THE MAXIMAL SUBSEMIGROUPS OF THE IDEALS OF SOME SEMIGROUPS
OF PARTIAL INJECTIONS
Ilinka Dimitrova
Faculty of Mathematics and Natural Science South-West University "Neofit Rilski"
Blagoevgrad, 2700, Bulgaria e-mail: ilinka_dimitrova@yahoo.com
and Jörg Koppitz
∗Institute of Mathematics, Potsdam University Potsdam, 14469, Germany
e-mail: koppitz@rz.uni-potsdam.de
Abstract
We study the structure of the ideals of the semigroup IO
nof all isotone (order-preserving) partial injections as well as of the semigroup IM
nof all monotone (order-preserving or order-reversing) partial in- jections on an n-element set. The main result is the characterization of the maximal subsemigroups of the ideals of IO
nand IM
n.
Keywords: finite transformation semigroup, isotone and monotone partial transformations, maximal subsemigroups.
2000 Mathematics Subject Classification: 20M20.
∗
Supported by Humboldt Foundation.
1. Introduction
Let X
n= {1, 2, . . . , n} be an n - element set ordered in the usual way. The monoid P T
nof all partial transformations of X
nis a very interesting object.
In this paper we will multiply transformations from the right to the left and use the corresponding notation for the right to the left composition of transformations: x(αβ) = (xα)β, for x ∈ X
n. We say that a transformation α ∈ P T
nis isotone (order-preserving) if x ≤ y =⇒ xα ≤ yα for all x, y from the domain of α, antitone (order-reversing) if x ≤ y =⇒ yα ≤ xα for all x, y from the domain of α and monotone if it is isotone or antitone.
In the present paper, we study the structure of the semigroups IO
nof all isotone partial injections and IM
nof all monotone partial injections of X
n. From the definition of monotone transformations, it is clear that IO
n⊆ IM
n.
Some semigroups of transformations have been studied since the sixties.
In fact, presentations of the semigroup O
nof all isotone transformations and of the semigroup P O
nof all isotone partial transformations (excluding the permutation in both cases) were established by A˘ızen˘stat ([1]) in 1962 and by Popova ([16]), respectively, in the same year. Some years later (1971), Howie ([14]) studied some combinatorial and algebraic properties of O
nand, in 1992, Gomes and Howie ([13]) established some more properties of O
n, namely its rank and idempotent rank. In recent years it has been studied in different aspects by several authors (for example [4, 15, 17, 18]). The monoid IO
nof all isotone partial injections of X
nhas been the object of study since 1997 by Fernandes in various papers ([7, 8, 9]). Some basic properties of IO
n, in particular, a description of Green’s relations, congruences and a presentation, were obtained in [2]. Ganyushkin and Mazorchuk ([12]) studied some properties of IO
nas describe ideals, systems of generators, maximal subsemigroups and maximal inverse subsemigroups of IO
n.
In [10], Fernandes, Gomes and Jesus gave a presentation of both the
semigroups M
nof all monotone transformations of X
nand the semigroup
P M
nof all monotone partial transformations. Dimitrova and Koppitz ([4])
considered the maximal subsemigroups of M
nand its ideals. Delgado and
Fernandes ([3]) have computed the abelian kernels of the semigroup IM
n.
Fernandes, Gomes and Jesus ([11]) exhibited some properties as well as a pre-
sentation for the semigroup IM
n. Dimitrova and Koppitz ([5]) characterized
the maximal subsemigroups of IM
n.
In this paper we consider the ideals of the semigroups IO
nand IM
n. In Section 2 we describe the maximal subsemigroups of the ideals of the semi- group IO
n. Each of the considered ideals has exactly 2(
nr) − 2 maximal subsemigroups. In Section 3 we characterize the maximal subsemigroups of the ideals of the semigroup IM
n. It happens that each of the considered ideals has exactly 2(
nr)
+1− 3 maximal subsemigroups.
We will try to keep the standard notation. For every partial transfor- mation α by dom α and im α we denote the domain and the image of α, re- spectively. If α is injective, the number rank α := |dom α| = |im α| is called the rank of α. Clearly, rank αβ ≤ min{rank α, rank β} and im β = im αβ as well as dom α = dom αβ if im α = dom β. From the definition of iso- tone and antitone transformation, it follows that every element α ∈ IM
nis uniquely determined by dom α and im α satisfying |dom α| = |im α|.
