Journal of Algebra 278 (2004) 342–359

www.elsevier.com/locate/jalgebra

Combinatorial results for semigroups of

order-preserving partial transformations
A. Laradji and A. Umar ∗
Department of Mathematical Sciences, King Fahd University of Petroleum and Minerals,
Dhahran 31261, Saudi Arabia
Received 5 August 2003
Available online 24 January 2004
Communicated by T.E. Hall

Abstract
Let POn be the semigroup of all order-preserving partial transformations of a finite chain. It is
shown that |POn | = cn satisfies the recurrence


(2n − 1)(n + 1)cn+1 = 4 3n2 − 1 cn − (2n + 1)(n − 1)cn−1

with initial conditions c0 = 1, c1 = 2. It is also shown that |E(POn )| = en satisfies the recurrence
en+1 = 5(en − en−1 ) + 1 with initial conditions e0 = 1, e1 = 2. Moreover, the cardinalities of the
Green’s relations L, R and J have been computed.
 2003 Elsevier Inc. All rights reserved.

1. Introduction
Consider a finite chain, say Xn = {1, 2, . . . , n} under the natural ordering and let Tn
and Pn be the full and partial transformation semigroups on Xn , respectively. We shall
call a partial transformation α : Dom α ⊆ Xn → Xn (order)-decreasing if xα
x for all
x in Dom α, and α is order-preserving if x
y implies xα
yα for x, y in Dom α.
Combinatorial properties of Cn , the semigroup of all decreasing and order-preserving full
transformations on Xn have been investigated by Higgins [8] and recently by Laradji
and Umar [11]. These papers motivated the study of combinatorial properties of PC n ,
the semigroup of all decreasing and order-preserving partial transformations on Xn by

* Corresponding author.
E-mail addresses: [email protected] (A. Laradji), [email protected] (A. Umar).

0021-8693/$ – see front matter  2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.jalgebra.2003.10.023
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 343

Laradji and Umar [12], where it is shown that |PC n | is the double Schröder number and
|E(PC n )| = (3n + 1)/2. This paper investigates combinatorial properties of POn , the
semigroup (necessarily larger than PC n ) of all order-preserving partial transformations
on Xn , by analogy with [12].
Unlike PC n , the semigroup POn has been studied by Gomes and Howie [6] and
Garba [4,5], mainly from algebraic point of view. After this introductory and preliminary
section, we obtain in Section 2 a recurrence satisfied by |POn | (similar to the one for
PC n ). In Section 3, we compute the total number of idempotents of POn via some
natural equivalences and a linear recurrence relation. Finally, in Section 4 we compute the
cardinalities of the Green’s relations L-, R- and J -classes in POn . For standard concepts
in semigroup theory we refer the reader to [10] or [7].
We now recall some basic definitions from [12] that we shall need in the coming
sections.

Definition 1.1. Consider Xn = {1, 2, . . . , n} and let α : Xn → Xn be a partial transforma-

tion. We shall denote by Dom α, the domain of α and by Im α the image set of α. The width
of α is |Dom α|, the height of α is |Im α| and the waist of α is max(Im α).

Let Pn , denote the semigroup of all partial transformations of Xn under the usual
composition. Formally, we define POn as
 
POn = α ∈ Pn : (∀x, y ∈ Dom α) x
y ⇒ xα
yα . (1.1)

We also record these two results that will be needed in Section 2. The first (Lemma 1.1)
known as the Vandermonde’s convolution identity is in the words of Riordan [14, p. 8]
perhaps the most widely used combinatorial identity, while the second (Lemma 1.2) can
be obtained by repeated application of the Pascal’s triangular identity.

