n
is neither one-to-one nor onto, so c
x
o
X
T
X
.
Now notice c
x
= c
x
for all T
X
.
We show this by letting y X, then
y(c
x
) = (y)c
x
= x = yc
x
.
Hence c
x
= c
x
. Thus c
x
is a right zero for T
X
. Furthermore c
x
= c
x
for all , T
X
so T
X
is not cancellative for [X[ 2.
If o
n
then can be written in two row notation or as a product of disjoint cycles.
For example
=
_
1 2 3 4 5
2 3 1 5 4
_
o
5
= (123)(45).
We can also express this pictorially as the cycle diagram of . For T
X
we can use two
row notation. For example if T
5
is given by 1 = 2, 2 = 2, 3 = 3, 4 = 1, 5 = 1.
Then,
=
_
1 2 3 4 5
2 2 3 1 1
_
.
For example,
c
2
=
_
1 2 3 4 5
2 2 2 2 2
_
T
5
.
We have map diagrams for T .
3. Subsemigroups / Submonoids
Denition: Let S be a semigroup and , = T S. Then T is a subsemigroup of S if
a, b T ab T. If S is a monoid then T is a submonoid of S if T is a subsemigroup
and 1 T.
Example 3.1. (N, +) is a subsemigroup of (Z, +).
8 VICKY G
Example 3.2. R = c
x
[ x X is a subsemigroup of T
X
called the right zero semigroup.
Proof. c
x
c
y
= c
y
for all x, y X.
Example 3.3 (Bicyclic Monoid). Put E(B) = (a, a) [ a N
0
, the set of all idempotents
in B. We claim that E(B) is a commutative submonoid of B. Clearly we have (0, 0) E(B)
and for (a, a), (b, b) E(B) we have
(a, a)(b, b) = (a a + t, b b + t) where t = maxa, b,
= (t, t),
= (b, b)(a, a).
3.1. Morphisms
Denition: Let S, T be semigroups then : S T is a semigroup (homo)morphism if,
for all a, b S,
(ab) = ab.
If S, T are monoids then is a monoid (homo)morphism if is a semigroup morphism and
1
S
= 1
T
.
Example 3.4. : B Z given by (a, b) = a b is a monoid morphism because
_
(a, b)(c, d)
_
= (a b + t, d c + t) t = maxb, c,
= (a b) (d c),
= (a b) + (c d),
= (a, b) + (c, d).
Furthermore (0, 0) = 0 0 = 0.
Example 3.5. Let T = I J be the rectangular band then dene : T T
J
by
(i, j) = c
j
. Then we have
_
(i, j)(k, )
_
= (i, ),
= c

,
= c
j
c

,
= (i, j)(k, ).
So, is a morphism.
Denition: A bijective morphism is an isomorphism.
Isomorphisms preserve algebraic properties (e.g. commutativity). Suppose f : S T is a
morphism then with A S we have
SEMIGROUP THEORY 9
Af = af [ a A.
If A = S then Sf is the image, Imf, of S. So, if A is a subsemigroup of S then Af will
be a subsemigroup of T. Now, conversely, if B T then we dene
Bf
1
= s S [ sf B.
If B is a subsemigroup of T then Bf
1
is either or Bf
1
is a subsemigroup of S, i.e.
endomorphisms pull back subsemigroups.
Embeddings
Let : S T be a morphism. From the handout, Im is a subsemigroup of T. If is
1:1 then : S Im is an isomorphism and S is embedded in T.
Theorem 3.1 (The Cayley Theorem for Semigroups). Every semigroup is embedded
in some T
X
Proof. Let S be a semigroup and set X = S
1
. We need a 1:1 morphism S T
X
. For
s S, we dene
s
T
X
by x
s
= xs. Now, dene : S T
X
by s =
s
. First show
that is 1:1. If s = t then
s
=
t
and so x
s
= x
t
for all x S
1
; in particular
1
s
= 1
t
and so 1s = 1t hence s = t and is 1:1.
Let u, b S. For any x X we have
x(
u

v
) = (x
u
)
v
) = (xu)v = x(uv) = x
uv
.
Hence
u

v
=
uv
and so uv =
u

v
=
uv
= (uv). Therefore is a morphism. Hence
: S T
X
is an embedding.
Theorem 3.2 (The Cayley Theorem - for Monoids). Let S be a monoid then there exists
an embedding S T
X
for some X.
Proof. We know from the proof of the Cayley Theorem for Semigroups that : S T
X
,
given by s =
s
where x
s
= xs, is a semigroup embedding. In the proof above we had
X = S
1
X = S because S is a monoid. We must check that is a monoid morphism,
i.e. we need to check 1 = I
X
. Now 1 =
1
and for all x X = S we have
x
1
= x1 = x = xI
X
and so 1 =
1
= I
X
.
Theorem 3.3 (The Cayley Theorem - for Groups). Let S be a group. Then there exists
an embedding S o
X
for some X.
Proof. Exercise.
10 VICKY G
3.2. Idempotents
Denition: e S is an idempotent if e
2
= e. Also we dene the set of idempotents in S
to be
E(S) = e S [ e
2
= e.
Now, E(S) may be empty, e.g. E(S) = (N under +) but E(S) may also be S. If S = I J
is a rectangular band then for any (i, j) S we have (i, j)
2
= (i, j)(i, j) = (i, j) and so
E(S) = S.
Denition: If E(S) = S, then S is a band.
For the bicyclic semigroup B we have E(B) =
_
(a, a) [ a N
0
_
from exercises 1. If S is
a monoid then 1 E(S). If S is a cancellative monoid, then 1 is the only idempotent: for
if e
2
= e then ee = e1 and so e = 1 by cancellation. In particular for S a group we have
E(S) = 1.
Lemma 3.1. Suppose ef = fe for all f, e E(S). Then E(S) = or E(S) is a subsemi-
group.
Proof. Let e, f E(S). Then (ef)
2
= efef = eeff = ef and hence ef E(S).
Denition: A commutative band is a semilattice.
From Lemma 3.1 if idempotents in S commute then E(S) is empty or its a semilattice.
Example 3.6. E(B) =
_
(a, a) [ a N
0
_
is a semilattice.
Example 3.7. A rectangular band I J is not a semilattice (unless [I[ = [J[ = 1) since
(i, j)(k, ) = (k, )(i, j) i = k and j = .
Denition: Let a S. Then we dene a) = a
n
[ n N, which is a commutative
subsemigroup of S. We call a) the monogenic subsemigroup of S generated by a.
Proposition. a)