Moreover, for every A, B ⊂ X
nof the same cardinality there exists one isotone transformation α ∈ IO
n⊆ IM
nand one antitone transformation β ∈ IM
nsuch that dom α = dom β = A and im α = im β = B. We will denote by α
A,Bthe unique isotone element α ∈ IM
nfor which A = dom α and B = im α, and by β
A,Bthe unique antitone element β ∈ IM
nfor which A = dom β and B = im β. The elements α
A,A, A ∈ X
n, exhaust all idempo- tents in IO
nas well as in IM
n. For the elements β
A,A, we have β
2A,A= α
A,A. In case A = B = X
n, we will use the notations α
nand β
ninstead of α
Xn,Xnand β
Xn,Xn.
The Green’s relations L, R, J and H on IO
nas well as on IM
nare characterized as follows:
αLβ ⇐⇒ im α = im β αRβ ⇐⇒ dom α = dom β αJβ ⇐⇒ rank α = rank β
H = L ∩ R.
Obviously, every H-class in IO
ncontains exactly one element and every H-
class in IM
n\ {α ∈ IM
n: rank α ≤ 1} contains exactly two elements. In
the set {α ∈ IM
n: rank α ≤ 1}, every H-class contains exactly one element.
2. Maximal subsemigroups of the ideals of IO
nThe semigroup IO
nis the union of the J-classes J
0, J
1, . . . , J
n, where J
r:= {α ∈ IO
n: rank α = r} for r = 0, . . . , n.
It is well known that the ideal I(n, r) (r = 0, . . . , n) of the semigroup IO
nis the union of J-classes J
0, J
1, . . . , J
r, i.e.
I(n, r) = {α ∈ IO
n: rank α ≤ r}.
Every principal factor on IO
nis a Rees quotient I(n, r)/I(n, r − 1) (1 ≤ r ≤ n) of which we think as J
r∪ {0} (as it is usually convenient), where the product of two elements of J
ris taken to be zero if it falls into I(n, r − 1).
Let us denote by Λ
rthe collection of all subsets of X
nof cardinality r.
The R -, L - and H - classes in J
rhave the following form:
R
A:= {α ∈ I(n, r) : dom α = A}, A ∈ Λ
r; L
B:= {α ∈ I(n, r) : im α = B}, B ∈ Λ
r; H
A,B:= {α
A,B} = R
A∩ L
B, A, B ∈ Λ
r.
Clearly, each R
A- class (L
A- class), A ∈ Λ
rcontains exactly one idempotent α
A,A. Thus if E
ris the set of all idempotents in the class J
r, then |E
r| =
nr.
Since the product αβ for all α, β ∈ J
rbelongs to the class J
rif and only if im α = dom β, it is obvious that
Lemma 1.
1. L
BR
A=
( J
r, if A = B, 0, if A 6= B.
2. α
A,Bα
C,D=
( α
A,D, if B = C, 0, if B 6= C.
Proposition 1 [7]. hJ
ri = I(n, r), for 0 ≤ r ≤ n − 1.
Now we begin with the description of the maximal subsemigroups of the ideals of the semigroup IO
n.
Let us denote by Dec(Λ
r) the set of all decompositions (N
1, N
2) of Λ
r, i.e. N
1∪ N
2= Λ
rand N
1∩ N
2= ∅ where N
1, N
26= ∅.
Definition 1. Let (N
1, N
2) ∈ Dec(Λ
r) (r = 1, . . . , n − 1). Then we put S
(N1,N2):= I(n, r − 1) ∪ {α
A,B: A ∈ N
1or B ∈ N
2}.
The maximal subsemigroups of the ideal I(n, n) = IO
nwere described by Ganyushkin and Mazorchuk:
Theorem 1 [12]. A subsemigroup S of IO
nis maximal if and only if S = I(n, n − 1) or S = {α
n} ∪ S
(N1,N2), where (N
1, N
2) ∈ Dec(Λ
n−1).