Lemma 1.2 [14, (3a), p. 8]. For all natural numbers k, m, n and p we have

n 
    
n p n+p
= .
m−k k m
k=0

Lemma 1.3. For all natural numbers k, n and r we have

n 
     
k+r −2 n+r −1 n+r −1
= = .
k−1 n−1 r
k=0

2. The order of PO n

Gomes and Howie [6] were the first to study POn (excluding the identity map) and
among other things they computed the order of POn , which we now record.
344 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359

Theorem 2.1 [6, Theorem 3.1]. Let POn be as defined in (1.1). Then |POn | is the
coefficient of x n in the series expansion of (1 + x)n (1 − x)−n . Equivalently,

n  
 
n n+r −1
|POn | = .
r r
r=0

However, from a computational point of view this result is not satisfactory if one were
to compute higher orders of POn . Recently, the authors in [12] computed the order of PC n
as rn , the double Schröder number given by

n    n   
1  n+1 n+r 1 n n+r
rn = = (2.1)
n+1 n−r r n r n−r
r=0 r=0

which also satisfies the recurrence

(n + 2)rn+1 = 3(2n + 1)rn − (n − 1)rn−1 (2.2)

for n  1, with initial conditions r0 = 1, r1 = 2. See [13] for a detailed account on

Schröder numbers. Moreover, in the process of discovering |PC n | (in [12]) the authors
also found four triangular arrays of numbers which are not in Sloane’s encyclopedia of
integer sequences [15], and it is this relative success that motivates the search for similar
results for POn . As in [12], we begin by defining f (n, r, k) as
 
f (n, r, k) =  α ∈ POn : |Dom α| = r ∧ max(Im α) = k . (2.3)

Then clearly we have

1 (k = 0), 1 (r = 0),
f (n, 0, k) = f (n, r, 0) =
0 (k > 0), 0 (r > 0),

and perhaps less clearly, we have

 
n
f (n, r, 1) = (1
r
n).
r

This holds because the number of maps α in POn of width r with Im α = {1}, is simply
the number of possible domains, that is, the number of subsets of Xn of size r. In general,
we have

Proposition 2.2. Let f (n, r, k) be as defined in (2.3). Then for n  r, k > 0,

  
n k+r −2
f (n, r, k) = .
r k−1
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 345

Proof. First note that for all α in POn and y in Im α, yα −1 is convex modulo Dom α. That
is, to say, if y1 , y2 ∈ yα −1 and z ∈ Dom α is such that y1 < z < y2 then −1

n z ∈ yα as well.
Next note that we can choose the elements of Dom α (from Xn ) in r ways. However,
since |Im α| = s, where 1
s
r and max(Im α) = k, it follows
that we  can choose the
remaining s − 1 elements of Im α \ {k} from {1, 2, . . . , k − 1} in k−1
s−1 ways, which can

r−1 
now be tied to Dom α in s−1 ways, by inserting s − 1 symbols between the r − 1 spaces
in Dom α, to get convex (modulo Dom α) partitions. Thus, in all we have

 
r     
r−1   
n k−1 r −1 n k−1 r −1
f (n, r, k) = =
r s −1 s −1 r k−1−j j
s=1 j =1
  
n k+r −2
= (by Lemma 1.1). 2
r k−1

n 
2r−2 
Corollary 2.3. f (n, r, r) = r r−1 .

Corollary 2.4. Let On be the semigroup of all order-preserving full transformations of Xn .

Then
 
 
 α ∈ On : max(Im α) = k  = f (n, n, k) = n + k − 2 .
k−1

n
Lemma 2.5. Let G(n, 0) = 1, and for n  k > 0, let G(n, k) = r=0 f (n, r, k). Then
G(n, 1) = 2n − 1,
n  
 
n n+r −2
G(n, n) =
r n−1
r=0

and, for 2
k
n,

G(n, k) = 2G(n − 1, k) − G(n − 1, k − 1) + G(n, k − 1).

Proof. First observe that G(n, 1) = 2n − 1 holds because the number of maps α in POn
with Im(α) = {1}, is simply the number of possible domains, that is, the number of
nonempty subsets of Xn . The formula

for G(n, n) follows from Proposition 2.2. To prove
the recurrence we let a(k, r) = k+r−2
k−1 . Then it is clear that a(k, 0) = 0 and

a(k, r) = a(k − 1, r) + a(k, r − 1). (2.4)

