Positive fuzzy quasi-orders on semigroups⋆

Miroslav Ćirića,∗, Jelena Ignjatovića , Žarko Popovićb , Aleksandar Stamenkovića

a University of Niš, Faculty of Sciences and Mathematics, Višegradska 33, P. O. Box 224, 18000 Niš, Serbia
b University of Niš, Faculty of Economics, Trga Kralja Aleksandra 11, 18000 Niš, Serbia

Abstract
In this paper we study certain types of positive fuzzy quasi-orders on semigroups and their links with completely
semiprime fuzzy ideals, completely prime fuzzy ideals, and fuzzy filters. We characterize various properties of a
positive fuzzy quasi-order Q in terms of the properties of its left and right eigen spaces IQ and CQ , the solution sets
of eigen fuzzy set equations corresponding to Q. We also demonstrate certain applications of the obtained results to
semilattice decompositions of semigroups. The results of this paper shed new light on the known links between positive
(crisp) quasi-orders, completely semiprime ideals, completely prime ideals, and filters of semigroups, and make these
links much clearer.
Keywords: Positive fuzzy quasi-order, cm-property, cp-property, lower-potency, completely semiprime fuzzy ideal,
completely prime fuzzy ideal, fuzzy filter, semilattice decomposition.

1. Introduction

Positive quasi-orders on semigroups have been introduced by B. M. Schein in [42], in the study of semi-
groups of binary relations, and later they have been studied in different contexts, mostly in relation to semi-
lattice decompositions of semigroups. Semilattice decompositions are one of the most powerful tools used
in the study of structural properties of semigroups. They were first defined and studied by A. H. Clifford in
[18], where completely regular semigroups were characterized as semilattices of simple semigroups. Par-
ticular impetus to the study of semilattice decompositions was given by the general results of T. Tamura
and N. Kimura [52] and T. Tamura [46], according to which each semigroup possess the greatest semilattice
decomposition, whose components are semilattice indecomposable semigroups. This result initiated inten-
sive study of the greatest semilattice decompositions and the corresponding smallest semilattice congru-
ences. From the point of view of this paper, the most interesting characterizations of the smallest semilattice
congruence are those given by means of completely prime ideals and filters (M. Petrich [35]), completely
semiprime ideals (M. Ćirić and S. Bogdanović [13, 14]), and the relation that we call here Tamura’s quasi-
order (T. Tamura [48, 49]).
The study of Tamura’s quasi-order has prompted more general research on positive quasi-orders and
their role in semilattice decompositions, which was conducted in [8–10, 12, 14, 37–40, 47–51]. It is worth
mentioning that T. Tamura established in [51] an isomorphism between the lattice of semilattice congruences
on a semigroup and the lattice of positive lower-potent half-congruences on this semigrop. In the same paper
he also showed that positive lower-potent half-congruences are the same as positive quasi-orders having the
cm-property. Both concepts, the lower-potency and the cm-property, were introduced by Tamura. On the
other hand, Ćirić and Bogranović in [14] linked positivity, lower-potency and cm-property with filters,
completely prime and completely semiprime ideals.

⋆ Researchsupported by Ministry Education and Science, Republic of Serbia, Grant No. 174013
∗ Corresponding author.
Email addresses: [email protected] (Miroslav Ćirić), [email protected] (Jelena Ignjatović),
[email protected] (Žarko Popović), [email protected] (Aleksandar Stamenković)

Preprint submitted to Filomat October 6, 2019

 In this paper we discuss the mentioned problems from the point of view of fuzzy set theory. The main
tool we use are fuzzy quasi-orders. They originated as a generalization of ordinary (crisp) quasi-orders, and
like their crisp counterparts, they have wide applications in numerous fields of mathematics and computer
science. Worth mentioning that fuzzy quasi-orders have recently been used as a major tool in reducing the
number of states of fuzzy automata [17, 44] and in the positional analysis of fuzzy social networks [15, 28, 45].
Also, they naturally emerge as the greatest solutions to various systems of fuzzy relation inequations and
equations [25, 26, 28, 45] and the major tool used in computing the greatest solutions to systems of eigen
fuzzy set inequations and equations [29, 30]. Here we study positive fuzzy quasi-orders on semigroups
and certain their varieties, such as those which are lower-potent and those that have the cm-property or
cp-property. It should be noted that the lower-potency and cm-property are generalizations of the concepts
already used in the study of ordinary quasi-orders, whereas the cp-property is a brand new concept. We
also investigate the links of all these fuzzy quasi-orders with fuzzy ideals, consistent fuzzy subsets and
related concepts, such as completely semiprime and completely prime fuzzy ideals and fuzzy filters, and
demonstrate their applications in semilattice decompositions of semigroups.
The paper is organized as follows. After this section and Section 2, in which we introduce preliminary
notions and results, in the consequent sections we present the main results of the paper. First, in Theorem
3.1, we prove the equivalence of the twelve conditions that define positive fuzzy quasi-orders, and then we
provide theorems that characterize their special types (with cm-property and cp-property, lower-potent,
etc.). Some of the results obtained are generalizations of the well-known results concerning ordinary quasi-
orders, but many of them are brand new and have not had their crisp counterparts so far. It should be
emphasized that, by Theorem 3.7, we establish the duality between fuzzy ideals and consistent fuzzy
subsets, and assuming that the underlying structure of membership values is linearly ordered, we establish
the duality between completely prime fuzzy ideals and fuzzy filters. In Section 4 we first prove that the
lattice of positive fuzzy quasi-orders can be dually (anti-isomorphically) embedded into the latices of fuzzy
ideals and consistent fuzzy subsets (Theorems 4.1 and 4.2), and then we provide the construction of positive
lower-potent fuzzy quasi-orders, starting from collections of completely semiprime fuzzy ideals, and the
construction of positive fuzzy quasi-orders with the cm-property, starting from collections of completely
prime fuzzy ideals and fuzzy filters (Theorems 4.4, 4.6 and 4.7). We also give new characterizations of
positive fuzzy quasi-orders with the cm-property and cp-property (Theorems 4.5 and 4.8).
In Section 5 we provide certain applications to semilattice decompositions of semigroups. By Theorem 5.1
we establish a link between positive lower-potent fuzzy half-congruences and semilattice fuzzy congru-
ences, which is bidirectional if the membership values are taken from a complete Heyting algebra, and by
Theorem 5.2 we show that under the same conditions the concepts of a positive fuzzy quasi-order with the
cm-property and a positive lower-potent fuzzy half-congruence are identical. By Theorem 5.4 we provide a
characterization of the smallest semilattice fuzzy congruence through completely semiprime fuzzy ideals,
completely prime fuzzy ideals, and fuzzy filters, whereas Theorem 5.5 provides a version of the well-known
Prime ideal theorem for completely semiprime fuzzy ideals. It should be noted that in lattice theory, ring
theory and semigroup theory Prime ideal theorem is usually proved using the Zorn’s lemma arguments,
but here we give a proof in which Zorn’s lemma is not used.
Finally, we point out that the results of this paper shed new light on the known links between positive
quasi-orders, completely semiprime ideals, completely prime ideals, and filters of semigroups, and make
these links much clearer.

2. Preliminaries

A residuated lattice is an algebra L = (L, ∧, ∨, ⊗, →, 0, 1) such that

(L1) (L, ∧, ∨, 0, 1) is a lattice with the least element 0 and the greatest element 1,
(L2) (L, ⊗, 1) is a commutative monoid with the unit 1,
(L3) ⊗ and → form an adjoint pair, i.e., they satisfy the adjunction property: for all x, y, z ∈ L,

x ⊗ y 6 z ⇔ x 6 y → z. (1)
2
Moreover, L is called a complete residuated lattice if it satisfies (L2), (L3), and
(L1’) (L, ∧, ∨, 0, 1) is a complete lattice with the least element 0 and the greatest element 1.
The operations ⊗ (called multiplication) and → (called residuum) are intended
W for modelingV the conjunction
and implication of the corresponding logical calculus, and supremum ( ) and infimum ( ) are intended
for modeling of the existential and general quantifier, respectively. An operation ↔ defined by

x ↔ y = (x → y) ∧ (y → x), (2)

called biresiduum (or biimplication), is used for modeling the equivalence of truth values.
If L is a complete residuated lattice, then for all x, y, z ∈ L and any {xi }i∈I , {yi }i∈I ⊆ L the following holds:

x ⊗ (x → y) 6 y, (3)
x 6 y if and only if x → y = 1, (4)
x 6 y implies x ⊗ z 6 y ⊗ z, (5)
x 6 y implies z → x 6 z → y and y → z 6 x → z, (6)
(x → y) ⊗ (y → z) 6 (x → z), (7)
_ _
x⊗ yi = (x ⊗ yi ), (8)
i∈I i∈I
^ ^
x⊗ yi 6 (x ⊗ yi ), (9)
i∈I i∈I
^ _ 
(xi → y) = xi → y, (10)
i∈I i∈I
_ ^ 
(xi → y) 6 xi → y, (11)
i∈I i∈I
^ ^ 
(x → yi ) = x → yi , (12)
i∈I i∈I
_ _ 
(x → yi ) 6 x → yi , (13)
i∈I i∈I
^ ^  ^ 
(xi → yi ) 6 xi → yi , (14)
i∈I i∈I i∈I

For other properties of complete residuated lattices we refer to [2, 3].