= (N, +) or a) is nite.
Proof. If a
i
,= a
j
for all i, j N with i ,= j then : a) N dened by a
i
= i is an
isomorphism. Suppose that in the list of elements a, a
2
, a
3
, . . . there is a repetition, i.e.
a
i
= a
j
for some i < j. Let k be least such that a
k
= a
n
for some n < k. Then k = n + r
for some n, r N where n is the index of a, r is the period of a. Then the elements
a, a
2
, a
3
, . . . , a
n+r1
are all distinct and a
n
= a
n+r
.
We can express this pictorially by a map diagram.
SEMIGROUP THEORY 11
a
a
2
a
3
a
4
a
n
a
n+1
a
n+r1
a
n+2
Notice that
a
n+2r
= a
n+r+r
= a
n+r
a
r
= a
n
a
r
= a
n+r
= a
n
and hence a
n+rk
= a
n
for all k N
0
. Let u N
0
. Write u = qr + t with 0 t < r,
q, t N
0
. Then a
n+u
= a
n+qr+t
= a
n+qr
a
t
= a
n
a
t
= a
n+t
and so we have
a) = a, a
2
, . . . , a
n+r1
and [a)[ = n + r 1.
Lemma 3.2 (The Idempotent Power Lemma). If a) is nite, then it contains an idem-
potent.
Proof. Let n, r be the index and period of a. Choose s N
0
with s n (mod r). Then
s + n 0 (mod r) and so s + n = kr for k N. Then
(a
n+s
)
2
= a
n+n+s+s
= a
n+kr+s
= a
n+kr
a
s
= a
n
a
s
= a
n+s
and so a
n+s
E(S). In fact, a
n
, a
n+1
, . . . , a
n+r1
is a group with identity a
n+s
.
Corollary 3.1. Any nite semigroup contains an idempotent.
3.3. Idempotents in T
X
We know c
x
c
y
= c
y
for all x, y X and hence c
x
c
y
= c
x
for all x X. Therefore c
x
E(T
X
)
for all x X.
Example 3.8. Let us dene an element
=
_
1 2 3
2 2 3
_
E(T
X
)
and a subset Z X. The restriction of to Z , denoted [
Z
is the map
Z
: Z Y
such that Z(
Z
) = Z.
Example 3.9. Let us dene an element
=
_
a b c d
1 1 1 2
_
[
{c,d}
=
_
c d
1 2
_
We can see that is not one-to-one but [
{c,d}
is.
12 VICKY G
Let T
X
(i.e. : X X). Recall that Im = x : x X X.
Example 3.10. In T
3
we have Imc
1
= 1, ImI
3
= 1, 2, 3 then
Im
_
1 2 3
3 2 3
_
= 2, 3.
Lemma 3.3 (The E(T
X
) Lemma). An element T
X
is idempotent [
Im
= I
Im
.
Proof. [
Im
means for all y Im we have y = y. Then
E(T
X
)
2
= ,
x
2
= x for all x X,
(x) = x for all x X,
y = y for all y Im,
[
Im
= I
Im
.
Example 3.11. Dene an element
=
_
1 2 3
2 2 3
_
T
3
,
this has image, Im = 2, 3. Now we can see that 2 = 2 and 3 = 3. Hence E(T
3
).
Example 3.12. We can similarly create another idempotent in T
5
,
_
1 2 3 4 5
5 5 3 3 5
_
E(T
5
).
Using Lemma 3.3 we can now list all the idempotents in T
3
. We start with the constant
maps, i.e. E(T
3
) such that [ Im[ = 1. These are
_
1 2 3
1 1 1
_
,
_
1 2 3
2 2 2
_
,
_
1 2 3
3 3 3
_
.
Now consider all elements E(T
3
) such that [ Im[ = 2. These are
_
1 2 3
2 2 3
_
,
_
1 2 3
1 1 3
_
,
_
1 2 3
3 2 3
_
,
_
1 2 3
1 3 3
_
,
_
1 2 3
1 2 1
_
,
_
1 2 3
1 2 2
_
.
Now there is only one idempotent such that [ Im[ = 3, that is the identity map
_
1 2 3
1 2 3
_
.
SEMIGROUP THEORY 13
4. Relations
Denition: A (binary) relation on A is a subset of AA.
Convention: we may write ab for (a, b) .
4.1. Partial Orders
We dene a partial ordering on R.
a a for all a R,
a b and b c a c for all a, b, c R,
a b and b a a = b for all a, b R.
Notice that for any a, b R we have a b or b a. For X a set then T(X) is the set of
all subsets of X. Now is a partial order on T(X). We have
A A for all A T(X)
A B and B C A C for all A, B, C T(X)
A B and B A A = B for all A, B T(X)
Notice that if [X[ > 2 and x, y X with x ,= y then x , y and y , x. If
= A A is the universal relation on A, so xy for all x, y A, then [x] = A for all
x A. We have that
=
_
(a, a) [ a A
_
is the equality relation and so xy x = y and so [x] = x for all x A.
4.2. Algebra of Relations
If , are relations on A, then so is . For all a, b A we have
a( )b (a, b) ( )
(a, b) and (a, b)
ab and ab.
We note that means ab ab. Note is reexive and so for any
equivalence relation . We see that is the smallest equivalence relation on A and is the
largest equivalence relation on A. We note that
[a] = b A [ ab.
If is an equivalence relation then [a] is the equivalence-class, a -class, of a.
14 VICKY G
Lemma 4.1. If , are equivalence relations on A then so is .
Proof. We have and , then , so is reexive. Suppose (a, b) .
Then (a, b) and (a, b) . So as , are symmetric, we have (b, a) and (b, a)
and hence (b, a) . Therefore is symmetric. By a similar argument we have
is transitive. Therefore is an equivalence relation.
Denoting by [a]

the -class of a and [a]

the -class of a we have that,

[a]

= b A [ b a,
= b A [ ba and ba,
= b A [ ba b A [ ba,
= [a]

[a]

.
We note that need not be an equivalence relation. On Z we have
3 1 (mod 2),
1 4 (mod 3).
If
_
(mod 2)
_

_
(mod 3)
_
were to be transitive then we would have
(3, 1)
_
(mod2)
_

_
(mod3)
_
(1, 4)
_
(mod2)
_

_
(mod3)
_
_
(3, 4)
_
(mod 2)
_

_
(mod 3)
_
3 4 (mod 2) or 3 4 (mod 3)
but this is a contradiction!
Kernels
Let : X Y be a function. Dene a relation ker on X by the rule
a ker b a = b.
We may sometimes write a

b. It is clear that ker is an equivalence relation on X.

The ker classes partition X into disjoint subsets; a, b lie in the same class i a = b.
Example 4.1. Let : 6 4 where
=
_
1 2 3 4 5 6
3 2 3 2 2 1
_
ker is the kernel of .
SEMIGROUP THEORY 15
Denition: An equivalence relation on a semigroup S is a congruence if
(ab and cd) acbd.
Lemma 4.2 (The Kernel Lemma). Let : S T be a semigroup morphism. Then ker
is a congruence on S.
Proof. We know ker is an equivalence relation on S. Suppose a, b, c, d S with
a ker b and c ker d.
Then a = b and c = d, so
(ac) = ac = bd = (bd).
Therefore ac ker bd, so that ker is a congruence.
Let be a congruence on S. Then we dene
S/ =
_
[a] [ a S
_
.
Dene a binary relation on S/ by
[a][b] = [ab].
We need to make sure that this is a well-dened relation. If [a] = [a

] and [b] = [b

] then
aa

and bb

; as is a congruence we have aba

and hence [ab] = [a

]. Hence our
operation is well dened. Let [a], [b], [c] S/ then we have
[a]
_
[b][c]
_
= [a][bc],
=
_
a(bc)

,
=
_
(ab)c

,
= [ab][c],
=
_
[a][b]
_
[c].
If S is a monoid, then so is S/ because we have
[1][a] = [1a] = [a] = [a1] = [a][1]
for any a S. Hence we conclude that S/ is a semigroup and if S is a monoid, then so is
S/. We call S/ the factor semigroup (or monoid) of S by . Now, dene

: S S/
by
s

= [s].
Then we have
16 VICKY G
s

= [s][t] denition of

,
= [st] denition of multiplication in S/,
= (st)

denition of

.
Hence

is a semigroup morphism. We now want to examine the kernel of

and so
s ker

t s

= t

denition of ker

,
[s] = [t] denition of

,
st denition of .
Therefore = ker

and so every congruence is the kernel of a morphism.