In the following, we will consider the maximal subsemigroups of the ideals I(n, r) for r = 1, . . . , n − 1.
Lemma 2. Every maximal subsemigroup in I(n, r) contains the ideal I(n, r − 1).
P roof. Let S be a maximal subsemigroup of I(n, r). Assume that J
r⊆ S, then according to Proposition 1 it follows that I(n, r) = hJ
ri ⊆ S, i.e.
S = I(n, r), a contradiction. Thus J
r* S. Then S ∪ I(n, r − 1) is a proper subsemigroup of I(n, r) since I(n, r−1) is an ideal, and hence S∪I(n, r−1) = S by maximality of S. This implies I(n, r−1) ⊆ S.
Theorem 2. Let 1 ≤ r ≤ n−1. Then a subsemigroup S of I(n, r) is maximal if and only if there is an element (N
1, N
2) ∈ Dec(Λ
r) with S = S
(N1,N2).
P roof. Let S = S
(N1,N2)for some (N
1, N
2) ∈ Dec(Λ
r). Then S = I(n, r − 1) ∪ {α
A,B: A ∈ N
1or B ∈ N
2}.
Therefore, if α
A,B∈ S then A ∈ N /
2and B ∈ N
1, and thus α
B,A∈ S.
From Lemma 1 it follows that S is a semigroup. Really, let α
A,B, α
C,D∈ S, i.e. A, C ∈ N
1or B, D ∈ N
2or A ∈ N
1, D ∈ N
2. Then we have α
A,Bα
C,D= α
A,D∈ S for B = C and α
A,Bα
C,D= 0 ∈ I(n, r − 1) ⊆ S for B 6= C.
Now we will show that S is maximal. Let α
C,D∈ I(n, r) \ S, i.e. C / ∈ N
1and D / ∈ N
2. Then D ∈ N
1, since N
1∪ N
2= Λ
rand so α
D,P∈ S for all P ∈ Λ
rand thus R
D= {α
D,P: P ∈ Λ
r} ⊆ S. Moreover, we have α
C,P= α
C,Dα
D,P, for all P ∈ Λ
r, by Lemma 1. Thus we obtain the R-class R
C= {α
C,P: P ∈ Λ
r} ⊆ hS ∪ {α
C,D}i. Moreover, C ∈ N
2and so L
C= {α
P,C: P ∈ Λ
r} ⊆ S. Using Lemma 1, we have L
CR
C= J
r⊆ hS ∪ {α
C,D}i.
Thus, we obtain that hS ∪ {α
C,D}i = I(n, r). Therefore, S is a maximal subsemigroup of the ideal I(n, r).
For the converse part let S be a maximal subsemigroup of the ideal I(n, r). From Lemma 2, we have that I(n, r − 1) ⊆ S. Then S = I(n, r − 1)
∪ T , where T ⊆ J
r.
Let α
A,B∈ S. Then hS∪{α /
A,B}i = I(n, r). Let now P, Q ∈ Λ
r. Suppose that α
P,Q∈ S. Then α /
P,Q∈ hS ∪ {α
A,B}i and α
P,Q= α
P,Aα
A,Bα
B,Q. Moreover, α
P,A= α
P,Aα
A,Bα
B,Aand α
B,Q= α
B,Aα
A,Bα
B,Q. This shows that we need α
P,Aand α
B,Qto generate α
P,Aand α
B,Q, respectively, with elements of S ∪ {α
A,B}. Hence α
P,A, α
B,Q∈ S.
Assume that α
Q,P∈ S. Then α /
Q,P= α
Q,Aα
A,Bα
B,Pand by the same arguments, we obtain that α
Q,A, α
B,P∈ S.
Further, from α
Q,P= α
Q,Aα
A,Pit follows that α
A,P∈ S. But α /
P,Q∈ / hS ∪ {α
A,P}i since α
P,Q= α
P,Aα
A,Pα
P,Q. This contradicts the maximality of S and thus α
Q,P∈ S. Hence if α
P,Q∈ S then α /
Q,P∈ S for any P, Q ∈ Λ
r. Therefore, for N
1= {B : α
A,B∈ S} and N /
2= {A : α
A,B∈ S} we have that / S = S
(N1,N2).