The most studied and applied structures of truth values, defined on the real unit interval [0, 1] with
x ∧ y = min(x, y) and x ∨ y = max(x, y), are the Łukasiewicz structure (where x ⊗ y = max(x + y − 1, 0),
x → y = min(1 − x + y, 1)), the Goguen (product) structure (x ⊗ y = x · y, x → y = 1 if x 6 y, and = y/x
otherwise), and the Gödel structure (x ⊗ y = min(x, y), x → y = 1 if x 6 y, and = y otherwise). More generally,
an algebra ([0, 1], ∧, ∨, ⊗, →, 0, 1) is a complete W residuated lattice if and only if ⊗ is a left-continuous t-norm
and the residuum is defined by x → y = {u ∈ [0, 1] | u ⊗ x 6 y} (cf. [3]). Another important set of truth
values is the set {a0 , a1 , . . . , an }, 0 = a0 < · · · < an = 1, with ak ⊗ al = amax(k+l−n,0) and ak → al = amin(n−k+l,n) .
A special case of the latter algebras is the two-element Boolean algebra of classical logic with the support
{0, 1}. The only adjoint pair on the two-element Boolean algebra consists of the classical conjunction and
implication operations. This structure of truth values we call the Boolean structure. Let us note that all of the
above mentioned structures are linearly ordered.
A residuated lattice L satisfying x ⊗ y = x ∧ y, for all x, y ∈ L, is called a Heyting algebra, and a complete
residuated lattice satisfying this condition is called a complete Heyting algebra.
Let L be a complete residuated lattice. A fuzzy subset of a set A is any function from A into L. For a fuzzy
subset f : A → L and a ∈ A, we say that f (a) is the membership degree of a in f (names such as membership
value or truth degree are also used). The set of all fuzzy subsets of A taking membership values in L, i.e., the
3
set of all functions from A to L is denoted by AL . Let f, 1 ∈ AL . The equality of f and 1 is defined as the usual
equality of functions, i.e., f = 1 if and only if f (x) = 1(x), for every x ∈ A. The inclusion f 6 1 is also defined
pointwise: f 6 1 if and only if f (x) 6 1(x), for every x ∈ A. If so, we say that f is contained in 1. Endowed with
this partial order L
forms a complete lattice, in which the meet (intersection) and the join (union)
V
A i∈I fi
i∈I fi of a family { fi }i∈I of fuzzy subsets of A are functions from A into L defined by
W
   
^  ^ _  _

 fi  (x) =
 fi (x), 

 fi  (x) = fi (x), (15)
i∈I i∈I i∈I i∈I

for every x ∈ A, the least element is the empty set ∅ and the greatest one is the whole set A. Let us denote
this lattice by F (A) = (AL , ∨, ∧, ∅, A).
A crisp subset of a set A is a fuzzy subset which takes values only in the two-element set {0, 1}. If f is a
crisp subset of A, then expressions “ f (x) = 1” and “x ∈ f ” have the same meaning, i.e., f is considered as an
ordinary subset of A. The crisp part of a fuzzy subset f of A is a crisp subset f c : A → L defined by f c (a) = 1,
if f (a) = 1, and f c (a) = 0, if f (a) < 1, i.e., f c = {x ∈ A | f (x) = 1}. A fuzzy subset f of A is normalized (or modal,
in some sources) if f (x) = 1 for at least one x ∈ A, i.e., if its crisp part is non-empty.
A fuzzy relation on A is any function from A × A into L, that is to say, any fuzzy subset of A × A, and
equality, inclusion, joins, meets and ordering of fuzzy relations are defined as for fuzzy sets. Let us note
that when U 6 V, for two fuzzy relations U and V on A, we say that U is contained in V. The set of all fuzzy
relations on a set A is denoted by R(A).
For fuzzy relations U, V ∈ R(A), their composition is a fuzzy relation U ◦ V on A defined by
_
(U ◦ V)(a, b) = U(a, c) ⊗ V(c, b), (16)
c∈A

for all a, b ∈ A, and for f ∈ F (A) and U ∈ R(A), the compositions f ◦ U and U ◦ f are fuzzy subsets of
A defined by
_ _
( f ◦ U)(a) = f (b) ⊗ U(b, a), (U ◦ f )(a) = U(a, b) ⊗ f (b), (17)
b∈A b∈A

for any a ∈ A. Finally, for f, 1 ∈ F (A) we write

_
f ◦1= f (a) ⊗ 1(a). (18)
a∈A

The value f ◦ 1 can be interpreted as the "degree of overlapping" of f and 1.

For arbitrary U, V, W ∈ R(A) we have that
(U ◦ V) ◦ W = U ◦ (V ◦ W), (19)
U 6 V implies U ◦ W 6 V ◦ W and W ◦ U 6 W ◦ V, (20)
and for arbitrary f, 1 ∈ F (A) and U, V ∈ R(A) it can be easily verified that
( f ◦ U) ◦ V = f ◦ (U ◦ V), ( f ◦ U) ◦ 1 = f ◦ (U ◦ 1), (21)
and hence, the parentheses in (21) can be omitted. For n ∈ N, where N denotes the set of all natural numbers
(without zero included), an n-th power of a fuzzy relation U ∈ R(A) is a fuzzy relation Un on A defined
inductively by U1 = U and Un+1 = Un ◦ U. We also define U0 to be the ordinary equality relation on A.
For all U, V ∈ R(A) and all {Ui }i∈I , {Vi }i∈I ⊆ R(A) we have that
_  _ _  _
U◦ Vi = (U ◦ Vi ), Ui ◦ V = (Ui ◦ V), (22)
i∈I i∈I i∈I i∈I
^  ^ ^  ^
U◦ Vi 6 (U ◦ Vi ), Ui ◦ V 6 (Ui ◦ V), (23)
i∈I i∈I i∈I i∈I

4
and then the system (R(A), ∧, ∨, ◦, ∅, ∇A, ∆A ) forms a quantale, where ∇A is the universal relation on A and
∆A is the equality relation on A, i.e., for all a, b ∈ A we have that ∇A (a, b) = 1, and ∆A (a, b) = 1, if a = b, and
∆A (a, b) = 0, if a , b (for the definition of a quantale we refer to [26]).
Notice that if A is a finite set with n elements, then U and V can be treated as n × n fuzzy matrices over
L and U ◦ V is the matrix product, whereas f ◦ U can be treated as the product of a 1 × n matrix f and an
n × n matrix U, and U ◦ f as the product of an n × n matrix U and an n × 1 matrix f t (the transpose of f ),
i.e., as vector-matrix products. Also, for f, 1 ∈ F (A) we can interpret f ◦ 1 as the scalar product of vectors
f and 1.
Let R be a fuzzy relation on a set A and let f be a fuzzy subset of A. The left residual of f by R is a fuzzy
subset f /R of A defined by
^
( f /R)(a) = R(a, b) → f (b), (24)
b∈A

for each a ∈ A, and the right residual of f by R is a fuzzy subset R\ f of A defined by

^
(R\ f )(a) = R(b, a) → f (b), (25)
b∈A

for each a ∈ A. We think of the left residual f /R as what remains of f on the left after “dividing” f on the
right by R, and of the right residual R\ f as what remains of f on the right after “dividing” f on the left by R.
It is easy to check that

f ◦ R 6 1 ⇔ f 6 1/R, R ◦ f 6 1 ⇔ f 6 R\1, (26)

for all fuzzy subsets f and 1 of A and every fuzzy relation R on A. We call (26) the residuation property.
For a fuzzy relation R on A, the fuzzy relation R−1 on A defined by R−1 (a, b) = R(b, a), for every a, b ∈ A,
is called the inverse of R. A fuzzy relation R on A is said to be
(R) reflexive (or fuzzy reflexive) if ∆A 6 R, i.e., if R(a, a) = 1, for every a ∈ A;
(S) symmetric (or fuzzy symmetric) if R−1 6 R, i.e., if R(a, b) = R(b, a), for all a, b ∈ A;
(T) transitive (or fuzzy transitive) if R ◦ R 6 R, i.e., if for all a, b, c ∈ A we have

R(a, b) ⊗ R(b, c) 6 R(a, c).

For a fuzzy relation R on a set A, a fuzzy relation R∞ on A defined by R∞ = n∈N Rn is the least transitive
W
fuzzy relation on A containing R, and it is called the transitive closure of R.
A reflexive and transitive fuzzy relation on A is called a fuzzy quasi-order, and a reflexive and transitive
crisp relation on A is called a quasi-order. In some sources quasi-orders and fuzzy quasi-orders are called
preorders and fuzzy preorders. Let us note that a reflexive fuzzy relation R is a fuzzy quasi-order if and only if
R2 = R. A reflexive, symmetric and transitive fuzzy relation on A is called a fuzzy equivalence, and a reflexive,
symmetric and transitive crisp relation on A is called an equivalence. A fuzzy equivalence E on A is called a
fuzzy equality if for any a, b ∈ A, E(a, b) = 1 implies a = b. If Q is a fuzzy quasi-order on A, then EQ = Q ∧ Q−1
is the greatest fuzzy equivalence contained in Q, and it is called the natural fuzzy equivalence of Q.
With respect to the ordering of fuzzy relations, the set Q(A) of all fuzzy quasi-orders on a set A, and
the set E (A) of all fuzzy equivalences on A, form complete lattices. The meet both in Q(A) and E (A) is the
ordinary intersection of fuzzy relations, but in the general case, the joins in Q(A) and E (A) do not coincide
with the ordinary union of fuzzy relations. Namely, if {Ri }i∈IWis a family of fuzzy quasi-orders (resp. fuzzy
equivalences) on A, then its join in Q(A) (resp. in E (A)) is ( i∈I Ri )∞ , the transitive closure of the union of
this family.
Let Q be a fuzzy quasi-order on a set A. For each a ∈ A, the Q-afterset of a is the fuzzy subset aQ of A
defined by (aQ)(x) = Q(a, x), for any x ∈ A, and the Q-foreset of a is the fuzzy subset Qa of A defined by
(Qa)(x) = Q(x, a), for any x ∈ A (cf. [1, 6, 20, 29, 33, 44]). The set of all Q-aftersets will be denoted by A/Q, and
5
the set of all Q-foresets by A\Q. If E is a fuzzy equivalence, then for every a ∈ A we have that aE = Ea, and
we usually denote this fuzzy set by Ea and we call it the equivalence class of a with respect to E (cf. [16]). The
set of all equivalence classes of E is denoted by A/E and called the factor set of A with respect to E. We have
the following:

Theorem 2.1 ([29, 44]). Let Q be a fuzzy quasi-order on a set A and E the natural fuzzy equivalence of Q. Then
(a) For arbitrary a, b ∈ A the following statements are equivalent:
(i) E(a, b) = 1;
(ii) Ea = Eb ;
(iii) aQ = bQ;
(iv) Qa = Qb.
(b) Functions Ea 7→ aQ and Ea 7→ Qa are bijective functions of A/E onto A/Q and of A/E onto A\Q, respectively.