Theorem 4.1 (The Fundamental Theorem of Morphisms for Semigroups). Let : S T
be a semigroup morphism. Then ker is a congruence on S, Im is a subsemigroup of T
and S/ ker

= Im.
Proof. Dene

: S/ ker Im by [a]

= a. We have
[a] = [b] a ker b
a = b
[a]

= [b]

.
Hence

is well dened and one-to-one. For any x Im we have x = a = [a]

and so

is onto. Finally,
_
[a][b]
_

= [ab]

= (ab) = ab = [a]

[b]

.
Therefore

is an isomorphism and S/ ker

= Im. Note that the analogous result holds
for monoids.
5. Ideals
Notation
If A, B S then we write
AB = ab [ a A, b B,
A
2
= AA = ab [ a, b A.
Note. A is a subsemigroup if and only if A ,= and A
2
A.
We write aB for aB = ab [ b B. For example
AaB = xay [ x A, y B.
SEMIGROUP THEORY 17
Facts:
(1) A(BC) = (AB)C therefore T(S) = S [ A S is a semigroup the power
semigroup of S.
(2) A B AC BC and CA CB for all A, B, C T(S).
(3) AC = BC , A = B and CA = CB , A = B, i.e. the power semigroup is not
cancellative.
Denition: Let , = I S then I is a right ideal if IS I (i.e. a I, s S as I).
Then I is a left ideal if SI I. Finally I is a (two sided) ideal if IS SI I.
Any (left/right) ideal is a subsemigroup. If S is commutative, all 3 concepts coincide.
Example 5.1. Some examples of ideals.
(1) Let i I then i J is a right ideal in a rectangular band I J.
(2) We dene a right ideal (x, y) [ x m, y N
0
in the bicyclic semigroup B (m
xed).
(3) Y X then we have T
X
[ Im Y is a left ideal of T
X
.
(4) For any n N we dene
S
n
= a
1
a
2
. . . a
n
[ a
i
S.
This is an ideal of S. If S is a monoid then S
n
= S for all n, since for any s S
we can write
s = s 11 . . . 1
. .
n1
S
n
.
(5) If S has a zero 0, then 0 (usually written 0), is an ideal.
Denition: For a semigroup S we have:
(1) S is simple if S is the only ideal.
(2) if S has a zero 0, then S is 0-simple if S and 0 are the only ideals and S
2
,= 0.
Example 5.2. Let G be a group and I a left ideal. Let g G, a I then we have
g = (ga
1
)a I
and so G = I. Therefore G has no proper left/right ideals. Hence G is simple.
Exercise: G
0
is 0-simple
Example 5.3. We have (N, +) is a semigroup. Now dene I
n
(N, +) to be
I
n
= n, n + 1, n + 2, . . . ,
which is an ideal. Hence N is not simple.
Note. 2, 4, 6, . . . is a subsemigroup but not an ideal.
Example 5.4. The bicyclic semigroup B is simple.
18 VICKY G
Proof. Let I B be an ideal, say (m, n) I. Then (0, n) = (0, m)(m, n) I. Thus
(0, 0) = (0, n)(n, 0) I. Let (a, b) B. Then
(a, b) = (a, b)(0, 0) I
and hence B = I B is simple.
5.1. Principle Ideals
We make note of how the S
1
notation can be used. For example
S
1
A = sa [ s S
1
, a A,
= sa [ s S 1, a A,
= sa [ s S, a A 1a [ a A,
= SA A.
In particular, if A = a then S
1
a = Sa a. So,
S
1
a = Sa a Sa,
a = ta
for some t S. We have S
1
a = Sa for a S if:
S is a monoid (then a = 1a).
a E(S) (then a = aa).
a is regular, i.e. there exists x S with a = axa (then a = (ax)a).
But in (N, +) we have 1 , 1 +N. Dually,
aS
1
= aS a
and similarly
S
1
aS
1
= SaS aS Sa a.
For , = I S then we have I is an ideal S
1
IS
1
I.
Claim. aS
1
is the smallest right ideal containing a.
Proof. a = a1 aS
1
and (aS
1
)S = a(S
1
S) aS
1
. So, aS
1
is a right ideal containing a. If
a I and I is a right ideal, then aS
1
IS
1
= I IS I. Then aS
1
is the principal right
ideal generated by a. S
1
a is the principal left ideal generated by a. S
1
aS
1
is the smallest
ideal containing a its the called the principal ideal generated by a.
If S is commutative then aS
1
= S
1
a = S
1
aS
1
.
SEMIGROUP THEORY 19
Example 5.5. In a group G we have
aG
1
= G = G
1
a = G
1
aG
1
for all a G.
Example 5.6. In N under addition we have
I
n
= n, n + 1, n + 2, . . . = n +N
1

Example 5.7. B is simple, so

B(m, n)B = B
1
(m, n)B
1
= B
Claim. (m, n)B = (m, n)B
1
=
_
(x, y) [ x m, y N
0
_
Proof. We have
(m, n)B =
_
(m, n)(u, v) [ (u, v) B
_

_
(x, y) [ x m, y N
0
_
.
Let x m then
(m, n)
_
n + (x m), y
_
=
_
mn + n + (x m), y
_
,
= (x, y).
Therefore (x, y) (m, n)B
_
(x, y) [ x m, y N
0
_
(m, n)B. hence we have proved
our claim.
Dually we have B(m, n) =
_
(x, y) [ x N
0
, y n
_
.
Lemma 5.1 (Principle Left Ideal Lemma). The following statements are equivalent;
i) S
1
a S
1
b,
ii) a S
1
b,
iii) a = tb for some t S
1
,
iv) a = b or a = tb for some t S.
Note. If S
1
a = Sa and S
1
b = Sb, then the Lemma can be adjusted accordingly.
Proof. It is clear that (i) (iii) (iv) (i) and so we prove (i) (ii).
(i) (ii): If S
1
a S
1
b then a = 1a S
1
a S
1
b a S
1
b.
(ii) (i): If a S
1
b, then as S
1
a is the smallest left ideal containing a, and as S
1
b is a
left ideal we have S
1
a S
1
b.
Lemma 5.2 (Principle Right Ideal Lemma). The following statements are equivalent:
i) aS
1
bS
1
,
ii) a bS
1
,
20 VICKY G
iii) a = bt for some t S
1
,
iv) a = b or a = bt for some t S.
Note. If aS = aS
1
and bS = bS
1
then aS bS a bS a = bt for some t S.
Denition: The relation L on a semigroup S is dened by the rule
aLb S
1
a = S
1
b
for any a, b S.
Note.
(1) L is an equivalence.
(2) If aLb and c S then S
1
a = S
1
b, so S
1
ac = S
1
bc and hence acLbc, i.e. L is right
compatible.
(3) a right (left) compatible equivalence is a right (left) congruence. Thus L is a right
congruence.
Corollary 5.1. We have that
aLb s, t S
1
with a = sb and b = ta.
Proof. We start with aLb
aLb S
1
a = S
1
b
S
1
a S
1
b and S
1
b S
1
a
s, t S
1
with a = sb, b = ta
by the Principle Left Ideal Lemma. We note that this statement about L can be used as
a denition of L.
Remark.
(1) aLb a = b or there exists s, t S with a = sb, b = ta.
(2) If Sa = S
1
a and Sb = S
1
b, then aLb s, t S with a = sb, b = ta.
Dually, the relation 1 is dened on S by
a1b aS
1
= bS
1
,
s, t S
1
with a = bs and b = at,
a = b or s, t S with a = bs and b = at.
We can adjust this if aS
1
= aS as before. Now 1 is an equivalence; it is left compatible
and hence a left congruence.
Denition: We dene the relation H = L 1 and note that H is an equivalence.
The relations L, 1, H are in fact three of Greens relations.
SEMIGROUP THEORY 21
Example 5.8. If S is commutative, L = 1 = H. In a group G,
G
1
a = G = G
1
b and aG
1
= G = bG
1
for all a, b G.
So aLb and a1b for all a, b G. Therefore L = 1 = = GG and hence we must have
H = .
Example 5.9. In N under + we have
a +N
1
= a, a + 1, . . .
and so a +N
1
= b +N
1
a = b. Hence L = 1 = H = .
Example 5.10. In B we know
(m, n)B =
_
(x, y) [ x m, y N
0
_
and so we have
(m, n)B = (p, q)B m = p.
Hence (m, n)1(p, q) m = p. Dually,
(m, n)L(p, q) n = q.
Thus (m, n)H(p, q) (m, n) = (p, q), which gives us H = .
5.2. L and 1 in T
X
Claim. T
X
T
X
Ker Ker . [Recall Ker =
_
(x, y) X X [ x = y
_
].
Proof. () Suppose T
X
T
X
. Then = for some T
X
. Let (x, y) Ker . Then
x = x() = (x) = (y) = y() = y.
Hence (x, y) Ker and so Ker Ker .
() Suppose Ker Ker . Dene : X X by
(x) = x