There are exactly 2(
nr) − 2 maximal subsemigroups of the ideal I(n, r), for r = 1, . . . , n − 1 and 2
n− 1 maximal subsemigroups of I(n, n).
3. Maximal subsemigroups of the ideals of IM
nThe semigroup IM
nis the union of the J-classes J
0, J
1, . . . , J
n, where
J
r:= {α ∈ IM
n: rank α = r} for r = 0, . . . , n.
It is well known that the ideal I(n, r) (r = 0, . . . , n) of the semigroup IM
nis the union of J-classes J
0, J
1, . . . , J
r, i.e.
I(n, r) = {α ∈ IM
n: rank α ≤ r}.
Every principal factor on IM
nis a Rees quotient I(n, r)/I(n, r − 1) (1 ≤ r ≤ n) of which we think as J
r∪ {0}, where the product of two elements of J
ris taken to be zero if it falls into I(n, r − 1).
The R -, L - and H - classes in J
rhave the following form:
R
A:= {α ∈ I(n, r) : dom α = A}, A ∈ Λ
r; L
B:= {α ∈ I(n, r) : im α = B}, B ∈ Λ
r; H
A,B:= {α
A,B, β
A,B} = R
A∩ L
B, A, B ∈ Λ
r.
The L-class, R-class and H-class, respectively, containing the element α ∈ IM
nwill be denoted by L
α, R
α, and H
α, respectively.
Since the product αβ for all α, β ∈ J
rbelongs to the class J
rif and only if im α = dom β, it is easy to show that
Lemma 3.
1. L
BR
A=
( J
r, if A = B, 0, if A 6= B.
2. H
A,BH
C,D=
( H
A,D, if B = C, 0, if B 6= C.
Let U be a subset of the semigroup IM
n. We denote by U
i(respectively U
a) the set of all isotone (respectively antitone) transformations in the set U . An immediate but important property is that the product of two isotone transformations or two antitone transformations is an isotone, and the prod- uct of an isotone transformation with an antitone transformation, or vice versa, is an antitone one.
Proposition 2. J
r⊆ hJ
rai and J
r⊆ hJ
ri∪ {β
A,B}i, for all A, B ∈ Λ
r.
P roof. Let A, B ∈ Λ
r. Then for all C ∈ Λ
r, we have α
A,B= β
A,Cβ
C,B. Therefore, J
r⊆ hJ
rai.
From L
iAβ
A,B= L
aBand L
aBR
Bi= J
ra, we have J
r⊆ hJ
ri∪ {β
A,B}i.
Proposition 3. hJ
ri = I(n, r), for 0 ≤ r ≤ n − 1.
P roof. Clearly hJ
0i = I(n, 0). In [5], it was shown that J
r−1i⊆ J
riJ
riand J
r−1a⊆ J
r−1iJ
raJ
r−1ifor 1 ≤ r ≤ n − 1. Since I(n, r) = J
0∪ J
1∪ · · · ∪ J
r, we have hJ
ri = I(n, r).
From Proposition 2 and Proposition 3 we have
Corollary 1. Let 1 ≤ r ≤ n − 1. Then hJ
rai = hJ
ri∪ {β
A,B}i = I(n, r), for all A, B ∈ Λ
r.
Now we begin with the description of the maximal subsemigroups of the ideals of the semigroup IM
n.
Clearly, the ideal I(n, 1) of IM
ncoincides with the ideal I(n, 1) of IO
n. Thus the maximal subsemigroups of this ideal are characterized in Theorem 2 and there are exactly 2
n− 2 such semigroups.
Now we will consider the maximal subsemigroups of the ideals I(n, r) for r = 2, . . . , n − 1.
Lemma 4. Every maximal subsemigroup in I(n, r) contains the ideal I(n, r − 1).
The proof is similar as that in Lemma 2.
Theorem 3 Let 2 ≤ r ≤ n−1. Then a subsemigroup S of I(n, r) is maximal if and only if it belongs to one of the following three types:
(1) S
(1):= I(n, r − 1) ∪ J
ri;
(2) S
(N(2)1,N2):= S{H
α: α ∈ S
(N1,N2)}, f or (N
1, N
2) ∈ Dec(Λ
r);
(3) S
(N(3)1,N2):= I(n, r − 1) ∪ {α
A,B: A, B ∈ N
1or A, B ∈ N
2} ∪
∪ {β
A,B: A ∈ N
1, B ∈ N
2or A ∈ N
2, B ∈ N
1} f or (N
1, N
2) ∈ Dec(Λ
r).