Now, by Proposition 2.2

n  
 n−1 
 
n n−1
G(n, k) − G(n − 1, k) = a(k, r) − a(k, r)
r r
r=1 r=1
346 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359

n−1 
 
n−1
= a(k, r) + a(k, n)
r −1
r=1
n 
 
n−1
= a(k, r) (2.5)
r −1
r=1

and so
n 
 
n−1
G(n, k − 1) − G(n − 1, k − 1) = a(k − 1, r). (2.6)
r −1
r=1

From (2.5) and (2.6) we have

G(n, k) − G(n − 1, k) − G(n, k − 1) + G(n − 1, k − 1)

 n  
n−1  
= a(k, r) − a(k − 1, r)
r −1
r=1
n 
 
n−1

= a(k, r − 1) by (2.4)
r −1
r=1
n 
 
n−1

= a(k, r − 1) since a(k, 0) = 0
r −1
r=2
n−1 
 
n−1
= a(k, r) = G(n − 1, k).
r
r=1

Hence the result follows. 2

Corollary 2.6.

n  
  n−1 
  
n n+r −2 n n+s −1
G(n, n) = =
r n−1 s +1 n−1
r=1 s=0
n−1 
  
n n+s −1
= = nrn−1 .
n−s −1 s
s=0

n
Proposition 2.7. Let F (n, r) = k=1 f (n, r, k). Then
  
n n+r −1
F (n, r) = .
r n−1
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 347

Proof.


n n  
 
n k+r −2
F (n, r) = f (n, r, k) =
r k−1
k=1 k=1
 
n     
n k+r −2 n n+r −1
= = (by Lemma 1.2). 2
r k−1 r n−1
k=1

Corollary 2.8 [9, Theorem 2.1]. Let On be the semigroup of all order-preserving full
transformations of Xn . Then
 
2n − 1
|On | = F (n, n) = .
n−1

Remark 2.9. The triangular arrays of numbers f (n, r, r), G(n, k) and F (n, r) are not yet
listed in [15] and so we believe they are new. For selected values of these numbers see
Tables 1–3.

It is now clear that we have also proved the last part of Theorem 3.1, that is


n n  
 
n n+r −1
|POn | = F (n, r) = .
r n−1
r=0 r=0

Table 1
f (n, r, r)
k

n
0 1 2 3 4 5 6 7 f (n, r, r)
0 1 1
1 1 1 2
2 1 2 2 5
3 1 3 6 6 16
4 1 4 12 24 20 61
5 1 5 20 60 100 70 256
6 1 6 30 120 300 420 252 1129
7 1 7 42 210 700 1470 1764 924 5118

Table 2
G(n, k)
k

n
0 1 2 3 4 5 6 7 G(n, k)
0 1 1
1 1 1 2
2 1 3 4 8
3 1 7 12 18 38
4 1 15 32 56 88 192
5 1 31 80 160 280 450 1002
6 1 63 192 432 832 1452 2364 5336
7 1 127 448 1120 2352 4244 7700 12642 28814
348 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359

Table 3
F (n, r)
r

n
0 1 2 3 4 5 6 7 F (n, r)
0 1 1
1 1 1 2
2 1 4 3 8
3 1 9 18 10 38
4 1 16 60 80 35 192
5 1 25 150 350 350 126 1002
6 1 36 315 1120 1890 1512 462 5336
7 1 49 588 2940 7350 9702 6468 1716 28814

Before we get a recurrence (similar to that for |PC n | = rn in [12]) satisfied by

|POn | = cn , first we establish the following lemma linking the two cardinalities.

Lemma 2.10. For all n > 0, we have

2cn = (n + 1)rn − (n − 1)rn−1 .