x
x
x
22 VICKY G
Thus we have y = x
o
, y , Im where x
0
is xed. If z = x = y, then (x, y) Ker
Ker so x = y. Hence is well-dened. So T
X
and = . Therefore T
X
so
that by the Principle Ideal Lemma, T
X
T
X
.
Corollary 5.2 (1T
X
-Lemma). 1 Ker = Ker .
Proof. We have
1 T
X
= T
X
T
X
T
X
and T
X
T
X
Ker Ker and Ker Ker
Ker = Ker .

Fact: T
X
T
X
Im Im (Exercise 4).
Corollary 5.3 ( T
X
-Lemma). L Im = Im.
Consequently H Ker = Ker and Im = Im.
Example 5.11. Let us dene
=
_
1 2 3
2 2 3
_
E(T
3
)
Now we have Im = 2, 3. We can see that Ker has classes 1, 2, 3. So
H Im = Im and Ker = Ker
Im = 2, 3 and Ker has classes 1, 2, 3.
So we have
=
_
1 2 3
3 3 2
_
or = =
_
1 2 3
2 2 3
_



which is the table of a 2-element group. Thus the H-class of is a group.
SEMIGROUP THEORY 23
5.3. Subgroups of Semigroups
S is a semigroup, H S. Then H is a subgroup of S if it is a group under (the restriction
of) the binary operation on S to H; i.e.
a, b H ab H
e H with ea = a = ae for all a H
a H b H with ab = e = ba
Remark.
(1) S does not have to be a monoid. Even if S is a monoid, e does not have to be 1.
However, e must be an idempotent, i.e. e E(S).
(2) If H is a subgroup with identity e, then e is the only idempotent in H.
S
H
e
Figure 2. e is the only idempotent in H.
(3) If e E(S), then e is a trivial subgroup.
(4) o
X
is a subgroup of T
X
; , (as above) is a subgroup of T
3
. Notice
HI
X
Im = ImI
X
and Ker = Ker I
X
,
Im = X and Ker = ,
is onto and is 1:1,
o
X
.
Therefore o
X
is the H-class of I
X
.
Denition: L
a
is the L-class of a; R
a
is the 1-class of a and H
a
is the H-class of a.
Now L
a
= L
b
aLb and H
a
= L
a
R
a
. In B, L
(2,3)
=
_
(x, 3) [ x N
0
_
.
5.4. Maximal Subgroup Theorem
Let e E(S). Then H
e
is the maximal subgroup of S with identity e.
Lemma 5.3 (Principle Ideal for Idempotents). Let a S, e E(S). Then
(i) S
1
a S
1
e ae = a
(ii) aS
1
eS
1
ea = a.
24 VICKY G
S
H
e
e
Figure 3. Existence of a Maximal Subgroup.
Proof. (We prove part (i) only because (ii) is dual). If ae = a, then a S
1
e so S
1
a S
1
e
by the Principle Ideal Lemma. Conversely, if S
1
a S
1
e then by the Principle Ideal Lemma
we have a = te for some t S
1
. Then
ae = (te)e = t(ee) = te = a.

Corollary 5.4. Let e ES. Then we have

a1e ea = a,
aLe ae = a,
aHe a = ae = ea.
Idempotents are left/right/two-sided identities for their 1/L/H-classes.
Proof. Let G be a subgroup with idempotent e. Then for any a G we have ea = a = ae
and there exists a
1
G with aa
1
= e = a
1
a. Then
ea = a
aa
1
= e
_
a1e
ae = a
a
1
a = e
_
aLe
aHe.
Therefore G He.
Theorem 5.1 (Maximal Subgroup Theorem). Let e E(S). Then H
e
is the maximal
subgroup of S with identity e.
Proof. We have shown G a subgroup with identity e G H
e
. We know H
e
is a subgroup
with identity, e. We know that e is an identity for H
e
. Suppose a, b H
e
. Then bHe,
so b1e hence ab1ae (1 is left compatible) so ab1ae = a1e. Also, aLe abLeb = bLe
hence abHe so ab H
e
. It remains to show that for all a H
e
there exists b H
e
with
ab = e = ba.
SEMIGROUP THEORY 25
Let a H
e
by denition of H = 1 L there exists s, t S
1
with
at =
..
aRe
e = sa
..
aLe
.
We have e = ee = ate
e ee = eee = a(ete) = (ese)a.
Let b = ete, c = ese so b, c S and eb = be = b, ec = ce = c. Also e = ab = ca. Now
b = eb = (ca)b = c(ab) = ce = c.
Finally,
eb = b ba = e
. .
bRe
be = b ab = e
. .
bLe
so bHe b H
e
. Hence H
e
is a subgroup.
Lets take 2 distinct idempotents e, f E(S) with e ,= f. Since H
e
and H
f
are subgroups
H
e
,= H
f
H
e
H
f
= .
Corollary 5.5 (Maximal Subgroup Theorem). Let a S. Then (i) (ii) (iii)
(iv).
(i) a lies in a subgroup,
(ii) aHe, some e E(S),
(iii) H
a
is a subgroup,
(iv) aHa
2
.
Proof. (i) (ii) a G G H
e
where e
2
= e is the identity for G. Therefore a H
e
so
aHe.
(ii) (iii) aHe H
a
= H
e
and by MST H
e
is a subgroup.
(iii) (i) a H
a
.
(iii) (iv) If H
a
is a subgroup, then certainly H
a
is closed. Hence a, a
2
H
a
therefore
aHa
2
.
Theorem 5.2 (Greens Theorem). a S, then a lies in a subgroup i aHa
2
.
Proof. see later.
Subgroups of T
n
We use Greens Theorem to show the following
Claim. lies in a subgroup of T
n
the map diagram has no tails of length 2.
26 VICKY G
Proof. lies in a subgroup H
2
L
2
, 1
2
Im = Im
2
, Ker = Ker
2
.
We know Im
2
Im (as T
n

2
T
n
). Let be an equivalence on a set X. Put
X

=
_
[x] [ x X
_
for X = 1, 2, . . . , n = n and T
n
[n/ Ker [ = [ Im[.
Claim (Mini Claimette). For T
n
, Im = Im
2
Ker = Ker
2
.
Proof. Ker Ker
2
(T
n
T
n
) so,
xKer y (x, y) Ker (x, y) Ker
2
Ker classes
Ker
2
classes
Figure 4. The classes of Ker and Ker
2
.
Suppose Im = Im
2
. If Ker Ker
2
then there exists (u, v) Ker
2
Ker . Then
[ Im
2
[ =

n
Ker
2

<

n
Ker

= [ Im[
a contradiction. Hence Ker = Ker
2
.
Hence: lies in a subgroup Im = Im
2
. An arbitrary element of T
n