P roof.
(1) It is obvious that S
(1)= I(n, r − 1) ∪ J
riis a semigroup, since I(n, r − 1) is an ideal and (J
ri)
2⊆ I
i(n, r) ⊆ I(n, r −1)∪J
ri. From Proposition 2, we have that J
r⊆ hJ
ri∪ {β
A,B}i for all β
A,B∈ J
ra. Since I(n, r) \ S
(1)= J
ra, we obtain hS
(1)∪ {β
A,B}i = I(n, r) for all β
A,B∈ J
ra. Therefore, S
(1)is maximal in I(n, r).
(2) Let S = S
(2)(N1,N2)for some (N
1, N
2) ∈ Dec(Λ
r). Then S = I(n, r − 1) ∪ {H
A,B: A ∈ N
1or B ∈ N
2}.
From Lemma 3 it follows that S is a semigroup. Really, let H
A,B, H
C,D⊆ S, i.e. A, C ∈ N
1or B, D ∈ N
2or A ∈ N
1, D ∈ N
2. Then we have H
A,BH
C,D= H
A,D⊆ S for B = C and H
A,BH
C,D⊆ I(n, r − 1) ⊆ S for B 6= C.
Now we will show that S is maximal. Let H
C,D= {α
C,D, β
C,D} ⊆ I(n, r) \ S, i.e. C / ∈ N
1and D / ∈ N
2. Then D ∈ N
1, since N
1∪ N
2= Λ
rand so H
D,P∈ S for all P ∈ Λ
rand thus R
D= S
P ∈Λr
H
D,P⊆ S.
Moreover, we have
H
C,P= H
C,DH
D,P, for P ∈ Λ
r, by Lemma 3. Thus we obtain the R-class R
C= S
P ∈Λr
H
C,P⊆ hS ∪ H
C,Di. Moreover, C ∈ N
2and so L
C= S
P ∈Λr
H
P,C⊆ S. Using Lemma 3, we have L
CR
C= J
r⊆ hS ∪ H
C,Di. Since α
C,D= β
C,Dβ
D,Dand β
C,D= α
C,Dβ
D,D, where β
D,D∈ R
D⊆ S, we obtain that hS∪{α
C,D}i = I(n, r) and hS ∪ {β
C,D}i = I(n, r). Therefore, S is a maximal subsemi- group of the ideal I(n, r).
(3) Let S = S
(N(3)1,N2)for some (N
1, N
2) ∈ Dec(Λ
r). From Lemma 3, it follows that S is a semigroup. We will show that S is maximal. Let
V := I(n, r) \ S = {β
A,B: A, B ∈ N
1or A, B ∈ N
2} ∪
∪ {α
A,B: A ∈ N
1, B ∈ N
2or A ∈ N
2, B ∈ N
1}
and let γ ∈ V . Then for the transformation γ we have four possibilities:
Let γ ∈ {β
A,B: A, B ∈ N
1}. Then α
C,A∈ S (since A ∈ N
1) and so α
C,Aβ
A,B= β
C,B∈ hS ∪ {γ}i for all C ∈ N
1. Also, we have β
C,A∈ S and thus β
C,Aβ
A,B= α
C,B∈ hS ∪ {γ}i for all C ∈ N
2. Since α
C,B∈ S for all C ∈ N
1and β
C,B∈ S for all C ∈ N
2, we obtain L
B= S
C∈Λr
H
C,B⊆ hS ∪ {γ}i. Further, β
B,B∈ L
Band β
B,Bβ
B,D= α
B,Dfor all D ∈ N
2as well as β
B,Bα
B,D= β
B,Dfor all D ∈ N
1. Thus since α
B,D∈ S for all D ∈ N
1and β
B,D∈ S for all D ∈ N
2, we obtain R
B= S
D∈Λr
H
B,D⊆ hS ∪ {γ}i.