Proof.

r.h.s. = (n + 1)rn − (n − 1)rn−1

n    n−1   
n+1 n n+r n−1 n+r −1

= − by (2.1)
n r n−1 r n−2
r=0 r=0
n−1
        
n+1 n n+r n−1 n+r −1 n+1 2n
= − +
n r n−1 r n−2 n n−1
r=0


n−1  
  (n + r − 1)! n+1 2n
= (n + 1)(n + r) − (n − 1)(n − r) +
r!(n − r)!(r + 1)! n n−1
r=0


n−1
2n(n + r − 1)! 2n(2n − 1)!  2n(n + r − 1)!
n
= + =
r!(n − r)!r! n!n! r!(n − r)!r!
r=0 r=0
n 
  
n n+r −1
=2 = 2cn = l.h.s. 2
r r
r=0

We now have

Proposition 2.11. Let POn be as defined in (1.1), and let cn = |POn |. Then c0 = 1, c1 = 2
and for all n > 0,


(2n − 1)(n + 1)cn+1 = 4 3n2 − 1 cn − (2n + 1)(n − 1)cn−1 .
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 349

Proof. From Lemma 2.10 and (2.2) successively we have

2cn+1 = (n + 2)rn+1 − nrn = 3(2n + 1)rn − (n − 1)rn−1 − nrn

= (5n + 3)rn − (n − 1)rn−1 . (2.7)

Eliminating rn from Lemma 2.10 and (2.7) gives

(n + 1)cn+1 − (5n + 3)cn = (2n + 1)(n − 1)rn−1 (2.8)

while eliminating rn−1 from Lemma 2.10 and (2.7) gives

cn+1 − cn = (2n + 1)rn

which in turn implies

cn − cn−1 = (2n − 1)rn−1 . (2.9)

Finally, eliminating rn−1 from (2.8) and (2.9) gives the required result. 2

3. The number of idempotents in PO n

As many ‘natural’ semigroups of transformations are idempotent-generated it is not sur-

prising that counting the number of idempotents in such semigroups has attracted the atten-
tion of Higgins [8], Howie [9], Tainiter [16] and Umar [17,18]. See also [2, Ex. 2.2.2(a)].
Gomes and Howie [6, Theorem 3.13] showed that POn is idempotent-generated, but did
not count all the idempotents in POn . To investigate this number we take a slightly differ-
ent approach (but essentially the same) from the previous section. First, we consider the
equivalence on E(POn ) given by the equality of widths and define
 
E(n, r) =  α ∈ POn : α 2 = α ∧ |Dom α| = r . (3.1)

Then clearly we have

E(n, 0) = 1 and E(n, 1) = n.

Moreover, we have from Howie [9] that

E(n, n) = f2n

where f2n is the alternate Fibonacci number. In general, we have

Lemma 3.1. E(n, r) = n−r E(n − 1, r)

n
(n > r  0).
350 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359

Proof. Let g(r, s) be the number of all idempotent order-preserving full transformations
with domain {x1 , x2 , . . . , xr } ⊆ Xn and of height s. To count all idempotents ε in POn of

r, we first note that we can choose the domain of ε (from Xn ), say {x1 , x2 , . . . , xr }
width
in nr ways. Next we choose the elements of Im ε ⊆ {x1 , x2 , . . . , xs } where s = | Im ε|


satisfies 1
s
r. Now since Im ε can be chosen in rs ways, it follows that
 
r  
n r
g(r, s) = E(n, r) (3.2)
r s
s=1

from which we deduce

  r  
n−1  r
g(r, s) = E(n − 1, r). (3.3)
r s
s=1

From (3.2) and (3.3), it follows that

   
n−1 n
E(n, r) = E(n − 1, r)
r r

which in turn gives the required result. 2

Consequently from Lemma 3.1 we deduce that

n
Corollary 3.2. E(n, r) = r E(r, r).

Next, we consider the equivalence in E(POn ) given by equality of waists and define
 
H (n, k) =  α ∈ POn : α 2 = α ∧ max(Im α) = k . (3.4)

Then clearly

H (n, 0) = 1 and H (n, 1) = 2n−1 .

In general, we have

Lemma 3.3. For 0 < k < n, H (n, k) = 2n−k H (k, k).