Im Im
2
So, Im = Im
2
, tails of length 2.
SEMIGROUP THEORY 27
Example 5.12.
(1) We take an element of T
5
to be
=
_
1 2 3 4 5
3 1 4 3 1
_
T
5
.
This has map diagram
2
5
1 3 4
Now has no tails with length 2 and therefore doesnt lie in any subgroup.
(2) Let us take the constant element c
1
T
5
=
_
1 2 3 4 5
1 1 1 1 1
_
= c
1
.
This has the following map diagram
1
2 3 4 5
Now has no tails of length 2, therefore lies in a subgroup and hence lies in
H

. Note
2
= .
Now for any
H

H,
1 and L,
Ker = Ker and Im = Im,
Ker has classes 1, 2, 3, 4, 5 and Im = 1,
= .
Therefore the maximal subgroup containing is H

= .
(3) Take the element
28 VICKY G
=
_
1 2 3 4 5
2 3 3 3 5
_
.
This has map diagram
1
3 4
5
2
No tails of length 2. Therefore lies in a subgroup. Hence lies in a maximal
subgroup. Hence the maximal subgroup containing is H

. For any
H

H,
1 and L,
Ker = Ker and Im = Im,
Im = 2, 3, 5 and Ker has classes 1, 3, 2, 4, 5.
We now gure out what the elements of H

. We start with the idempotent. We

know that the image of the idempotent is 2, 3, 5 and that idempotents are iden-
tities on their images. Thus we must have
=
_
1 2 3 4 5
2 3 5
_
.
We also know that 1 and 3 go to the same place and 2 and 4 go to the same place.
Thus we must have
=
_
1 2 3 4 5
3 2 3 2 5
_
.
We now have what the idempotent is and then the other elements of H

are:
=
_
1 2 3 4 5
2 3 2 3 5
_
=
5

2
=
_
1 2 3 4 5
5 2 5 2 3
_

3
=
_
1 2 3 4 5
3 5 3 5 2
_
=
_
1 2 3 4 5
5 3 5 3 2
_

2
=
_
1 2 3 4 5
2 5 2 5 3
_
.
These are all 6 elements because there are 3 options for the image
SEMIGROUP THEORY 29
_
3
2
__
2
2
_
= 6.
Check H

S
3
.
5.5. T and
Recall S
1
aS
1
= xay [ x, y S
1
.
Denition: We say that a is related to b if and only if
ab S
1
aS
1
= S
1
bS
1
s, t, u, v S
1
with a = sbt b = uav
Note. If aLb, then S
1
a = S
1
b so
S
1
aS
1
= S
1
bS
1
and so ab, i.e. L , dually 1 .
Recall: S is simple if S is the only ideal of S. If S is simple and a, b S then
S
1
aS
1
= S = S
1
bS
1
so ab
and = (the universal relation). Conversely if = and InormS, then pick any a I
and any s S. We have
s S
1
sS
1
= S
1
aS
1
I.
Therefore I = S and S is simple. Thus we have that S is simple = .
Similarly if S has a zero, then 0 and S 0 are the only -classes i 0 and S are the
only ideals.
5.6. Composition of Relations
Denition: If , are relations on A we dene
=
_
(x, y) AA [ z A with (x, z) and (z, y)
_
.
Claim. If , are equivalence relations and if = then is an equivalence
relation. Also, its the smallest equivalence relation containing .
Proof. Put = =
for any a A, aaa so aa and is reexive.
Symmetric - an exercise.
30 VICKY G
Suppose that abc then there exists x, y A with
axbyc.
From xby we have xy, so
axyc.
Therefore xc hence there exists z A such that axzc, therefore azc and
hence ac. Therefore is transitive.
We have shown that is an equivalence relation. If (a, b) then abb so (a, b) .
Similarly if (a, b) then aab so (a, b) . Hence .
Now, suppose where is an equivalence relation. Let (a, b) . Then we have
acb for some c. Hence acb so ab as is transitive. Therefore .
Denition: T = 1 L, i.e. aTb c S with a1cLb.
Lemma 5.4 (The T Lemma). 1 L = L 1
Proof. We prove , the proof of being dual. Suppose that a1 Lb. Then there exists
c S with
a1cLb
There exists u, v, s, t S
1
with
a = cu
(1)
c = av
(2)
c = sb
(3)
b = tc
(4)
.
Put d = bu then we have
a =
(1)
cu =
(3)
sbu = sd,
d = bu =
(4)
tcu =
(1)
ta.
Therefore aLd. Also
b =
(4)
tc =
(2)
tav =
(1)
tcuv =
(4)
buv = dv.
Therefore b1d and hence a(L 1)b.
Hence T is an equivalence relation T = L 1. Now by denition
H = L 1 L T,
H = L 1 1 T.
As is an equivalence relation and L 1 we must have T . This has Hasse
Diagram
SEMIGROUP THEORY 31
L
H
1

T
Notation: D
a
is the T class of a S and J
a
is the -class of a S.
Note. H
a
L
a
D
a
J
a
and also H
a
R
a
D
a
J
a
.
Egg-Box Pictures
Let D be a T-class. Let a D then D = D
a
. Consider R
u
L
v
. Since uTv there exists
h S with u1hLv. No cell is empty. Moreover
R
u
L
v
= R
h
L
h
= H
h
.
Every class is a H-class. As T is an equivalence, S is the union of such egg-boxes.
5.7. Structure of T-classes
Let S be a semigroup, s S
1
. We dene
s
: S S by a
s
= as for all a S
Lemma 5.5 (Greens Lemma). Let a, b S be such that a1b and let s, s

S be such that
as = b and bs

= a.
Then
s
: L
a
L
b
and
s
: L
b
L
a
are mutually inverse, 1-class preserving bijections
(i.e. if c L
a
, then c1c
s
and if s L
b
then d1d
s
)
Proof. If c L
a
then
c
s
= csLas = b
because L is a right congruence. So c
s
Lb therefore
s
: L
a
L
b
. Dually
s
: L
b
L
a
.
Let c K
a
. Then c = ta for some t S. Now
c
s

s
= tas
s
= tass

= tbs

= ta = c.
So
s

s
= I
La
, dually,
s

s
= I
L
b
.
Again, let c L
a
. Then
32 VICKY G
cs = c s,
c = cs s

.
Therefore c1cs = c
s
.
Lemma 5.6 (Continuing Greens Lemma). For any c L
a
we have
s
: H
c
H
cs
is a
bijection with inverse
s
: H
cs
H
c
. In particular put c = a then

s
: H
a
H
b
and
s
: H
b
H
a
are mutually inverse bijections. For any s S
1
,
s
: S S is given by a
s
= sa.
Lemma 5.7 (Dual of Greens Lemma). Let a, b S be such that aLb and let t, t