From Lemma 3, we have L
BR
B= J
rand therefore hS ∪ {γ}i = I(n, r).
– For γ ∈ {β
A,B: A, B ∈ N
2}, the proof is similar.
– Let γ ∈ {α
A,B: A ∈ N
1, B ∈ N
2}. Then α
C,A∈ S (since A ∈ N
1) and so α
C,Aα
A,B= α
C,B∈ hS ∪ {γ}i for all C ∈ N
1. Also, we have β
C,A∈ S and thus β
C,Aα
A,B= β
C,B∈ hS ∪ {γ}i for all C ∈ N
2. Since α
C,B∈ S for all C ∈ N
2and β
C,B∈ S for all C ∈ N
1, we obtain L
B= S
C∈Λr
H
C,B⊆ hS ∪ {γ}i. Further, β
B,B∈ L
Band β
B,Bα
B,D= β
B,Dfor all D ∈ N
2as well as β
B,Bβ
B,D= α
B,Dfor all D ∈ N
1. Thus since α
B,D∈ S for all D ∈ N
2and β
B,D∈ S for all D ∈ N
1, we obtain R
B= S
l∈Λr
H
B,D⊆ hS ∪ {γ}i. From Lemma 3, we have L
BR
B= J
rand therefore hS ∪ {γ}i = I(n, r).
– For γ ∈ {α
A,B: A ∈ N
2, B ∈ N
1}, the proof is similar.
Altogether, this shows that S is maximal.
For the converse part let S be a maximal subsemigroup of the ideal I(n, r). From Lemma 4, we have that I(n, r − 1) ⊆ S. Then S = I(n, r − 1)
∪ T , where T ⊆ J
r. We consider two cases for the set T . 1. Let T = J
ri. Then S = I(n, r − 1) ∪ J
ri= S
(1).
2. Let now T 6= J
ri. Assume that J
ri⊆ T . Then T = J
ri∪ T
0where
∅ 6= T
0⊆ J
ra. From Corollary 1, we have S = I(n, r), a contradiction.
Thus J
ri* T . We also have that J
ra* T since hJ
rai = I(n, r).
Admit that H
A,B⊆ S or H
A,B∩ S = ∅, for all A, B ∈ Λ
r. Assume that
S
i= S ∩ I
i(n, r) is not a maximal subsemigroup of I
i(n, r). Then there is an
isotone transformation α
A,B∈ I(n, r) \ S such that hS
i∪ {α
A,B}i is a proper
subset of I
i(n, r). Therefore, there exists an α
C,D∈ I(n, r) \ S such that
α
C,D∈ hS /
i∪ {α
A,B}i. But hS ∪ {α
A,B}i = I(n, r) since S is maximal and
α
C,D= β
C,Aα
A,Bβ
B,D. Moreover, β
C,A= β
C,Aα
A,Bα
B,A= α
C,Aα
A,Bβ
B,Aand β
B,D= β
B,Aα
A,Bα
B,D= α
B,Aα
A,Bβ
B,D. This shows that we need
β
C,Aor α
C,Aand β
B,Dor α
B,Dto generate β
C,Aand β
B,D, respectively, with elements of S ∪ {α
A,B}. This implies that β
C,A, α
C,A, β
B,D, α
B,D∈ S, since we assume that H
A,B⊆ S or H
A,B∩ S = ∅, for all A, B ∈ Λ
r. Hence α
C,D= α
C,Aα
A,Bα
B,D∈ hS
i∪ {α
A,B}i, a contradiction. Therefore, we obtain that S
iis maximal in I
i(n, r). Since all maximal subsemigroups of the ideal I
i(n, r) are of type S
(N1,N2)we have S = ∪{H
α: α ∈ S
i} = S
(N(2)1,N2), for some (N
1, N
2) ∈ Dec(Λ
r).
Now, admit that |H
A,B∩ S| = 1, for some A, B ∈ Λ
r. Suppose that α
A,B∈ S and β /
A,B∈ S. Then from α
A,B= β
A,Bβ
B,Band α
A,B= β
A,Aβ
A,B, it follows that β
A,A, β
B,B∈ S. Moreover, from β /
A,Bα
B,A= β
A,A∈ S, we get α /
B,A∈ S. Assume that β /
B,A∈ S. Then β /
B,A∈ hS ∪ {α
B,A}i, because of the maximality of S, and since β
B,A= β
B,Bα
B,Aα
A,A= α
B,Bα
B,Aβ
A,A, we obtain β
A,A∈ S or β
B,B∈ S, a contradiction, and thus β
B,A∈ S.