According to the previous theorem, the sets A/Q, A\Q and A/E have the same cardinality. This cardinality
will be called the index of Q, and it will be denoted by ind(Q).
If A is a finite set with n elements and a fuzzy quasi-order Q on A is treated as an n × n fuzzy matrix
over L , then Q-aftersets are row vectors, whereas Q-foresets are column vectors of this matrix.
It is important to note that a fuzzy quasi-order on a set A is uniquely determined both by the family of
all its aftersets and the family of all its foresets, since it can be reconstructed from these families as follows
^ ^
Q(a, b) = Q(c, a) → Q(c, b) = cQ(a) → cQ(b) (27)
c∈A c∈A
^ ^
= Q(b, c) → Q(a, c) = Qc(b) → Qc(a), (28)
c∈A c∈A

for all a, b ∈ A (for the proof we refer to [21, 53]). Besides, we have that
^ ^
EQ (a, b) = Q(c, a) ↔ Q(c, b) = cQ(a) ↔ cQ(b) (29)
c∈A c∈A
^ ^
= Q(b, c) ↔ Q(a, c) = Qc(b) ↔ Qc(a), (30)
c∈A c∈A

for all a, b ∈ A.
For any fuzzy subset f of A, let fuzzy relations Q f , Q f , and E f on A be defined by

Q f (a, b) = f (a) → f (b), Q f (a, b) = f (b) → f (a), E f (a, b) = f (a) ↔ f (b), (31)

for all a, b ∈ A. We have that Q f and Q f are fuzzy quasi-orders, and E f is a fuzzy equivalence on A. In par-
ticular, if f is a normalized fuzzy subset of A, then it is an afterset of Q f , a foreset of Q f , and an equivalence
class of E f .
For undefined notions and notation from fuzzy set theory we refer to [2, 3], for those from semigroup
theory we refer to [11, 24], and for those concerning ordered sets and lattices we refer to [4, 19, 41].

3. Fuzzy ideals, consistent fuzzy subsets and fuzzy quasi-orders

Let S be a semigroup and f a fuzzy subset of S. We say that f is a fuzzy ideal of S if f (a) ∨ f (b) 6 f (ab), for
all a, b ∈ S, or equivalently, if f (a) 6 f (ab) and f (b) 6 f (ab), for all a, b ∈ S, and it is a consistent fuzzy subset of
S if f (ab) 6 f (a) ∧ f (b), for all a, b ∈ S, or equivalently, if f (ab) 6 f (a) and f (ab) 6 f (b), for all a, b ∈ S.
On the other hand, f is said to be a fuzzy subsemigroup of S if f (a) ∧ f (b) 6 f (ab), for all a, b ∈ S, and a
completely prime fuzzy subset of S if f (ab) 6 f (a) ∨ f (b), for all a, b ∈ S.
6
 Lastly, f is a completely prime fuzzy ideal of S if it is both a fuzzy ideal and a completely prime fuzzy
subset, i.e., if f (ab) = f (a) ∨ f (b), for all a, b ∈ S, and it is a fuzzy filter of S if f (ab) = f (a) ∧ f (b), for all a, b ∈ S.
It is easy to verify that f is both a fuzzy ideal and a consistent fuzzy subset if and only if it is a fuzzy point,
i.e., if there exists λ ∈ L such that f (a) = λ, for every a ∈ S. Similarly, f is both a fuzzy subsemigroup and a
completely prime fuzzy subset if and only if it is a fuzzy point. It should be noted that fuzzy filters can be
viewed as homomorphisms of S into the meet-subsemilattice of L , and completely prime fuzzy ideals as
homomorphisms of S into the join-subsemilattice of L .
A slight modification of the definition of a completely prime fuzzy subset gives the notion of a completely
semiprime fuzzy subset. Namely, a fuzzy subset f of a semigroup S is said to be a completely semiprime fuzzy
subset of S if f (a2 ) 6 f (a), for every a ∈ S. If f is both a fuzzy ideal and a completely semiprime fuzzy subset,
then it is said to be a completely semiprime fuzzy ideal of S. Clearly, every completely prime fuzzy subset is
completely semiprime.
The division relation on a semigroup S is a crisp relation D : S → {0, 1} ⊆ L defined by

 1 if b = paq, for some p, q ∈ S1 ,


D(a, b) =  (32)
 0 otherwise.


Note that S1 denotes the semigroup S ∪ {1} arising from S by the adjunction of an identity element 1, unless
S already has an identity, in which case S1 = S.
The first theorem in this section associates fuzzy ideals, consistent fuzzy subsets and fuzzy quasi-orders
on a semigroup.

Theorem 3.1. Let Q be a fuzzy quasi-order on a semigroup S. Then the following statements are equivalent:
(i) f ◦ Q is a fuzzy ideal of S, for every fuzzy subset f of S;
(ii) f /Q is a fuzzy ideal of S, for every fuzzy subset f of S;
(iii) every Q-afterset is a fuzzy ideal of S;
(iv) Q satisfies the inequality

Q(c, a) ∨ Q(c, b) 6 Q(c, ab), for all a, b, c ∈ S; (33)

(v) Qa ∨ Qb 6 Qab, for all a, b ∈ S;

(vi) Q ◦ f is a consistent fuzzy subset of S, for every fuzzy subset f of S;
(vii) Q\ f is a consistent fuzzy subset of S, for every fuzzy subset f of S;
(viii) every Q-foreset is a consistent fuzzy subset of S;
(ix) Q satisfies the inequality

Q(ab, c) 6 Q(a, c) ∧ Q(b, c), for all a, b, c ∈ S; (34)

(x) abQ 6 aQ ∧ bQ, for all a, b ∈ S;

(xi) Q satisfies the equalities

Q(a, ab) = Q(b, ab) = 1, for all a, b ∈ S; (35)

(xii) Q contains the division relation on S.

Proof. It is enough to prove sequences of implications (i)⇒(iii)⇒(iv)⇒(i), (ix)⇒(ii)⇒(iii) and (v)⇒(xi)⇒(iv),

and the equivalences (iv)⇔(v), since (vi)⇒(viii)⇒(ix)⇒(vi), (iv)⇒(vii)⇒(viii) and (ix)⇔(x) can be proved
in a similar manner. The verification of (xi)⇔(xii) is straightforward.
(i)⇒(iii). Consider an arbitrary a ∈ S. For each b ∈ S we have that
_ _
((aQ) ◦ Q)(b) = (aQ)(c) ⊗ Q(c, b) = Q(a, c) ⊗ Q(c, b) = Q2 (a, b) = Q(a, b) = (aQ)(b),
c∈S c∈S

7
so (aQ) ◦ Q = aQ, and according to (i), aQ is a fuzzy ideal of S.
(iii)⇒(iv). This follows directly from definitions of a Q-afterset and a fuzzy ideal.
(iv)⇒(i). Notice first that (33) is equivalent to Q(a, b) 6 Q(a, bc) and Q(a, c) 6 Q(a, bc), for all a, b, c ∈ S.
Consider now an arbitrary fuzzy subset f of S and arbitrary a, b ∈ S. Then
_ _
( f ◦ Q)(a) = f (c) ⊗ Q(c, a) 6 f (c) ⊗ Q(c, ab) = ( f ◦ Q)(ab),
c∈S c∈S

and similarly, ( f ◦ Q)(b) 6 ( f ◦ Q)(ab). Therefore, we conclude that f ◦ Q is a fuzzy ideal of S.

(iv)⇔(v). This equivalence can be easily verified.
(ix)⇒(ii). Since the residuum operation is antitone in the first argument, for any fuzzy subset f of S and
all a, b ∈ S we obtain that
^ ^
( f /Q)(a) = Q(a, c) → f (c) 6 Q(ab, c) → f (c) = ( f /Q)(ab),
c∈S c∈S

and
^ ^
( f /Q)(b) = Q(b, c) → f (c) 6 Q(ab, c) → f (c) = ( f /Q)(ab).
c∈S c∈S

Hence, f /Q is a fuzzy ideal of S.

(ii)⇒(iii). Consider arbitrary a, b ∈ S. For each c ∈ S we have that Q(a, b) ⊗ Q(b, c) 6 Q(a, c), which is equi-
valent to Q(a, b) 6 Q(b, c) → Q(a, c), whence it follows
^
Q(a, b) 6 Q(b, c) → Q(a, c) 6 Q(b, b) → Q(a, b) = 1 → Q(a, b) = Q(a, b),
c∈S

and therefore,
^
Q(a, b) = Q(b, c) → Q(a, c).
c∈S

Now we obtain that

^ ^
((aQ)/Q)(b) = Q(b, c) → (aQ)(c) = Q(b, c) → Q(a, c) = Q(a, b) = (aQ)(b).
c∈S c∈S

Hence, for each a ∈ S we have that (aQ)/Q = aQ, and according to (ii), aQ is a fuzzy ideal of S.
(v)⇒(xi). Consider arbitrary a, b ∈ S. According to (v) we have that Qa 6 Qab, which implies that

1 = Qa(a) 6 Qab(a) = Q(a, ab).