Proof. Let ε be an idempotent in POn satisfying max(Im ε) = k. Then by the order-

preserving property, for all x in {k, k + 1, . . . , n} we have xε = k, if x ∈ Dom ε. Thus to
compute H (n, k) we first look at the set W of all idempotents η on {1, 2, . . . , k} such that
max(Im η) = k. Then |W | = H (k, k). Now multiply this number by 2n−k to get H (n, k),
where 2n−k is the total number of degrees of freedom for members of {k + 1, k + 2, . . . , n},
that is, for each x in {k + 1, k + 2, . . . , n} either x ∈ Dom ε (in which case xε = k) or
x∈/ Dom ε. 2
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 351

Now since H (n, k) depends on H (k, k), we focus our attention to finding an expression
for H (n, n). In fact we have

n−2
Proposition 3.4. H (n, n) = H (n − 1, n − 1) + 2n−1 + t =1 (n − t + 1)2n−t −2H (t, t).

Proof. Since max(Im ε) = n, then n ∈ Dom ε and nε = n, by idempotency. Now we

consider cases:

Case 1. If
min(nε−1 ) = {n}, then from the remaining {1, 2, . . . , n − 1} elements we can
construct n−1 −1
t =0 H (n − 1, t) idempotents to each of which we adjoin (nε )ε = nε = n.

Case 2. If min(nε −1 ) = {n − 1}, then from the remaining {1, 2, . . . , n − 2} elements

n−2
we can construct t =0 H (n − 2, t) idempotents to each of which we adjoin (nε−1 )ε =
{n − 1, n}ε = n.

Now, in general, if min(nε−1 ) = {n − m + 1}, where 2
m
n, it is clear that {n −

m + 1, n} ⊆ nε−1 ⊆ {n − m + 1, . . . , n}. However, for each of the m − 2 middle elements
{n−m+2, . . . , n−1} there are two degrees of freedom: either x (in {n−m+2, . . . , n−1})
belongs to Dom ε (in which case xε = n) or it does not belong to Dom ε. Thus there
are 2m−2 degrees of freedom for these m − 2 middle
elements. Next, considering the
remaining elements {1, 2, . . . , n − m} we can construct n−m t =0 H (n − m, t) idempotents,
to each of which we adjoin (nε −1 )ε ⊆ {n − m + 1, . . . , n}ε = n, thus giving rise to

2m−2 n−mt =0 H (n − m, t) idempotents in all. Finally, adding all the sums from all the cases
we get


n−1 
n 
n−m
H (n, n) = H (n − 1, t) + 2 m−2
H (n − m, t)
t =0 m=2 t =0


n−2 
n
= H (n − 1, n − 1) + H (n − 1, t) + H (n − 1, 0) + 2m−2 H (n − m, 0)
t =1 m=2

 
n−1 n−m
+ 2m−2 H (n − m, t)
m=2 t =1


n−2 
n−2 
n−1
= H (n − 1, n − 1) + 2n−1 + H (n − 1, t) + 2m−2 H (n − m, t)
t =1 t =1 m=t +1


n−2 
n−2 
n−1
= H (n − 1, n − 1) + 2n−1 + 2n−t −1 H (t, t) + 2n−t −2 H (t, t)
t =1 t =1 m=t +1


n−2
= H (n − 1, n − 1) + 2n−1 + 2n−t −2 · 2H (t, t)
t =1
352 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359


n−2
+ (n − t − 1)2n−t −2H (t, t)
t =1


n−2
= H (n − 1, n − 1) + 2n−1 + (n − t + 1)2n−t −2 H (t, t),
t =1

using Lemma 3.3 along the way. 2

However, a simple linear recurrence satisfied by H (n, n) = bn is given by the following

lemma.

Lemma 3.5. Let H (n, n) = bn . Then b1 = 1, b2 = 3 and

bn+1 = 5(bn − bn−1 ).

Proof. From Proposition 3.4, we have


n−2
bn = bn−1 + 2n−1 + (n − t + 1)2n−t −2 bt ,
t =1

so that


n−1
 
bn+1 = bn + 2n + (n − t + 1) + 1 2n−t −1 bt
t =1


n−2
 
= bn + 2n + (n − t + 1) + 1 2n−t −1 bt + 3bn−1
t =1
 

n−2 
n−2
n−t −2
= bn + 2 2 n−1
+ (n − t + 1)2 bt + bn−1 + 2n−t −1 bt + bn−1
t =−1 t =1