Further, let P, Q ∈ Λ
r. Suppose that α
P,Q∈ S. Then from α /
P,Q= α
P,Aβ
A,Bβ
B,Q, it follows that if α
P,A∈ S then β
B,Q∈ S and vice versa. / Also from α
P,Q= β
P,Aβ
A,Bα
B,Q, it follows that if β
P,A∈ S then α
B,Q∈ S / and vice versa. Moreover, α
P,Q∈ hS ∪ {α
A,B}i since S is maximal. Hence α
P,Q= α
P,Aα
A,Bα
B,Q= β
P,Aα
A,Bβ
B,Q. Therefore, we have α
P,A, α
B,Q∈ S and β
P,A, β
B,Q∈ S or vice versa. /
Assume that β
P,Q∈ S. Then β /
P,Q∈ hS ∪ {α
A,B}i and so β
P,Q= α
P,Aα
A,Bβ
B,Q= β
P,Aα
A,Bα
B,Q. But we obtain already that if α
P,A, α
B,Q∈ S then β
P,A, β
B,Q∈ S or vice versa. Therefore, β /
P,Q∈ hS ∪ {α /
A,B}i. This contradicts the maximality of S and thus β
P,Q∈ S.
Further, from α
P,Q= β
P,Qβ
Q,Qand α
P,Q= β
P,Pβ
P,Q, it follows that β
P,P, β
Q,Q∈ S. Moreover, from β /
P,Qα
Q,P= β
P,P∈ S, we get α /
Q,P∈ / S. Assume that β
Q,P∈ S. Then β /
Q,P∈ hS ∪ {α
Q,P}i, because of the maximality of S, and since β
Q,P= β
Q,Qα
Q,Pα
P,P= α
Q,Qα
Q,Pβ
P,P, we obtain β
P,P∈ S or β
Q,Q∈ S, a contradiction, and thus β
Q,P∈ S.
Analogously, if β
P,Q∈ S we have that β /
Q,P∈ S and α /
P,Q, α
Q,P∈ S.
Suppose that α
P,Q∈ S for some P, Q ∈ Λ
r. Then β
P,Q∈ S. Otherwise, / from α
A,B= α
A,Pβ
P,Qβ
Q,B∈ S it follows /
i) α
A,P∈ S and β /
Q,B∈ S, i.e. β
A,P∈ S and β
Q,B∈ S;
ii) α
A,P∈ S and β
Q,B∈ S, i.e. α /
A,P∈ S and α
Q,B∈ S;
iii) α
A,P∈ S and β /
Q,B∈ S, i.e. β /
A,P∈ S and α
Q,B∈ S.
Then α
A,B= β
A,Pα
P,Qβ
Q,B= α
A,Pα
P,Qα
Q,B= β
A,Pβ
P,Qα
Q,B, which con- tradicts that α
A,B∈ S. /
The proof when α
A,B∈ S and β
A,B∈ S is similar. / Finally, we obtain that
(1) α
P,Q∈ S ⇐⇒ β
P,Q∈ S /
for P, Q ∈ Λ
r.
Let ρ
r:= {(P, Q) : α
P,Q∈ S}. Obviously, ρ
ris an equivalence relation on Λ
rwith Λ
r/ρ
r= {N
1, N
2, . . . , N
m} (m ≥ 2). Indeed, ρ
ris reflexive since E
r⊆ S, symmetric because of the previous considerations and transitive since α
P,Qα
Q,R= α
P,R∈ S for α
P,Q, α
Q,R∈ S. Moreover, m ≥ 2 becomes clear by J
ri* T . Assume that the decomposition contains more than two elements, i.e. m > 2. Then there are N
1, N
2, N
3in our decomposition such that A ∈ N
1, B ∈ N
2and C ∈ N
3. Thus α
A,B= β
A,Cβ
C,B∈ S, a con- tradiction. Therefore, Λ
r/ρ
r= {N
1, N
2} and S = S
(N(3)1,N2), because of (1).