Thus, Q(a, ab) = 1. In the exactly same way we show that Q(b, ab) = 1.
(xi)⇒(iv). Let a, b, c ∈ S. Because of the assumption Q(b, bc) = 1 and the transitivity of Q, we get

Q(c, a) = Q(c, a) ⊗ Q(a, ab) 6 Q(c, ab),

and analogously Q(c, b) 6 Q(c, ab). Thus, Q(c, a) ∨ Q(c, b) 6 Q(c, ab).
This completes the proof of the theorem.
A fuzzy quasi-order on a semigroup S satisfying any of the twelve equivalent conditions of Theorem 3.1
(particularly the condition (35)) will be called a positive fuzzy quasi-order.
Let us note that condition (xi) of Theorem 3.1 means that the crisp part of Q is a positive crisp quasi-order
(cf. [14]). In other words, a fuzzy quasi-order Q on a semigroup S is positive if and only if its crisp part is a
positive crisp quasi-order on S.
8
Theorem 3.2. Let Q be a fuzzy quasi-order on a semigroup S. Then the following statements are equivalent:
(i) f ◦ Q is a completely prime fuzzy subset of S, for every fuzzy subset f of S;
(ii) every Q-afterset is a completely prime fuzzy subset of S;
(iii) Q satisfies the inequality

Q(c, ab) 6 Q(c, a) ∨ Q(c, b), for all a, b, c ∈ S; (36)

(iv) Qab 6 Qa ∨ Qb, for all a, b ∈ S;

(v) Q satisfies the equality

Q(ab, a) ∨ Q(ab, b) = 1, for all a, b ∈ S. (37)

Proof. The implication (i)⇒(ii) can be proved in the same way as (i)⇒(iii) in Theorem 3.1, and the implication
(ii)⇒(iii) and the equivalence (iii)⇔(iv) are straightforward.
(iii)⇒(i). Consider an arbitrary fuzzy subset f of S and arbitrary a, b ∈ S. According to (36) and (8) we
have that
_ _   _   
( f ◦ Q)(ab) = f (c) ⊗ Q(c, ab) 6 f (c) ⊗ Q(c, a) ∨ Q(c, b) = f (c) ⊗ Q(c, a) ∨ f (c) ⊗ Q(c, b)
c∈S c∈S c∈S
_  _ 
= ( f (c) ⊗ Q(c, a) ∨ f (c) ⊗ Q(c, b) = ( f ◦ Q)(a) ∨ ( f ◦ Q)(b),
c∈S c∈S

and thus, f ◦ Q is a completely prime fuzzy subset of S.

(iii)⇒(v). For arbitrary a, b ∈ S we have that

1 = Q(ab, ab) 6 Q(ab, a) ∨ Q(ab, b),

so we conclude that (37) holds.

(v)⇒(iii). Consider arbitrary a, b, c ∈ S. Then
     
Q(c, ab) = Q(c, ab) ⊗ 1 = Q(c, ab) ⊗ Q(ab, a) ∨ Q(ab, b) = Q(c, ab) ⊗ Q(ab, a) ∨ Q(c, ab) ⊗ Q(ab, b)
6 Q(c, a) ∨ Q(c, b),

and hence, we have that (36) holds.

A fuzzy quasi-order on a semigroup S satisfying any of the equivalent conditions of Theorem 3.2 (par-
ticularly the condition (36)) is said to have the cp-property (the abbreviation for complete primeness property).

Theorem 3.3. Let Q be a fuzzy quasi-order on a semigroup S. Then the following statements are equivalent:
(i) every Q-foreset is a fuzzy subsemigroup of S;
(ii) Q satisfies the inequality

Q(a, c) ∧ Q(b, c) 6 Q(ab, c), for all a, b, c ∈ S; (38)

(iii) aQ ∧ bQ 6 abQ, for all a, b ∈ S.

Proof. The equivalence (i)⇔(ii) follows directly from definitions of a Q-foreset and a fuzzy subsemigroup,
and the equivalence (ii)⇔(iii) is straightforward.
A fuzzy quasi-order on a semigroup satisfying the inequality (38) is said to have the cm-property (the
abbreviation for common multiple property).
It is worth emphasizing once again that the inequalities (33) and (34) are equivalent to each other.
However, we will show that the reverse inequalities (36) and (38) are not necessarily equivalent.
9
Theorem 3.4. Let Q be a fuzzy quasi-order on a semigroup S.
If Q has the cp-property, then it also has the cm-property, but the reverse implication does not necessarily hold.

Proof. Let Q have the cp-property. According to this property and equalities (27) and (10) we have that
^  ^ 
Q(a, c) ∧ Q(b, c) = Q(d, a) → Q(d, c) ∧ Q(d, b) → Q(d, c)
d∈S d∈S
^    ^ 
= Q(d, a) → Q(d, c) ∧ Q(d, b) → Q(d, c) = Q(d, a) ∨ Q(d, b) → Q(d, c)
d∈S d∈S
^
6 Q(d, ab) → Q(d, c) = Q(ab, c),
d∈S

which means that Q has the cm-property.

On the other hand, in any commutative semigroup the division relation has the cm-property. However,
if S is a null semigroup, i.e., S has a zero 0 and xy = 0, for all x, y ∈ S, and S has at least two non-zero elements
a and b, then the division relation on S does not have the cp-property since

D(ab, a) ∨ D(ab, b) = D(0, a) ∨ D(0, b) = 0.

Therefore, the cm-property does not necessarily imply the cp-property.

The next theorem characterizes positive fuzzy quasi-orders having the cm-property.

Theorem 3.5. Let Q be a fuzzy quasi-order on a semigroup S. Then the following statements are equivalent:

(i) every Q-foreset is a fuzzy filter of S;

(ii) Q is positive and has the cm-property, that is, it satisfies the equality

Q(ab, c) = Q(a, c) ∧ Q(b, c), for all a, b, c ∈ S; (39)

(iii) aQ ∧ bQ = abQ, for all a, b ∈ S.

Proof. This theorem follows directly from Theorems 3.1 and 3.3.

A fuzzy quasi-order Q on a set A is said to be prelinear if for arbitrary a, b ∈ A we have Q(a, b) ∨Q(b, a) = 1,
and it is said to be linear if for all a, b ∈ A we have Q(a, b) = 1 or Q(b, a) = 1. It is clear that any linear fuzzy
quasi-order is prelinear, and if the underlying complete residuated lattice L is linearly ordered, then these
two notions coincide.
Now we characterize positive fuzzy quasi-orders having the cp-property.

Theorem 3.6. Let Q be a fuzzy quasi-order on a semigroup S. Then the following statements are equivalent:

(i) f ◦ Q is a completely prime fuzzy ideal of S, for every fuzzy subset f of S;

(ii) every Q-afterset is a completely prime fuzzy ideal of S;
(iii) Q is positive and has the cp-property;
(iv) Q is positive, prelinear and has the cm-property;
(v) Q satisfies the equalities

Q(ab, c) = Q(a, c) ∧ Q(b, c), Q(c, ab) = Q(c, a) ∨ Q(c, b), for all a, b, c ∈ S; (40)

(vi) Qab = Qa ∨ Qb, for all a, b ∈ S.

10
Proof. The equivalence of statements (i), (ii), (iii) and (vi) follows directly from Theorems 3.1 and 3.2, while
the equivalence (iii)⇔(v) follows from Theorem 3.4. It remains to prove the equivalence of (iii) and (iv).
(iii)⇔(iv). If Q is positive and has the cm-property, then by (39) we obtain that
   
Q(ab, b) ∨ Q(ab, a) = Q(a, b) ∧ Q(b, b) ∨ Q(a, a) ∧ Q(b, a) = Q(a, b) ∨ Q(b, a). (41)

Now, if Q is positive and has the cp-property, then it also has the cm-property, so by (37) and (41) it follows
that Q is prelinear. Conversely, if Q is positive, prelinear and has the cm-property, then (41) yields (37), so
we conclude that Q has the cp-property.
An important connection between fuzzy ideals and consistent fuzzy subsets, completely prime fuzzy
subsets and fuzzy subsemigroups, as well as between completely prime fuzzy ideals and fuzzy filters is
given by the following theorem.

Theorem 3.7. Let S be a semigroup, let f be a fuzzy subset of S and λ ∈ L a fixed scalar, and let a fuzzy subset 1 of S
be defined by 1(a) = f (a) → λ, for each a ∈ S. Then the following is true:

a) fuzzy ideal consistent fuzzy subset

b) consistent fuzzy subset fuzzy ideal
If f is a , then 1 is a .
c) completely prime fuzzy subset fuzzy subsemigroup
d) completely prime fuzzy ideal fuzzy filter

If the membership values are taken in a linearly ordered complete residuated lattice, then the following is also true:

e) fuzzy subsemigroup completely prime fuzzy subset

If f is a , then 1 is a .
f) fuzzy filter completely prime fuzzy ideal

Proof. a) Let f be a fuzzy ideal of S. According to (6) and (10), for arbitrary a, b ∈ S we have that

1(ab) = f (ab) → λ 6 ( f (a) ∨ f (b)) → λ = ( f (a) → λ) ∧ ( f (b) → λ) = 1(a) ∧ 1(b),

and thus, 1 is a consistent fuzzy subset of S.

In a similar way we prove c) and d).
b) Let f be a consistent fuzzy subset of S. According to (6) and (11), for arbitrary a, b ∈ S we get

1(ab) = f (ab) → λ > ( f (a) ∧ f (b)) → λ > ( f (a) → λ) ∨ ( f (b) → λ) = 1(a) ∨ 1(b),

and we conclude that 1 is a fuzzy ideal of S.

Further, suppose that the underlying complete residuated lattice L is linearly ordered.
e) Let f be a fuzzy subsemigroup of S. Consider arbitrary a, b ∈ S. Since the underlying complete resid-
uated lattice L is linearly ordered, we can distinguish two cases: f (a) 6 f (b) and f (b) 6 f (a).
If f (a) 6 f (b), then f (a) = f (a) ∧ f (b) 6 f (ab), and according to (6) we get

1(b) = f (b) → λ 6 f (a) → λ = 1(a),

and

1(ab) = f (ab) → λ 6 f (a) → λ = 1(a) = 1(a) ∨ 1(b).

In the same way we show that f (b) 6 f (a) implies 1(ab) 6 1(a)∨1(b). Thus, we conclude that 1 is a completely
prime fuzzy subset of S.
The claim f) follows directly from b) and e).
11
 It should be noted that the negation in a residuated lattice is given by ¬x = x → 0, which means that the
fuzzy set 1, defined in the previous theorem by 1(a) = f (a) → λ, is some kind of a generalized complement of
f (the usual complement is obtained for λ = 0).
We prove the following theorem in an analogous way as the previous one.