n−2
= 3bn + bn−1 + 2n−t −1 bt . (3.5)
t =1

This in turn implies


n−3
bn = 3bn−1 + bn−2 + 2n−t −2 bt . (3.6)
t =1

However, (3.5) may be written as

 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 353


n−3
bn+1 = 3bn + bn−1 + 2 2n−t −2 bt + 2bn−2
t =1


= 3bn + bn−1 + 2(bn − 3bn−1 − bn−2 ) + 2bn−2 by (3.6)
= 5bn − 5bn−1

as required. 2

Remark 3.6. The triangular arrays of numbers E(n, r) and H (n, k) are not yet in [15]. For
selected values of these numbers see Tables 4 and 5.

Now to obtain a formula for the total number of idempotents in POn we observe that

  
n
E(POn ) = en = H (n, k). (3.7)
k=0

Then by Lemma 3.3 and (3.7), we have


n 
n 
n−2
en = 1 + 2n−k H (k, k) = 1 + 2n−k bk = 1 + 2 2n−k−1 bk + 2bn−1 + bn
k=1 k=1 k=1

Table 4
E(n, r)
r

n
0 1 2 3 4 5 6 7 E(n, r)
0 1 1
1 1 1 2
2 1 2 3 6
3 1 3 9 8 21
4 1 4 18 32 21 76
5 1 5 30 80 105 55 276
6 1 6 45 160 315 330 144 1001
7 1 7 63 280 735 1155 1008 377 3626

Table 5
H (n, k)
k

n
0 1 2 3 4 5 6 7 H (n, k)
0 1 1
1 1 1 2
2 1 2 3 6
3 1 4 6 10 21
4 1 8 12 20 35 76
5 1 16 24 40 70 125 276
6 1 32 48 80 140 250 450 1001
7 1 64 96 160 280 500 900 1625 3626
354 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359


= 1 + 2(bn+1 − 3bn − bn−1 ) + 2bn−1 + bn by (3.5)
= 1 + 2bn+1 − 5bn , (3.8)

so that

en+1 = 1 + 2bn+2 − 5bn+1 = 1 + 2(5bn+1 − 5bn ) − 5bn+1 (by Lemma 3.5)

= 1 + 5bn+1 − 10bn . (3.9)

From (3.8) and (3.9) we deduce

en+1 − en = 3bn+1 − 5bn . (3.10)

But by (3.9) we have

en+2 = 1 + 5bn+2 − 10bn+1 = 1 + 5(5bn+1 − 5bn ) − 10bn+1

= 1 + 15bn+1 − 25bn = 1 + 5(3bn+1 − 5bn )


= 1 + 5(en+1 − en ) by (3.10) .

Thus we have shown that

Lemma 3.7. For all n > 0, en+1 = 1 + 5(en − en−1 ) with initial conditions, e0 = 1, e1 = 2.

By the standard method of solving linear recurrence relations (see [1]) we deduce

Theorem 3.8. Let en be as defined in (3.7). Then

 √  √  
√ n−1 5+1 n 5−1 n
en = ( 5 ) − + 1.
2 2

Remark 3.9. The sequence {bn } (n  1) has been recorded (March 2003) as [15, A081567]
but {en } is not yet listed in [15]. For selected values see Table 5.

The following curious result is worth recording.

Lemma 3.10. en ≡ 1 (mod 5) (n  2).

Alternatively, we may get the formula for en by using E(n, r), since


n 
n
en = E(n, r) = 1 + E(n, r)
r=0 r=1
n  
 n
=1+ E(r, r) (by Lemma 3.2)
r
r=1
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 355

n  
 n
=1+ f2r
r
r=1

where f2r = ar is the alternate Fibonacci number and it satisfies the recurrence

ar = 3ar−1 − ar−2 ,

from which we get

1

ar = √ p r − q r
5
with
√ √
3+ 5 3− 5
p= and q = .
2 2
Hence
 n   n  

1  n  n 1  
en = 1 + √ p −
r
q = 1 + √ (1 + p)n − (1 + q)n
r
5 r=1 r r=1
r 5
 √   √  
1 5+ 5 n 5− 5 n
=1+ √ −
5 2 2
 √ n  √  
√ 5+1 5−1 n
= 1 + ( 5)n−1 − .
2 2