S
be such that ta = b and t

b = a. Then
t
: R
a
R
b
,
t
: R
b
R
a
are mutually
inverse L-class preserving bijections. In particular, for any c R
a
we have
t
: H
c
H
tc
,

t
: H
tc
H
c
are mutually inverse bijections. So, if c = a we have
t
: H
a
H
b
,

t
: H
b
H
a
are mutually inverse bijections.
Corollary 5.6. If aTh then there exists a bijection H
a
H
b
Proof. If aTb then there exists h S with a1hLb. There exists a bijection H
a
H
h
by
Greens Lemma and we also have that there exists a bijection H
h
H
b
by the Dual of
Greens Lemma. Therefore there exists a bijection H
a
H
b
.
Thus any two H-classes in the same T-class have the same cardinality.
Theorem 5.3 (Greens Theorem Strong Version). Let H be an H-class of a semigroup
S. Then either H
2
H = or H is a subgroup of S.
Proof. Suppose H
2
H ,= then there exists a, b, c H with ab = c. Since a1c,
b
:
H
a
H
c
is a bijection. But H
a
= H
c
= H so
b
: H H is a bijection. Hence Hb = H.
Dually, aH = H.
Since b H, b = db for some d H. As b1d, d = bs for some s S
1
and then
bs = dbs d = d
2
. Hence H contains an idempotent (so by the Maximal Subgroup
Theorem, its a subgroup).
We have d is an identity for H. Let h, k H. Then hLd so
hd = h
h
: H H
is a bijection. Hence k
h
= hk H. So H is closed. Let a H, then a
2
H, hence
aH = H = Ha. So, there exists a

, a

H with
d = aa

= a

a
and we have
a

= a

d = a

(aa

) = (a

a)a

= da

= a

.
SEMIGROUP THEORY 33
Corollary 5.7.
(i) aHa
2
H
a
is a subgroup,
(ii) e E(S) H
e
is a subgroup.
Proof.
(i) We know H
a
is a subgroup a, a
2
H
a
so aHa
2
. If aHa
2
, then a
2
H
a
(H
a
)
2
.
Hence H
a
(H
a
)
2
,= . So, by Green H
a
is a subgroup.
6. Rees Matrix Semigroups
Construction: Let G be a group, I, be non-empty sets and P = (p
i
) a matrix. P is a
I matrix over G0 such that every row / column of P contains at least one non-zero
entry.
Denition: m
0
= m
0
(G; I, ; P) is the set
I G 0
with binary operation given by 0n = 0 = n0 for all n m
0
and
(i, a, )(k, b, ) =
_
0 if p
k
= 0,
(i, ap
k
b, ) if p
k
,= 0.
Then m
0
is a semigroup with zero 0 a Rees Matrix Semigroup over G.
Denition: a S is regular if there exists x S with
a = axa.
S is regular if every a S is regular.
If S is regular then a1b aS = bS there exists s, t S with a = bs and b = at, etc.
6.1. Rees Matrix Facts
Let m
0
= m
0
(G; I, ; P) be a Rees Matrix Semigroup over a group G. Then
(1) (i, a, ) is idempotent p
i
,= 0 and p
i
= a
1
.
(2) m
0
is regular.
(3) (i, a, )1(j, b, ) i = j.
(4) (i, a, )L(j, b, ) = .
(5) (i, a, )H(j, b, ) i = j and = .
(6) The T = -classes are 0 and m
0
0 (so 0 and m
0
are the only ideals).
(7) m
0
is 0-simple.
(8) The rectangular property;
xyTx xy1x
xyTy xyLy
_
x, y m
0
34 VICKY G
(9) Put H
i
=
_
(i, a, ) [ a G
_
by (5) we have H
i
is an H-class (H
i
= H
(i,e,)
). If

i
,= 0 we know (i,
1
i
, ) is an idempotent H
i
is a group, by the Maximal
Subgroup Theorem. The identity is (i
1
i
, ) and (i, a, )
1
= (i,
1
i
a
1
,
1
i
, ).
(10) If
i
,= 0 and
j
,= 0 then (exercise) H
i
H
j
(the bijection is (i, a, )
(j, a, )). The homomorphism is a bit more sophisticated.
Proof.
(1) We have that
(i, a, ) E(m
0
) (i, a, ) = (i, a, )(i, a, ),
(i, a, ) = (i, ap
i
a, ),
a = ap
i
a,
p
i
,= 0 and p
i
= a
1
.
(2) 0 = 000 so 0 is regular. Let (i, a, ) m
0
0 then there exists j I with p
j
,= 0
and there exists with p
i
,= 0. Now,
(i, a, )(j, p
1
j
a
1
p
1
i
, )(i, a, ) = (i, a, )
and hence m
0
is regular.
(3) 0 is an 1-class. If (i, a, )1(j, b, ) then there exists (k, c, ) m
0
with
(i, a, ) = (j, b, )(k, c, ) = (j, bp
k
c, )
and so i = j. Conversely, if i = j, pick k with p
k
,= 0. Then
(i, a, ) = (j, b, )(k, p
1
k
b
1
a, )
and consequently (i, a, )1(j, b, )
(4) Dual.
(5) This comes from (3) and (4) above.
(6) 0 is a T-class and a -class. If (i, a, ), (j, b, ) m
0
then
(i, a, )1(i, a, )L(j, b, )
so (i, a, )T(j, b, ) and so (i, a, )(j, b, ). Therefore T = and 0 and m
0
0
are the only classes.
(7) Let i I, then there exists with p
i
,= 0 so (i, 1, )
2
,= 0. Therefore (m
0
)
2
,= 0
and so m
0
is 0-simple.
(8) The 2 T-classes are zero and everything else. If xy = x = 0 then xyTx and xy1x
always. If xy1x then xyTx as 1 T. Suppose xyT and xy, x ,= 0. Then y ,= 0,
so
x = (i, a, ) y = (j, b, )
say. Then as xy = (i, a
j
b, ) so xy1x. The result for L is dual.
SEMIGROUP THEORY 35
A nitary property is a property that captures nite nature.
Chain conditions
Denition: S a semigroup has M
L
if there are no innite chains
S
1
a
1
S
1
a
2
S
1
a
3
. . .
if principal left ideals. M
L
is the descending chain condition (d.c.c.) on principal left
ideals. Note that M
R
is the dual of this.
Claim (The Chain Claim). S has M
L
if and only if any chain
S
1
a
1
S
1
a
2
. . .
terminates stabilizes with
S
1
a
n
= S
1
a
n+1
= . . .
Proof. If every chain with terminates, then clearly we cannot have an strict chain
S
1
a
1
S
1
a
2
. . .
So S has M
L
. Conversely; suppose S has M
L
and we have a chain
S
1
a
1
S
1
a
2
. . .
The strict inclusions are at the j
i
th steps
S
1
a
1
= S
1
a
2
= = S
1
a
j
1
S
1
a
j
1
+1
= S
1
a
j
1
+2
= = S
1
a
j
2
S
1
a
j
2
+1
= . . .
Then S
1
a
j
1
S
1
a
j
2
. . . . As S has M
L
, this chain is nite with length n say. Then
S
1
a
jn+1
= S
1
a
jn+2
= . . .
and our sequence has stabilised.
Denition: The ascending chain condition (a.c.c.) on principal ideals on left / right ideals
M
L
(M
R
) is dened as above but with the inclusions reversed. The analogue of the chain
claim holds.
Example 6.1. Every nite semigroup has M
L
, M
R
, M
L
, M
R
. A chain
S
1
a
1
S
1
a
2
is in particular, a chain of distinct subsets of S but a nite semigroup S has at most 2
|s|
subsets.
36 VICKY G
Example 6.2. The Bicyclic semigroup B has M
L
& M
R
. We know B(x, y) =
_
(p, q) [
q y
_
and so B(x, y) B(u, v) y v, inclusion is strict if and only if y > v. If we
had an chain
B(x
1
, y
1
) B(x
2
, y
2
) B(x
3
, y
3
) . . .
then we would have
y
1
> y
2
> y
3
> . . .
hence M
L
holds, dually M
R
holds. However, since 0 < 1 < 2 < . . . we have
B(0, 0) B(1, 1) B(2, 2) . . .
so there exists descending chains. Hence B doesnt have M
L
or M
R
.
Example 6.3. Let m
0
= m
0
(G; I; ; P) be a Rees Matrix Semigroup over a group G.
Then m
0
has M
L
, M
R
, M
L
and M
R
.
Proof. We show that the length of the strict chains is at most 2. Suppose m
0
m
0
.
We could have = 0. If ,= 0 then m
0
,= 0 so ,= 0 and we have = (i, g, ),
= (j, h, ) and = for some = (, k, ) so that
(i, g, ) = (j, h, )(, k, ) = (j, h