Theorem 3.8. Let S be a semigroup, let f be a fuzzy subset of S and λ ∈ L a fixed scalar, and let a fuzzy subset 1 of S
be defined by 1(a) = λ → f (a), for each a ∈ S. Then the following is true:

(a) If f is a fuzzy ideal (resp. consistent fuzzy subset, fuzzy subsemigroup, fuzzy filter) of S, then 1 has the same
property.

If the membership values are taken in a linearly ordered complete residuated lattice, then the following is also true:

(b) If f is a completely prime fuzzy subset (resp. completely prime fuzzy ideal) of S, then 1 has the same property.

This section will be completed by a theorem which characterizes fuzzy quasi-orders whose aftersets are
completely semiprime fuzzy subsets and ideals. The theorem can be proved in a similar way as Theorem 3.2,
so its proof will be omitted.

Theorem 3.9. Let Q be a fuzzy quasi-order on a semigroup S. Then the following statements are equivalent:

(i) f ◦ Q is a completely semiprime fuzzy subset of S, for every fuzzy subset f of S;

(ii) every Q-afterset is a completely semiprime fuzzy subset of S;
(iii) Q satisfies the inequality

Q(b, a2 ) 6 Q(b, a), for all a, b ∈ S; (42)

(iv) Qa2 6 Qa, for each a ∈ S;

(v) Q satisfies the inequality

Q(a, b) 6 Q(a2 , b), for all a, b ∈ S; (43)

(vi) Q satisfies the equality

Q(a2 , a) = 1, for each a ∈ S. (44)

In addition, if Q is positive, then the term “fuzzy set” in (i) and (ii) can be replaced by the term “fuzzy ideal”, and (44)
is equivalent to

Q(an , a) = 1, for all a ∈ S and n ∈ N. (45)

A fuzzy quasi-order satisfying any of the equivalent conditions of Theorem 3.9 is called lower-potent. Each
fuzzy quasi-order having the cm-property is lower-potent, because (38) implies (43).

4. Eigen spaces of a positive fuzzy quasi-order

Let R be a fuzzy relation on a set A. A fuzzy subset f of A is called a left eigen fuzzy set of R if f ◦ R = f ,
and the set of all left eigen fuzzy sets of R is called the left eigen space of R. Similarly, f is called a right eigen
fuzzy set of R if R ◦ f = f , and the set of all right eigen fuzzy sets of R is called the right eigen space of R. Here
we consider (left and right) eigen spaces of positive fuzzy quasi-orders.
Let us consider the complete lattice F (A) = (AL , ∨, ∧, ∅, A) of all fuzzy subsets of a set A with membership
values in the complete residuated lattice L . For any λ ∈ L and f ∈ AL let us define the scalar multiplication
12
λ f as follows: λ f (a) = λ ⊗ f (a), for any a ∈ A. With respect to this scalar multiplication, the monoid (AL , ∨, ∅)
forms a left L-semimodule (cf. [22, 23]), i.e., for all λ, λ1 , λ2 ∈ L and f, f1 , f2 ∈ AL the following is true:

(λ1 ⊗ λ2 ) f = λ1 (λ2 f ), (46)

λ( f1 ∨ f2 ) = λ f1 ∨ λ f2 , (47)
(λ1 ∨ λ2 ) f = λ1 f ∨ λ2 f, (48)
1 f = f, (49)
λ∅ = 0 f = ∅. (50)

It is easy to see that (47) and (48) hold for arbitrary joins, including the infinite ones. The lattice F (A) equipped
with this scalar multiplication will be denoted by F⊗ (A) and called the L -lattice of fuzzy subsets of the set A.
A fuzzy subset f ∈ AL is said to be a linear combination of fuzzy subsets fi ∈ AL (i ∈ I) if there exist scalars
λi ∈ L (i ∈ I) such that f is expressed in the form
_
f = λi fi . (51)
i∈I

Any subset of AL which is closed under scalar multiplication and arbitrary meets and joins, and it contains
the least and the greatest element of F (A) will be called a complete L -sublattice of F⊗ (A).
A subset X of AL is called a closure system in F (A) if A ∈ X and X is closed under arbitrary meets. If
X is a closure system, then for any a ∈ A the family { f ∈ X | f (a) = 1} is non-empty, since it contains A, so
^n  o
Xa = f ∈ X  f (a) = 1 ∈ X . (52)

The fuzzy set Xa will be called the principal element of C generated by a, and the set P(X ) = {Xa | a ∈ A}
will be called the principal part of X . If for arbitrary a, b ∈ A we have that
^
Xa (b) = I(Xa , Xb ) = Xa (c) → Xb (c), (53)
c∈A

then we say that the principal elements of X satisfy the inclusion property, and if
^
Xa (b) = I(Xb , Xa ) = Xb (c) → Xa (c), (54)
c∈A

then we say that the principal elements of C satisfy the reverse inclusion property.
Let I(S) denote the set of all fuzzy ideals of a semigroup S. It is easy to verify that I(S) is a complete
L -sublattice of F⊗ (S), and this L -lattice is denoted by I⊗ (S) and called the L -lattice of fuzzy ideals of S.
For any positive fuzzy quasi-order Q on a semigroup S we define a collection IQ of fuzzy ideals of S by
n  o
IQ = f ∈ I(S)  f ◦ Q = f .

In other words, IQ is the left eigen space of Q.

Left eigen spaces of positive fuzzy quasi-orders are characterized by the next theorem. The proof of the
first part of the theorem is similar to the proof of Theorem 6.4 from [29], but for the sake of completeness
we give a a full proof.

Theorem 4.1. Let I be a collection of fuzzy ideals of a semigroup S. Then there exists a positive fuzzy quasi-order
Q on S such that I = IQ if and only if the following hold:
(1) I is a complete L -sublattice of I⊗ (S);
(2) every element of I can be expressed as a linear combination of principal elements of I ;
(3) principal elements of I satisfy the reverse inclusion property.
13
Furthermore, the function Q 7→ IQ is a dual order isomorphism of the lattice of positive fuzzy quasi-orders on S onto
the partially ordered set of all complete L -sublattices of I⊗ (S) satisfying conditions (2) and (3), i.e.,

Q1 6 Q2 ⇔ IQ2 ⊆ IQ1 , (55)

for arbitrary positive fuzzy quasi-orders Q1 and Q2 on S.

Proof. Let I = IQ , for some positive fuzzy quasi-order Q on S. For any family { fi }i∈I ⊆ IQ we have that
_  _ _ ^  ^ ^ ^ 
fi ◦ Q = fi ◦ Q = fi , fi ◦ Q 6 fi ◦ Q = fi 6 fi ◦ Q,
i∈I i∈I i∈I i∈I i∈I i∈I i∈I

which means that IQ is closed under arbitrary joins and meets, and it is clear that it is closed under scalar
multiplication. Therefore, IQ is a complete L -sublattice of I⊗ (S).
Next, for arbitrary a, b ∈ S we have that
_ _
(aQ ◦ Q)(b) = aQ(c) ⊗ Q(c, b) = Q(a, c) ⊗ Q(c, b) = Q(a, b) = aQ(b),
c∈S c∈S

so aQ ◦ Q = aQ. This means that aQ ∈ IQ , for each a ∈ S. It is clear that aQ(a) = 1, for every a ∈ S. Let a ∈ S
and let f ∈ IQ such that f (a) = 1. Then for any b ∈ S we have that
_
f (b) = ( f ◦ Q)(b) = f (c) ⊗ Q(c, b) > f (a) ⊗ Q(a, b) = Q(a, b) = aQ(b),
c∈S

which implies that aQ 6 f . Therefore, aQ = Ia , for every a ∈ S. Now, according to (28), for arbitrary a, b ∈ S
we obtain that
^ ^ ^
Ib (c) → Ia (c) = bQ(c) → aQ(c) = Qc(b) → Qc(a) = Q(a, b) = Ia (b),
c∈S c∈S c∈S

which means that principal elements of IQ satisfy the reverse inclusion property.
Finally, for arbitrary f ∈ IQ and a ∈ A we have that
_ _ _ 
f (a) = ( f ◦ Q)(a) = f (b) ⊗ Q(b, a) = f (b) ⊗ Ib (a) = f (b)Ib (a),
b∈S b∈S b∈S

and this implies

_
f = f (b)Ib .
b∈S

Thus, every element of I can be expressed as a linear combination of principal elements of I . Accordingly,
we have proved that (1), (2) and (3) hold.
Conversely, let I satisfy conditions (1), (2) and (3). Let us define a fuzzy relation Q on S by
^
Q(a, b) = Ia (b) = Ib (c) → Ia (c), (56)
c∈S

for all a, b ∈ S. It is clear that Q is reflexive. Consider arbitrary a, b, c, d ∈ S. According to (7), we have that

Q(a, b) ⊗ Q(b, c) 6 (Ib (d) → Ia (d)) ⊗ (Ic (d) → Ib (d)) 6 Ic (d) → Ia (d),

and since this holds for every d ∈ S, we conclude that

^
Q(a, b) ⊗ Q(b, c) 6 Ic (d) → Ia (d) = Q(a, c).
d∈S

14
Therefore, Q is transitive, i.e., Q is a fuzzy quasi-order. According to (56), for an arbitrary a ∈ S we have
that Ia = aQ, and by Theorem 3.1 we obtain that Q is a positive fuzzy quasi-order.
Consider an arbitrary f ∈ IQ , i.e., a fuzzy ideal of S satisfying f ◦ Q = f . Then for each a ∈ S we have
_ _ _ 
f (a) = f (b) ⊗ Q(b, a) = f (b) ⊗ Ib (a) = f (b)Ib (a),
b∈S b∈S b∈S

whence
_
f = f (b)Ib .
b∈S

Since Ib ∈ I , for every b ∈ S, and I is closed under scalar products and arbitrary joins, we conclude that
f ∈ I . Therefore, IQ ⊆ I .
On the other hand, let f ∈ I . According to (2), there are scalars λc ∈ L, c ∈ S, such that
_
f = λc Ic ,
c∈S

and for any a ∈ S we have that

_ __  _ _ 
( f ◦ Q)(a) = f (b) ⊗ Q(b, a) = λc ⊗ Ic (b) ⊗ Q(b, a) = λc ⊗ Q(c, b) ⊗ Q(b, a)
b∈S b∈S c∈S c∈S b∈S
_ _
= λc ⊗ Q(c, a) = λc ⊗ Ic (a) = f (a).
c∈S c∈S

Thus, f ∈ IQ , so I ⊆ IQ . Consequently, we have proved that I = IQ .