There are exactly 2(
nr) − 2 maximal subsemigroups of the ideal I
i(n, r) and exactly 2(
n
r
) −2 maximal subsemigroups of type (3). Taking I(n, r − 1) ∪ J
riinto account, we get 2(
nr)
+1− 3 maximal subsemigroups of the ideal I(n, r), for r = 2, . . . , n − 1.
Finally, we characterize the maximal subsemigroups of the ideal I(n, n) = IM
n.
For A ∈ Λ
n−1we put A := {n + 1 − i : i ∈ A} and for N ⊆ P(X
n) we set N := {A : A ∈ N }. Then we have
(2)
β
A,Aα
A,B= β
nα
A,B= β
A,B,
β
A,Aβ
A,B= β
nβ
A,B= α
A,B,
α
B,Aβ
A,A= α
B,Aβ
n= β
B,A,
β
B,Aβ
A,A= β
B,Aβ
n= α
B,A.
Theorem 4. A subsemigroup S of IM
nis maximal if and only if it belongs to one of the following three types:
(1) T := I(n, n − 1) ∪ {α
n};
(2) T
(N1,N2):= J
n∪ {H
α: α ∈ S
(N1,N2)}, f or (N
1, N
2) ∈ Dec(Λ
n−1) with N
1= N
1and N
2= N
2;
(3) T
(N,N ):= J
n∪ I(n, n − 2) ∪ {α
A,B: A, B ∈ N or A, B ∈ N }
∪ {β
A,B: A ∈ N, B ∈ N or A ∈ N , B ∈ N } f or (N, N ) ∈ Dec(Λ
n−1).
P roof. It is clear that T is a maximal subsemigroup of IM
n. Further, we put
Inv := {β
A,A: A ∈ Λ
n−1}.
Let (N
1, N
2) ∈ Dec(Λ
n−1) be a decomposition with the required proper- ties. Since Inv ⊆ T
(N1,N2)and by (2) it is easy to verify that T
(N1,N2)is a subsemigroup of IM
n. Since T
(N1,N2)\ J
nis a maximal subsemigroup of I(n, n − 1) by Theorem 3 and J
n⊆ T
(N1,N2), it follows that T
(N1,N2)is a maximal subsemigroup of IM
n. Analogously, one can show that T
(N,N )is a maximal subsemigroup of IM
n.
For the converse part, let S be maximal in IM
n. Admit that J
n* S.
Then it is easy to see that S = T . Now suppose that J
n⊆ S. Assume that Inv * S. Then there is an A ∈ Λ
n−1with β
A,A∈ S. Since S is maximal, we / have IM
n= hS ∪{β
A,A}i = S ∪{β
A,A} by (2). Thus S = IM
n\{β
A,A}. But β
A,A= α
A,Bβ
B,Afor some B ∈ Λ
n−1with B 6= A. Since α
A,B, β
B,A∈ S, we have S = IM
n, a contradiction. Hence Inv ⊆ S. Let S
n−1:= S ∩I(n, n−1).
Assume that S
n−1is not a maximal subsemigroup of I(n, n − 1). Clearly, S
n−16= I(n, n − 1). Let γ ∈ I(n, n − 1) \ S
n−1. Then for all δ ∈ I(n, n − 1), we have δ ∈ hS ∪ {γ}i = hS
n−1∪ {γ}i ∪ J
nby (2) and since Inv ⊆ S. This shows that δ ∈ hS
n−1∪ {γ}i and thus hS
n−1∪ {γ}i = I(n, n − 1). Conse- quently, S
n−1is a maximal subsemigroup of I(n, n−1). Using Theorem 3 we choose all decompositions (N
1, N
2) ∈ Dec(Λ
n−1) such that Inv ⊆ S
(N(2)1,N2)
and Inv ⊆ S
(3)(N1,N2)
, respectively. In this way we obtain the semigroups
T
(N1,N2)and T
(N,N ).
It is straightforward to calculate that there are exactly 2
n+12− 1 maximal subsemigroups of IM
nif n is odd and exactly
32
(2
n2