k, )
i = j and so 1 and m
0
= m
0
. So we have 0m
0
m
0
for all non-zero . But
,= 0, m
0
m
0
m
0
= m
0
. Hence m
0
has M
R
and M
R
; dually m
0
has M
L
and
M
L
.
Denition: A 0-simple semigroup is completely 0-simple if it has M
R
and M
L
.
By above, any Rees Matrix Semigroup over a group is completely 0-simple. Our aim is to
show that every completely 0-simple semigroup is isomorphic to a Rees Matrix Semigroup
over a group.
Theorem 6.1 (The T = Theorem). Suppose
()
_
a S, n N with a
n
La
n+1
,
a S, m N with a
m
1a
m+1
.
Then T = .
Example 6.4.
(1) If S is a band, a = a
2
for all a S and so () holds.
Proof. We know T . Let a, b S with ab. Then there exists x, y, u, v S
1
with
b = xay a = ubv.
Then
SEMIGROUP THEORY 37
b = xay = (xu)b(vy) = (xu)(xubvy)(vy) = (xu)
2
b(vy)
2
= = (xu)
n
b(vy)
n
for all n N. By () there exists n with (xu)
n
L(xu)
n+1
. Therefore
b = (xu)
n
b(vy)
n
L(xu)
n+1
b(vy)
n
= xu
_
(xu)
n
b(vy)
n
_
= xub.
Therefore bLxub, so
S
1
b = S
1
xub S
1
ub S
1
b.
So S
1
b = S
1
ub and bLub. Dually, b1bv. Therefore a = ubv1ubLb. So aTb and T.
Consequently, T = .
The Rectangular Property: Let S satisfy (). Then for all a, b S we have
(i) aab aTab a1ab,
(ii) bab bTab bLab.
Proof. We prove (i), (ii) being dual. Now,
aab aTab
as T = . Then a1ab aTab; as 1 T. If aab then there exists x, y S
1
with
a = xaby = xa(by) = x
n
a(by)
n
for all n. Pick n with (by)
n
1(by)
n+1
. Then
a = x
n
a(by)
n
1x
n
a(by)
n+1
= x
n
a(by)
n
by = aby
a1aby. Now aS
1
= abyS
1
abS
1
aS
1
. Hence aS
1
= abS
1
and a1ab.
Lemma 6.1 (0-Simple Lemma). Let S have a 0 and S
2
,= 0. Then the following are equal
(i) S is 0-simple,
(ii) SaS = S for all a S 0,
(iii) S
1
aS
1
= S for all a S 0,
(iv) the -classes are 0 and S 0.
Proof. (i) (iv) is a standard exercise.
(ii) (iii): Let a S 0. Then
S = SaS S
1
aS
1
S
and therefore S = S
1
aS
1
.
(iii) (iv): We know J
0
= 0. Let a, b S 0. Then
S
1
aS
1
= S = S
1
bS
1
and hence ab. Therefore 0 and S 0 are the only -classes.
38 VICKY G
(i) (ii): Since S
2
,= 0 and S
2
is an ideal, then S
2
= S. Therefore
S
3
= SS
2
= S
2
= S ,= 0.
Let I = x S [ SxS = 0. Clearly 0 I and hence I ,= . If x I and s S, then
0 SxsS SxS = 0.
Therefore SxsS = 0 and so xs I. Dually sx I; therefore I is an ideal. If I = S, then
S
3
= SIS,
=
_
xI
SxS,
= 0.
This is a contradiction, therefore I ,= S. Hence I = 0. Let a S 0. Then SaS is an
ideal and as a , I we have SaS ,= 0. Hence SaS = S.
Corollary 6.1. Let S be completely 0-simple. Then S contains a non-zero idempotent.
Proof. Let a S 0. Then SaS = S, therefore there exists a u, v S with a = uav. So,
a = uav = u
2
av
2
= = u
n
av
n
for all n. Hence u
n
,= 0 for all n N. Pick n, m with u
n
1u
n+1
, u
m
Lu
m+1
. Notice
u
n+1
1u
n+2
as 1 is a left congruence. Similarly,
u
n+2
1u
n+3
we deduce that u
n
1u
n+t
for all t 0. Similarly u
m
Lu
m+t
for all t 0. Let s = maxm, n.
Then u
s
1u
2s
, u
s
Lu
2s
so u
s
Hu
2s
= (u
s
)
2
. Hence u
s
lies in a subgroup. Therefore u
s
He for
some idempotent e. As u
s
,= 0 and H
0
= 0, we have e ,= 0.
Theorem 6.2 (Rees Theorem - 1941). Let S be a semigroup with zero. Then S is com-
pletely 0-simple S is isomorphic to a Rees Matrix Semigroup over a group.
Proof. If S

= m
0
(G; I; ; P) where G is a group, we know m
0
is completely 0-simple (by
Rees Matrix facts), hence S is completely 0-simple.
Conversely, suppose S is completely 0-simple. By the T = Theorem, T = (as S has
M
R
and M
L
, it must have ()). As S is 0-simple, the T = -classes are 0 and S 0.
Let D = S 0. By the Corollary to the 0-simple Lemma, D contains e = e
2
.
Let R
i
[ i I be the set of 1-classes in D (so I indexes the non-zero 1-classes). Let
L

[ be the set of L-classes in D (so index the non-zero L-classes).

SEMIGROUP THEORY 39
Denote the H-class R
i
L

by H
i
. Since D contains an idempotent e, D contains the sub-
group H
e
(Maximum Subgroup Theorem or Greens Theorem). Without Loss of Generality
assume e H
11
. Put G = H
11
, which is a group.
For each let q

H
1
(take q
1
= e). For each i I let r
i
H
i1
(take r
1
= e).
e = e
2
, e1q

eq

= q

By Greens Lemma,

: H
e
= G H
1
is a bijection. Now, e = e
2
, eLr
i
so r
i
e = r
i
. By the dual of Greens Lemma

r
i
: H
1
H
i
is a bijection. Therefore for any i I, we have

r
i
: G H
i
is a bijection.
Note. a
q

r
i
= r
i
aq

.
So, each element of H
i
has a unique expression as r
i
aq

where a G. Hence the mapping

: (I G) 0 S
given by 0 = 0, (i, a, ) = r
i
aq

is a bijection. Put p
i
= q

r
i
. If p
i
,= 0 then q

r
i
Tq

.
By the rectangular property q

r
i
1q

1e. Also by the rectangular property, if q

r
i
,= 0 then
as q

r
i
Tr
i
we have
q

r
i
Lr
i
Le.
Therefore q

r
i
= 0 or q

r
i
G. So, P = (p
i
) = (q

r
i
) is a I matrix over G0. For
any i I, by the 0-simple Lemma we have Sr
i
S = S. So, ur
i
v ,= 0 for some u, v S. Say,
u = r
k
bq

for some k, and b. Then

p
i
= q

r
i
,= 0
as r
k
bq

r
i
v ,= 0. Therefore every element of P has a non-zero entry. Dually for rows.
Therefore
m
0
= m
0
(G; I; ; P)
is a Rees Matrix Semigroup over a group G. For any x m
0
(x = 0 or x a triple) then
(0x) = 0 = 0 = 0(x) = 0x.
Also, (x0) = x0. For (i, a, ), (k, b, ) m
0
we have
40 VICKY G
_
(i, a, )(k, b, )
_
=
_
0 if p
k
= 0,
(i, ap
k
b, ) if p
k
,= 0,
=
_
0 if p
k
= 0,
r
i
ap
k
bq

if p
k
,= 0,
= r
i
ap
k
bq

,
= r
i
aq

r
k
bq

,
= (i, a, )(k, b, ).
Therefore is an isomorphism.
7. Regular Semigroups
Denition: a S is regular if a = axa for some x S. S is regular if every a S is
regular.
Denition: a