Consider arbitrary positive fuzzy quasi-orders Q1 and Q2 on S. If Q1 6 Q2 and f ∈ IQ2 , i.e., f ◦ Q2 = f ,
then f 6 f ◦ Q1 6 f ◦ Q2 = f , so f ◦ Q1 = f , and hence, f ∈ IQ1 . Conversely, let IQ2 ⊆ IQ1 . Then for
each a ∈ S we have that aQ2 ∈ IQ2 ⊆ IQ1 , which means that aQ2 ◦ Q1 = aQ2 . Thus, for arbitrary a, b ∈ S we
have that
_
Q2 (a, b) = aQ2 (b) = (aQ2 ◦ Q1 )(b) = aQ2 (c) ⊗ Q1 (c, b) > aQ2 (a) ⊗ Q1 (a, b) = Q1 (a, b),
c∈S

whence Q1 6 Q2 . Therefore, we have proved that (57) holds.

It is worth noting that in the crisp case conditions (2) and (3) in Theorem 4.1 are trivially satisfied. Namely,
in this case condition (3) means that b ∈ Ia if and only if Ib ⊆ Ia , and condition (2) means that every f ∈ I
is the union of all principal elements of I contained in f . Moreover, in the crisp case the only scalars are 0
and 1, and according to (49) and (50), closeness under scalar multiplication is redundant.
Next we consider the lattice of consistent fuzzy subsets. For any non-empty set A, let F⊗′ (A) denote
the lattice F (A) equipped with the scalar multiplication written on the right (a right L-semimodule), i.e.,
f λ(a) = f (a) ⊗ λ, for all f ∈ AL , λ ∈ L and a ∈ A. It is easy to see that in this case conditions dual to (46)–(50)
are satisfied.
Let C(S) denote the set of all consistent fuzzy subsets of a semigroup S. It is easy to verify that C(S) is
a complete L -sublattice of F⊗′ (S), and this L -lattice will be denoted by C⊗ (S) and called the L -lattice of
consistent fuzzy subsets of S.
For any positive fuzzy quasi-order Q on S we define a collection CQ of consistent fuzzy subsets of S by
n  o
CQ = f ∈ C(S)  Q ◦ f = f .

In other words, CQ is the right eigen space of Q.

The proof of the following theorem is similar to the proof of the Theorem 4.1 and it will be omitted.
15
Theorem 4.2. Let C be a collection of consistent fuzzy subsets of a semigroup S. Then there exists a positive fuzzy
quasi-order Q on S such that C = CQ if and only if the following hold:
(1) C is a complete L -sublattice of C⊗ (S);
(2) every element of C can be expressed as a linear combination of principal elements of C ;
(3) principal elements of C satisfy the inclusion property.
Furthermore, the function Q 7→ CQ is a dual order isomorphism of the lattice of positive fuzzy quasi-orders on S onto
the partially ordered set of all complete L -sublattices of C⊗ (S) satisfying conditions (2) and (3), i.e.,

Q1 6 Q2 ⇔ CQ2 ⊆ CQ1 , (57)

for arbitrary positive fuzzy quasi-orders Q1 and Q2 on S.

For any fuzzy subset f of a semigroup S and any fuzzy quasi-order Q on S, by the reflexivity of Q and
the residuation property (26) we get the following chain of equivalences:

f ◦Q= f ⇔ f ◦Q6 f ⇔ f 6 f /Q ⇔ f = f /Q.

Therefore, f ◦ Q = f if and only if f = f /Q, and analogously, Q ◦ f = f if and only if f = Q\ f . This means
that for any positive fuzzy quasi-order Q on S we have that
n  o n  o
IQ = f ∈ I(S)  f ◦ Q = f = f ∈ I(S)  f /Q = f , (58)
n  o n  o
CQ = f ∈ C(S)  Q ◦ f = f = f ∈ C(S)  Q\ f = f . (59)

It is important to note that functions f 7→ f ◦ Q and f 7→ Q ◦ f are closure operators on the lattice F (S), and
f 7→ f /Q and f 7→ Q\ f are opening operators on F (S) (under assumption that Q is a fuzzy quasi-order,
cf. [5, 7, 29]), so IQ is the set of all closed elements w.r.t. f 7→ f ◦ Q and open elements w.r.t. f 7→ f /Q,
whereas CQ is the set of all closed elements w.r.t. f 7→ Q ◦ f and open elements w.r.t. f 7→ Q\ f .
The following theorem characterizes positive fuzzy quasi-orders which determine complete L -sublattices
of F⊗ (S) consisting of completely semiprime fuzzy ideals.

Theorem 4.3. Let Q be a positive quasi-fuzzy order on a semigroup S. Then IQ consists of completely semiprime
fuzzy ideals if and only if Q is lower-potent.

Proof. If IQ consists of completely semiprime fuzzy ideals, then all Q-aftersets are completely semiprime
fuzzy ideals, and according to Theorem 3.9 we have that Q is lower-potent.
Conversely, if Q is lower-potent, by Theorem 3.9 it follows that principal elements of IQ are completely
semiprime fuzzy ideals, and since every element of IQ can be represented as a linear combination of prin-
cipal elements and the set of all completely semiprime fuzzy ideals is closed under linear combinations, we
conclude every element of IQ is a completely semiprime fuzzy ideal.
The followsing theorem provides the construction of a positive lower-potent fuzzy quasi-order starting
from a collection of completely semiprime fuzzy ideals.

Theorem 4.4. Let X be a collection of completely semiprime fuzzy ideals of a semigroup S and let Q be a fuzzy
relation on S defined by
^
Q(a, b) = f (a) → f (b), for all a, b ∈ S. (60)
f ∈X

Then Q is a positive lower-potent fuzzy quasi-order on S such that X ⊆ IQ .

If X is the collection of all completely semiprime fuzzy ideals of S, then X = IQ and Q is the smallest positive
lower-potent fuzzy quasi-order on S.
16
Proof. In accordance with the remarks made at the end of Section 2 (see (31)) we have that Q is the meet of
the family of fuzzy quasi-orders {Q f } f ∈X , so Q is also a fuzzy quasi-order.
For a fixed a ∈ S we have that
^
aQ(b) = f (a) → f (b),
f ∈X

for each b ∈ S, and by Theorem 3.8 we obtain that aQ is a fuzzy ideal, for each a ∈ S. Hence, by Theorem 3.1
we conclude that Q is a positive fuzzy quasi-order.
From (4) we conclude that f (a2 ) → f (a) = 1, for all a ∈ S and f ∈ X , which means that Q(a2 , a) = 1, for
each a ∈ S. Consequently, Q is a lower-potent fuzzy quasi-order.
According to (9) and (3), for arbitrary 1 ∈ X and a ∈ S we have that
_ _ ^ ! _ ^ !
 
(1 ◦ Q)(a) = 1(b) ⊗ Q(b, a) = 1(b) ⊗ f (b) → f (a) 6 1(b) ⊗ f (b) → f (a)
b∈S b∈S f ∈X b∈S f ∈X
_  
6 1(b) ⊗ 1(b) → 1(a) 6 1(a),
b∈S

which yields 1 ◦ Q 6 1. Since the opposite inequality is an immediate consequence of the reflexivity of Q, we
conclude that 1 ◦ Q = 1, and thus 1 ∈ IQ , which finally yields X ⊆ IQ .
Furthermore, let X be the collection of all completely semiprime fuzzy ideals of S. As we have just
shown, X ⊆ IQ , and since Theorem 4.3 implies the opposite inclusion, we conclude that X = IQ . Let R be
an arbitrary positive lower-potent fuzzy quasi-order on S. Then from Theorem 3.9 it follows that cR ∈ X ,
for each c ∈ S, and according to (27) we get
^ ^
Q(a, b) = f (a) → f (b) 6 cR(a) → cR(b) = R(a, b),
f ∈X c∈S

for all a, b ∈ S. Therefore, Q 6 R. This completes the proof of the theorem.

The next theorem gives new characterizations of positive fuzzy quasi-orders having the cm-property in
terms of properties of the corresponding lattices of fuzzy ideals and consistent fuzzy subsets. Note that the
theorem is proved under assumption that the underlying complete residuated lattice is linearly ordered. If
we take into account that in the practical application of the theory of fuzzy sets, the most commonly used
structures of truth-values are those defined by left-continuous t-norms on the real unit interval, which are
linearly ordered, this assumption does not seem too restrictive.

Theorem 4.5. Let Q be a positive quasi-fuzzy order on a semigroup S, and let the underlying complete residuated
lattice L be linearly ordered. Then the following statements are equivalent:
(i) Q has the cm-property;
(ii) every principal element of CQ is a fuzzy filter;
(iii) every element of IQ can be expressed as the meet of some family of completely prime fuzzy ideals from IQ .

Proof. (i)⇔(ii). This follows directly from Theorem 3.5.