4. The number of L-, R- and J -classes

It is clear that POn is a regular subsemigroup of Pn [6]. Hence by [10, Proposi-

tion 2.4.2] and [3, Section 2] we have

(α, β) ∈ L if and only if Im α = Im β, (4.1)

−1 −1
(α, β) ∈ R if and only if α◦α =β ◦β . (4.2)

Moreover, it can be easily shown that

(α, β) ∈ J if and only if |Im α| = |Im β|. (4.3)

Now let α in POn be such that | Im α| = s, then

since
 POn is aperiodic [6], it follows that
|Hα | = 1, and by (4.1) we deduce that |Lα | = ns . However, |Rα | is less clear and the next
lemma provides a formula.

n
n 
r−1 
Lemma 4.1. Let α in POn be such that |Im α| = s. Then |Rα | = r=s r s−1 .
356 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359


Proof. First observe that we can choose the r elements of Dom α in nr ways, where
s

r
n and that we can partition Dom α into s ‘convex’ (modulo Dom α) subsets in
r−1
s−1 ways. Thus multiplying these two binomial coefficients and taking the sum from
r = s to r = n yields the required result. 2

Next we obtain a linear recurrence satisfied by |Rα |.



Lemma 4.2. Let e(n, s) = nr=s nr r−1s−1 . Then e(n, 0) = 1 = e(n, n) and e(n, s) =
2e(n − 1, s) + e(n − 1, s − 1).

Proof.

2e(n − 1, s) + e(n − 1, s − 1) − e(n, s)

n−1 
   n−1 
  
n−1 r −1 n−1 r −1
= 2 +
r s−1 r s −2
r=s r=s−1
n−1 
    
n r −1 n−1
− −
r s −1 s−1
r=s
n−1 
        
n−1 r −1 n−1 r −1 n r −1
= 2 + −
r s−1 r s−2 r s −1
r=s
n−1   
 
1 n r −1  
= (n − 2r)(r − s + 1) + (n − r)(s − 1)
n(s − 1) r=s r s − 2
n−1   
1  n−1 r −1
= (n − 2r + s − 1)
s − 1 r=s r − 1 s−2
m   
1  m r −1  
= (m − r) − (r − s) (m = n − 1)
s − 1 r=s r − 1 s−2

1  m
m!  
= (m − r) − (r − s)
(s − 1)! r=s (m − r + 1)!(r − s + 1)!
m 
m!  m−r +1 r −s +1
= −
(s − 1)! r=s (m − r + 1)!(r − s + 1)! (m − r + 1)!(r − s + 1)!
m 
m!  1 1
= −
(s − 1)! r=s (m − r)!(r − s + 1)! (m − r + 1)!(r − s)!
 
m! 
m
1
= (ar − ar−1 ) ar =
(s − 1)! r=s (m − r)!(r − s + 1)!
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 357

m!
= (am − as−1) = 0
(s − 1)!

as required. 2

Two further recurrences satisfied by e(n, s) are given by the next two lemmas whose
proofs we omit because they are easy.

n−1 
Lemma 4.3. e(n − 1, s) + e(n − 1, s − 1) = 2n−s s−1 .

n−1 
Lemma 4.4. e(n, s) + e(n − 1, s − 1) = 2n−s+1 s−1 .

Now it follows from (4.3) that

 
n
J (n, s) = |Jα | = e(n, s). (4.4)
s

However, a recurrence satisfied by J (n, s) is given by the following lemma:

Lemma 4.5. J (n, 0) = 1 = J (n, n) and for n > s > 0

     
n−1 2(n − s + 1) n n
J (n, s) = J (n − 1, s) + J (n − 1, s − 1).
s −1 (n − s) s −1 s

Proof.
 