S is an inverse of a if a = aa

a and a

= a

aa

. We denote V (a) to be
the set of inverses of a.
Caution: Inverses need not be unique. In a rectangular band T = I then
(i, j)(k, )(i, j) = (i, j)
(k, )(i, j)(k, ) = (k, )
for any (i, j) and (k, ). So every element is an inverse of every other element. If G is a
group then V (a) = a
1
for all a G.
Lemma 7.1 (Lemma R). a is regular V (a) ,= .
Proof. If V (a) ,= , clearly a is regular. Conversely suppose that a is regular. Then there
exists x S with a = axa. Put a

= xax. Then
aa

a = a(xax)a = (axa)xa = axa = a,

a

aa

= (xax)a(xax) = x(axa)(xax) = xa(xax) = x(axa)x = xax = a.

So a

V (a).
Note. If a = axa then
(ax)
2
= (ax)(ax) = (axa)x = ax
so ax E(S) and dually, xa E(S). Moreover
SEMIGROUP THEORY 41
a = axa ax = ax a1ax,
a = axa xa = xa aLxa.
a
xa
ax
Figure 5. The egg box diagram of E(S).
Denition: S is inverse if [V (a)[ = 1 for all a S, i.e. every element has a unique
inverse.
Example 7.1.
(1) Groups are inverse; V (a) = a
1
,
(2) A rectangular band T is regular; but (as every element of T is an inverse of every
other element) T is not inverse (unless T is trivial),
(3) If S is a band then S is regular as e = e
3
for all e S; S need not be inverse,
(4) B is regular because (a, b) = (a, b)(b, a)(a, b) for all (a, b) B then B is inverse
see later,
(5) m
0
is regular (see Rees Matrix Facts),
(6) T
X
is regular (see Exercises),
(7) (N, +) is not regular as, for example 1 ,= 1 + a + 1 for any a N.
Theorem 7.1 (Inverse Semigroup Theorem). A semigroup S is inverse i S is regular
and E(S) is a semilattice (i.e. ef = fe for all e, g E(S)).
Proof. () Let a S. As S is regular, a has an inverse by Lemma R. Suppose x, y V (a).
Then
a =
(1)
axa x =
(2)
xax a =
(3)
aya y =
(4)
yay,
so ax, xa, ay, ya E(S). This gives us that
x =
(2)
xax =
(3)
x(aya)x = (xa)(ya)x = (ya)(xa)x = y(axa)x
=
(1)
yax =
(3)
y(aya)x = y(ay)(ax) = y(ax)(ay) = y(axa)y =
(1)
yay =
(4)
y.
42 VICKY G
So [V (a)[ = 1 and S is inverse. Conversely suppose S is inverse. Certainly S is regular.
Let e E(S). Then e = e

as e = eee, e = eee. Let x = (ef)

. Consider fxe. Then

(fxe)
2
= (fxe)(fxe) = f(xefx)e = fxe
as x = (ef)

. So (fxe) E(S) and therefore fxe = (fxe)

. We want to show that fxe

and ef are mutually inverse, i.e.
ef(fxe)ef = ef
2
xe
2
f = efxef = ef,
(fxe)ef(fxe) = fxe
2
f
2
xe = fx(efx)e = fxe.
Therefore we have ef = (fxe)

= fxe E(S). Therefore E(S) is a band. Let e, f E(S).

Then
ef(fe)ef = ef
2
e
2
f = efef = ef,
fe(ef)fe = fe
similarly. Therefore we have ef = (fe)

= fe.
Example 7.2.
(1) E(B) =
_
(a, a) [ a N
0
_
because
(a, a)(b, b) = (t, t) = (b, b)(a, a)
where t = maxa, b. Therefore B is inverse and note (a, b)

= (b, a.
(2) T
X
we know T
X
is regular. For [X[ 2 let x, y X with x ,= y we have
c
x
, c
y
E(T
X
). Then c
x
c
y
,= c
y
c
x
so T
X
is not inverse.
(3) S a band then S is regular. We have
S is inverse ef = fe for all e, f E(S),
ef = fe for all e, f S,
S is a semilattice.
7.1. Greens Theory for Regular T-classes
If e E(S) then H
e
is a subgroup of S (by the Maximal Subgroup Theorem or Greens
Theorem). If eTf then [H
e
[ = [H
f
[ (by the Corollary to Greens Lemmas). We will show
that H
e

= H
f
.
Lemma 7.2. We have that
(i) If a = axa then ax, xa E(S) and a1xa,
(ii) If b1f E(S), then b is regular,
(iii) If bLf E(S), then b is regular.
Proof.
SEMIGROUP THEORY 43
(i) we know this.
(ii) If b1f then fb = b. Also, f = bs for some s S
1
. Therefore b = fb = bsb and so
b is regular.
(iii) this is dual to (ii).

Lemma 7.3 (Regular T-class Lemma). If aTb then if a is regular, so is b.

Proof. Let a be regular with aTb . Then a1cLb for some c S.
a
b
c e
f
Figure 6. The egg box diagram of T.
There exists e = e
2
with e1a1c by (ii) above. By (ii), c is regular. By (i), cLf = f
2
. By
(iii), b is regular.
Corollary 7.1 (Corollary to Greens Lemmas). Let e, f E(S) with eTf. Then H
e

= H
f
.
Proof. Suppose e, f E(S) and eTf. There exists a S with e1aLf.
a
f
e
As e1a there exists s S
1
with e = as and ea = a. So a = asa. Put x = fse. Then
ax = afse = ase = e
2
= e
and so a = ea = axa. Since aLf there exists t S
1
with ta = f. Then
44 VICKY G
xa = fsea = fsa = tasa = ta = f.
Also
xax = fx = ffse = fse = x.
So we have the diagram
a e
x f

x
e = ax a = axa x = xax f = xa
We have ea = a therefore
a
: H
e
H
a
is a bijection. From aLf and xa = f we have

x
: H
a
H
f
is a bijection. Hence
a

x
: H
e
H
f
is a bijection. Let h, j H
e
. Then
h(
a

x
)k(
a

x
) = (xha)(xka) = xh(ax)ka = xheka = xhka = hk(
a

x
).
So,
a

x
is an isomorphism and H
e

= H
f
.
Example 7.3.
(1) m
0
= m
0
(G; I; ; P) then m
0
0 is a T-class. We have H
i
=
_
(i, g, ) [ g G
_
.
If p
i
,= 0, H
i
is a group H-class. If p
i
, p
j
,= 0 then H
i

= H
j
(seen directly).
(2) B (the Bicyclic Monoid). B is bisimple. E(B) =
_
(a, a) [ a N
0
_
. Then H
(a,a)
=
_
(a, a)
_
. Clearly H
(a,a)

= H
(b,b)
.
(3) In T
n
, then T () = () where () = [ Im()[. By the Corollary, if
, E(T
n
) and () = () = m say, then H

= H

. In fact H

= H

= o
m
.