(ii)⇒(iii). Consider an arbitrary f ∈ IQ . By (58) it follows that f = f /Q, and for each a ∈ S we have that
^ ^ ^
f (a) = ( f /Q)(a) = Q(a, b) → f (b) = Qb(a) → f (b) = fb (a),
b∈S b∈S b∈S

where for each b ∈ S we put fb (a) = Qb(a) → f (b). By assumption, Qb is a fuzzy filter, and by Theorem 3.7
we obtain that fb is a completely prime fuzzy ideal of S. Therefore, we have expressed f as the meet of the
family { fb }b∈S of completely prime fuzzy ideals of S.
17
 (iii)⇒(i). In order to prove that Q has the cm-property, consider arbitrary a, b ∈ S. By abQ ∈ IQ and the
assumption (4) we obtain that
^
abQ = fi ,
i∈I

where for each i ∈ I, fi is a completely prime fuzzy ideal of S, and we have that
^
fi (ab) = abQ(ab) = Q(ab, ab) = 1,
i∈I

which implies that fi (ab) = 1, for each i ∈ I. Since fi is a completely prime fuzzy ideal of S we get
fi (a) ∨ fi (b) = fi (ab) = 1,
for each i ∈ I, and as the assumption of the theorem states that L is linearly ordered, we have that fi (a) = 1
or fi (b) = 1, for any i ∈ I. Now, let us set Ia = {i ∈ I | fi (a) = 1} and Ib = {i ∈ I | fi (b) = 1}. Then I = Ia ∪ Ib , and
without loss of generality we can suppose that Ia , ∅ and Ib , ∅ (we can take S to be one of fi ’s). Since aQ
and bQ are principal elements of IQ , we have that aQ 6 fi and bQ 6 f j , for all i ∈ Ia and j ∈ Ib , whence
^  ^  ^
aQ ∧ bQ 6 fi ∧ fj = fi = abQ.
i∈Ia j∈Ib i∈I

This means that aQ ∧ bQ 6 abQ, for all a, b ∈ S, and in accordance with Theorem 3.3 we conclude that Q has
the cm-property.
We further prove the following:
Theorem 4.6. Let X be a collection of completely prime fuzzy ideals of a semigroup S and let Q be a fuzzy relation
on S defined as in (60). Then Q is a positive fuzzy quasi-order on S having the cm-property such that X ⊆ IQ .
If X is the collection of all completely prime fuzzy ideals of S, then Q is the smallest positive fuzzy quasi-order on
S having the cm-property.
Proof. According to Theorem 4.4, Q is a positive lower-potent fuzzy quasi-order and X ⊆ IQ .
Let us consider arbitrary a, b, c ∈ S. Using (10) we get
^  ^  ^   
Q(a, c) ∧ Q(b, c) = f (a) → f (c) ∧ f (b) → f (c) = f (a) → f (c) ∧ f (a) → f (c)
f ∈X f ∈X f ∈X
^  ^
= f (a) ∨ f (b) → f (c) = f (ab) → f (c) = Q(ab, c).
f ∈X f ∈X

Hence, Q has the cm-property.

Furthermore, let X be the collection of all completely prime fuzzy ideals of S, and let R be an arbitrary
positive fuzzy quasi-order on S with the cm-property. According to Theorem 4.5, for any c ∈ S we have that
^
cR = fi ,
i∈Ic

where { fi }i∈Ic is some family of completely prime fuzzy ideals. Put I = c∈S Ic . Using (27), (14) and the fact
S
that { fi }i∈I ⊆ X we get
^ ^^  ^  ^ ^  ^
R(a, b) = cR(a) → cR(b) = fi (a) → fi (b) > fi (a) → fi (b) = fi (a) → fi (b)
c∈S c∈S i∈Ic i∈Ic c∈S i∈Ic i∈I
^
> f (a) → f (b) = Q(a, b).
f ∈X

Thus, we have proved that Q 6 R, and we conclude that Q is the smallest positive fuzzy quasi-order on S
having the cm-property.
18
 The following theorem gives the construction of a positive fuzzy quasi-order having the cm-property
starting from a collection of fuzzy filters.
Theorem 4.7. Let X be a collection of fuzzy filters of a semigroup S and let Q be a fuzzy relation on S defined by
^
Q(a, b) = f (b) → f (a), for all a, b ∈ S. (61)
f ∈X

Then Q is a positive fuzzy quasi-order on S having the cm-property such that X ⊆ CQ .

If X is the collection of all fuzzy filters of S, then Q is the smallest positive fuzzy quasi-order on S having the
cm-property.
Proof. In an analogous way as in Theorem 4.4 we conclude that Q is a fuzzy quasi-order.
Consider arbitrary a, b, c ∈ S. According to (12) we get
^  ^  ^   
Q(a, c) ∧ Q(b, c) = f (c) → f (a) ∧ f (c) → f (b) = f (c) → f (a) ∧ f (c) → f (b)
f ∈X f ∈X f ∈X
^   ^
= f (c) → f (a) ∧ f (b) = f (c) → f (ab) = Q(ab, c),
f ∈X f ∈X

and on the basis of Theorem 3.5 we have that Q is a positive fuzzy quasi-order with the cm-property.
Moreover, in an analogous way as in Theorem 4.4 we show that X ⊆ CQ , and if X is the collection of all
fuzzy filters of S, we show that Q is the smallest positive fuzzy quasi-order on S having the cm-property.
The last theorem of this section gives new characterizations of positive fuzzy quasi-orders having the
cp-property in terms of properties of the corresponding lattices of fuzzy ideals and consistent fuzzy subsets.
Theorem 4.8. Let Q be a positive quasi-fuzzy order on a semigroup S. Then the following statements are equivalent:
(i) Q has the cp-property;
(ii) IQ consists of completely prime fuzzy ideals;
(iii) CQ consists of fuzzy filters.
Proof. (i)⇔(ii). This follows directly from Theorems 3.6 and 4.1, and the fact that the set of all completely
prime fuzzy ideals is closed under linear combinations.
(ii)⇒(iii). Consider an arbitrary f ∈ CQ . By (59) it follows that f = Q\ f , and for any a ∈ S we get
^ ^
f (a) = (Q\ f )(a) = Q(b, a) → f (b) = bQ(a) → f (b).
b∈S b∈S

For each b ∈ S we have that bQ is a completely prime fuzzy ideal, and according to Theorem 3.7, a fuzzy
subset fb of S defined by fb (a) = bQ(a) → f (b) is a fuzzy filter. Therefore, f is the meet of a family of fuzzy
filters { fb }b∈S , and since the set of all fuzzy filters is closed under meets, we conclude that f is a fuzzy filter.
(iii)⇒(i). Let CQ consist of fuzzy filters. To prove that Q has the cp-property, we will prove that (40)
holds. By assumption, every Q-foreset is a fuzzy filter, and by Theorem 3.5 we obtain that (39) holds. On
the other hand, by Theorem 3.1 it follows that (33) also holds, and hence, to prove (40) it is sufficient to
prove inequality opposite to the inequality (33).
Consider arbitrary a, b ∈ S. Set f = Qa ∨ Qb. Since CQ is closed under joins we conclude that f ∈ CQ , and
thus, f is a fuzzy filter of S. Further,
^    
Q(c, ab) → f (c) = (Q\ f )(ab) = f (ab) = f (a) ∧ f (b) = Qa(a) ∨ Qb(a) ∧ Qa(b) ∨ Qb(b)
c∈S
   
= 1 ∨ Qb(a) ∧ Qa(b) ∨ 1 = 1 ∧ 1 = 1,
which means that Q(c, ab) → f (c) = 1, for every c ∈ S, that is, Q(c, ab) 6 f (c), for every c ∈ S. Therefore,
Q(c, ab) 6 f (c) = Qa(c) ∨ Qb(c) = Q(c, a) ∨ Q(c, b),
for all a, b, c ∈ S, which was to be proved.
19
5. Application to semilattice decompositions of semigroups

A fuzzy relation R on a semigroup S is compatible if for all a, b, c ∈ S we have that R(a, b) 6 R(ac, bc) and
R(a, b) 6 R(ca, cb). A compatible fuzzy equivalence is called a fuzzy congruence, and a compatible fuzzy quasi-
order is called a fuzzy half-congruence. Clearly, in the crisp case these notions come down to notions of a
congruence and a half-congruence. If Q is a fuzzy half-congruence on a semigroup S, it is easy to check that
its natural fuzzy equivalence EQ is a fuzzy congruence.
If E is a fuzzy congruence on a semigroup S, the binary operation on S can be transferred to the factor set
S/E by letting Ea Eb = Eab , for all a, b ∈ S. Because of the compatibility of E, that operation is well-defined and
S/E becomes a semigroup called the factor semigroup of S with respect to E. If Q is a fuzzy half-congruence,
in the same way we can define an operation on the set S/Q of all Q-aftersets, by letting (aQ)(bQ) = (ab)Q,
for all a, b ∈ S, and and operation on the set S\Q of all Q-foresets, by letting (Qa)(Qb) = Q(ab), for all
a, b ∈ S. However, we are essentially getting nothing new because, in light of Theorem 2.1, the afterset semi-
group S/Q, the foreset semigroup S\Q, and the factor semigroup S/EQ , are isomorphic to each other.
A fuzzy congruence E on a semigroup S is called a semilattice fuzzy congruence on S if the corresponding
factor semigroup S/E is a semilattice, i.e., an idempotent commutative semigroup. It is easy to verify that
this holds if and only if E(a2 , a) = 1, for each a ∈ S, and E(ab, ba) = 1, for all a, b ∈ S.
Analogously, a fuzzy half-congruence Q on a semigroup S is called a semilattice fuzzy half-congruence on
S if the corresponding afterset semigroup S/Q (resp. foreset semigoup S\Q, factor semigroup S/EQ ) is a
semilattice. In accordance with Theorem 2.1, this is true if and only if Q(a2 , a) = Q(a, a2 ) = 1, for each a ∈ S,
and Q(ab, ba) = 1, for all a, b ∈ S.
The following theorem shows the relationship between semilattice fuzzy congruences and positive
lower-potent fuzzy half-congruences.

Theorem 5.1. For any positive lower-potent fuzzy half-congruence on a semigroup S, its natural fuzzy equivalence
is a semilattice fuzzy congruence.
If the membership values are taken from a complete Heyting algebra, then the converse is also true, i.e., any
semilattice fuzzy congruence on S can be represented as the natural fuzzy equivalence of some positive lower-potent
fuzzy half-congruence.