n−1
l.h.s. = J (n, s)
s −1
  
n−1 n

= e(n, s) by (4.4)
s −1 s
  
n−1 n  
= 2e(n − 1, s) + e(n − 1, s − 1) (by Lemma 4.2)
s −1 s
    
2n! n−1 n n−1
= e(n − 1, s) + e(n − 1, s − 1)
(n − s)!(s − 1)!(n − s) s s s −1
   
2(n − s + 1) n n
= J (n − 1, s) + J (n − 1, s − 1)
(n − s) s −1 s
= r.h.s. 2

Remark 4.6. The triangular array of numbers e(n, s) and J (n, s) are not yet listed in [15].
For selected values of these numbers see Tables 6 and 7.
358 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359

Table 6
e(n, s)
s

n
0 1 2 3 4 5 6 7 e(n, s)
0 1 1
1 1 1 2
2 1 3 1 5
3 1 7 5 1 14
4 1 15 17 7 1 41
5 1 31 49 31 9 1 122
6 1 63 129 111 49 11 1 365
7 1 127 321 351 209 71 13 1 1094

Table 7
J (n, s)
s

n
0 1 2 3 4 5 6 7 J (n, s)
0 1 1
1 1 1 2
2 1 6 1 8
3 1 21 15 1 38
4 1 60 102 28 1 192
5 1 155 490 310 45 1 1002
6 1 378 1935 2220 735 66 1 5336
7 1 889 6741 12285 7315 1491 91 1 28814

We conclude the section by observing that


n n  
 n
|POn | = cn = J (n, s) = e(n, s).
s
s=0 s=0

Acknowledgments

We gratefully acknowledge support from the King Fahd University of Petroleum and
Minerals. We also thank the referee for providing useful suggestions.

References

[1] I. Anderson, A First Course in Combinatorial Mathematics, Oxford University Press, 1974.
[2] A.H. Clifford, G.B. Preston, The Algebraic Theory of Semigroups, vol. 1, in: Math. Surveys, vol. 7, Amer.
Math. Soc., Providence, RI, 1961.
[3] G.U. Garba, Idempotents in partial transformation semigroups, Proc. Roy. Soc. Edinburgh 116 (1990) 359–
366.
[4] G.U. Garba, On the idempotent ranks of certain semigroups of order-preserving transformations, Port.
Math. 51 (1994) 185–204.
[5] G.U. Garba, The subsemigroup generated by nilpotents in a semigroup of partial order-preserving
transformations, J. Bihar Math. Soc. 21 (2001) 1–6.
 A. Laradji, A. Umar / Journal of Algebra 278 (2004) 342–359 359

[6] G.M.S. Gomes, J.M. Howie, On the ranks of certain semigroups of order-preserving transformations,
Semigroup Forum 45 (1992) 272–282.
[7] P.M. Higgins, Techniques of Semigroup Theory, Oxford University Press, 1992.
[8] P.M. Higgins, Combinatorial results for semigroups of order-preserving mappings, Math. Proc. Cambridge
Philos. Soc. 113 (1993) 281–296.
[9] J.M. Howie, Products of idempotents in certain semigroups of order-preserving transformations, Proc.
Edinburgh Math. Soc. 17 (2) (1971) 223–236.
[10] J.M. Howie, Fundamentals of Semigroup Theory, Clarendon Press, Oxford, 1995.
[11] A. Laradji, A. Umar, On certain finite semigroups of order-decreasing transformations I, Semigroup Forum,
in press.
[12] A. Laradji, A. Umar, Combinatorial results for semigroups of order-decreasing partial transformations,
submitted for publication.
[13] E. Pergola, R.A. Sulanke, Schröder triangles, paths and parallelogram polyominomes, J. Integer Seq. 1
(1998).
[14] J. Riordan, Combinatorial Identities, Wiley, New York, 1968.
[15] N.J.A. Sloane, The On-Line Encyclopedia of Integer Sequences, available at http://www.research.att.
com/~njas/sequences/.
[16] M. Tainiter, A characterisation of idempotents in semigroups, J. Combin. Theory 5 (1968) 370–373.
[17] A. Umar, On the semigroups of order-decreasing finite full transformations, Proc. Roy. Soc. Edinburgh
Sect. A 120 (1992) 129–142.
[18] A. Umar, Enumeration of certain finite semigroups of transformations, Discrete Math. 189 (1998) 291–297.