Proof. Let Q be a positive lower-potent half-congruence on S and let E = Q ∧ Q−1 . For arbitrary a, b, c ∈ S se
have that

E(a, b) 6 Q(a, b) 6 Q(ac, bc) and E(a, b) = E(b, a) 6 Q(b, a) 6 Q(bc, ac),

and therefore, E(a, b) 6 Q(ac, bc) ∧ Q(bc, ac) = E(ac, bc). In the same way we show that E(a, b) 6 E(ca, cb).
Hence, E is a fuzzy congruence.
Next, for an arbitrary a ∈ S we have that Q(a2 , a) = 1, because of the lower-potency of Q, and Q(a, a2) = 1,
because of the positivity of Q. This means that E(a2 , a) = 1. Let us now consider arbitrary a, b ∈ S. Due to the
positivity of Q we get Q(ab, bab) = 1 and Q(bab, (ba)2 ) = 1, which yields Q(ab, (ba)2 ) = 1, because of the tran-
sitivity. Since Q((ba)2 , ba) = 1, we conclude that Q(ab, ba) = 1, which is what we wanted to prove. Therefore,
E is a semilattice fuzzy congruence.
Further, let the underlying complete residuated lattice L be a complete Heyting algebra, that is, let the
multiplication operation ⊗ and the meet operation ∧ coincide, and let E be an arbitrary semilattice fuzzy
congruence on S. Define a fuzzy relation Q on S as follows
_
Q(a, b) = E(ua, b), for all a, b ∈ S.
u∈S

We will prove that Q is a positive lower-potent half-congruence on S and E = Q ∧ Q−1 .

First, for each a ∈ S we have that
_
Q(a, a) = E(ua, a) > E(a2 , a) = 1,
u∈S

20
which means that Q is reflexive. Next, consider arbitrary a, b ∈ S. Due to the compatibility and transitivity
of E, for each pair u, v ∈ S we have that

E(ua, b) ∧ E(vb, c) 6 E(vua, vb) ∧ E(vb, c) 6 E(vua, c),

whence
_  _  __  _
Q(a, b) ∧ Q(b, c) = E(ua, b) ∧ E(vb, c) 6 E(ua, b) ∧ E(vb, c) 6 E(vua, c)
u∈S v∈S u∈S v∈S u,v∈S
_
6 E(wa, c) = Q(a, c),
w∈S

and we conclude that Q is transitive. Therefore, Q is a fuzzy quasi-order.

Further, for arbitrary a, b ∈ S we have that
_
Q(a, ab) = E(ua, ab) > E(ba, ab) = 1,
u∈S

whence Q(a, ab) = 1, and in a similar way we get Q(b, ab) = 1. Thus, Q is positive. Besides, for an arbitrary
a ∈ S we have
_
Q(a2 , a) = E(ua2 , a) > E(a3 , a) = 1,
u∈S

which means that Q is lower-potent.

Due to the compatibility of E, for arbitrary a, b, c ∈ S we have that
_ _
Q(a, b) = E(ua, b) 6 E(uac, bc) = Q(ac, bc),
u∈S u∈S

and similarly, Q(a, b) 6 Q(ca, cb). Hence, Q is a positive lower-potent fuzzy half-congruence on S.
To prove that Q ∧ Q−1 = E, consider arbitrary a, b ∈ S. First, we have that
_
E(a, b) = E(a2 , b) 6 E(ua, b) = Q(a, b),
u∈S

and analogously, E(a, b) = E(b, a) 6 Q(b, a), so E(a, b) 6 Q(a, b) ∧ Q(b, a).
On the other hand, due to the compatibility and transitivity of E, for each pair u, v ∈ S we have

E(ua, b) 6 E(vbua, vb2 ) = E(uvab, vb) and E(vb, a) 6 E(uavb, ua2 ) = E(uvab, ua),

whence

E(ua, b) ∧ E(vb, a) 6 E(uvab, vb) ∧ E(vb, a) 6 E(uvab, a),

E(ua, b) ∧ E(vb, a) 6 E(uvab, ua) ∧ E(ua, b) 6 E(uvab, b),

which yields

E(ua, b) ∧ E(vb, a) 6 E(uvab, a) ∧ E(uvab, b) = E(a, uvab) ∧ E(uvab, b) 6 E(a, b).

Now we have that

_  _  __ 
Q(a, b) ∧ Q(b, a) = E(ua, b) ∧ E(vb, a) 6 E(ua, b) ∧ E(vb, a) 6 E(a, b),
u∈S v∈S u∈S v∈S

and therefore, Q(a, b) ∧ Q(b, a) 6 E(a, b). Hence, Q ∧ Q−1 = E. This completes the proof of the theorem.
21
 As we noted above, every fuzzy quasi-order having the cm-property is lower-potent. The following theo-
rem explains in more detail the connection between cm-property and lower-potency.

Theorem 5.2. Every positive fuzzy quasi-order on a semigroup S which has the cm-property is a lower-potent fuzzy
half-congruence.
If the membership values are taken from a complete Heyting algebra, then the converse is also true, i.e., every
positive lower-potent fuzzy half-congruence on S has the cm-property.

Proof. Let Q be a positive fuzzy quasi-order on S having the cm-property. We can easily derive from (39)
that Q is lower-potent.
To prove the compatibility of Q consider arbitrary a, b, c ∈ S. According to (39) and (33) we have that

Q(a, b) 6 Q(a, b) ∨ Q(a, c) 6 Q(a, bc) = Q(a, bc) ∧ 1 = Q(a, bc) ∧ Q(b, bc) = Q(ac, bc),

and in a similar way we prove that Q(a, b) 6 Q(ca, cb). Thus, Q is a fuzzy half-congruence.
Conversely, let the underlying complete residuated lattice L be a complete Heyting algebra, and let Q
be a lower-potent fuzzy half-congruence on S. To prove that Q has the cm-proprety, let us consider arbitrary
a, b, c ∈ S. Due to the compatibility, lower-potency and transitivity of Q we have that

Q(a, c) 6 Q(ab, cb) and Q(b, c) 6 Q(cb, c2 ) = Q(cb, c2 ) ∧ Q(c2 , c) 6 Q(cb, c),

whence it follows that

Q(a, c) ∧ Q(b, c) 6 Q(ab, cb) ∧ Q(cb, c) 6 Q(ab, c).

Therefore, Q has the cm-property. This completes the proof of the theorem.
As an immediate consequence of the previous theorem we obtain the well-known result of T. Tamura [51]
according to which a positive (crisp) quasi-order on a semigroup has the cm-property if and only if it is a
lower-potent half-congruence.

Remark 5.3. The reason why in the converse parts of Theorems 5.1 and 5.2 we required that membership
values be taken from a Heyting algebra is that the concepts used in their proofs are defined by two different
types of fuzzy conjunction. Specifically, transitivity is defined by the multiplication operation in L (strong
conjunction), while cm-property and natural fuzzy equivalence are defined by the meet operation in L
(weak conjunction). This creates difficulties in proving the converse parts of Theorems 5.1 and 5.2, which
disappear when it is assumed that membership values are taken from a Heyting algebra, where these two
operations (conjunctions) coincide.
It should be pointed out that the use of the multiplication operation in defining the transitivity, as well
as the composition of fuzzy relations, is of enormous importance, given that many significant properties of
fuzzy relations are ensured by the distributivity of the multiplication over arbitrary joins.
On the other hand, in many sources dealing with fuzzy algebra, with complete residuated lattices as the
underlying structures of membership values, the concept of a fuzzy subalgebra was defined by means of
the multiplication operation (for instance, see [27]). However, such an approach would not be appropriate
here. The use of the meet operation is crucial in defining a consistent fuzzy subset, fuzzy subsemigroup and
fuzzy filter, because it provides a duality between consistent fuzzy subsets and fuzzy ideals, as well as
other related concepts considered in Theorems 3.7 and 3.8, and links semilattice decompositions of semi-
groups with fuzzy filters. In particular, this enables semilattice homomorphic images of a semigroup to
be characterized as semilattices of fuzzy filters which are principal elements in sublattices of the lattice of
fuzzy consistent subsets corresponding to positive fuzzy quasi-orders having the cm-property.

According to Theorem 3.1, the ordinary (crisp) division relation D, defined by (32), is the smallest positive
fuzzy quasi-order on a semigroup S. Moreover, Theorem 3.1 says that the set of all positive fuzzy quasi-
orders on S is the principal dual ideal of the lattice of fuzzy quasi-orders on S generated by D (or the principal
22
filter, as it is said in many sources). On the other hand, the crisp relation P : S → {0, 1} ⊆ L defined by

 1 if a = bn , for some n ∈ N,


P(a, b) =  (62)
 0 otherwise,


is the smallest fuzzy quasi-order on S having the cm-property. Indeed, for an arbitrary fuzzy quasi-order Q
on S we have that Q(bn , b) = 1, for all b ∈ S and n ∈ N, and if P(a, b) = 1, for some a, b ∈ S, that is, if a = bn ,
for some n ∈ N, then Q(a, b) = Q(bn , b) = 1, which means that P 6 Q.
As each fuzzy quasi-order with the cm-property is lower-potent, we have that P is lower-potent, but
it is clear that P is not the smallest lower-potent fuzzy quasi-order on S, since the smallest lower-potent
fuzzy quasi-order is obtained when in (62) we replace "a = bn , for some n ∈ N" with a = b2 .
T. Tamura in [48] defined a relation on a semigroup S that is the composition of the division relation D
and the relation P, namely

 1 if paq = bn , for some p, q ∈ S and n ∈ N,


(D ◦ P)(a, b) =  (63)
 0 otherwise,


and a quasi-order T = (D ◦ P)∞ , the transitive closure of D ◦ P. We will call D ◦ P Tamura’s relation, whereas T
will be called Tamura’s quasi-order. In the mentioned paper Tamura proved that, on an arbitrary semigroup S,
the natural equivalence of Tamura’s quasi-order T is the smallest semilattice congruence. He also noted that
T is the smallest positive lower-potent quasi-order on S, as well as the smallest positive lower-potent half
congruence (cf. [49–51]). It is important to note that the crisp part of an arbitrary positive lower-potent fuzzy
quasi-order is a positive lower-potent quasi-order. Since this quasi-order is contained in the original fuzzy
quasi-order, we can conclude that the smallest positive lower-potent fuzzy quasi-order on S is a crisp quasi-
order. Thus, Tamura’s quasi-order T is also the smallest positive lower-potent fuzzy quasi-order on S. As we
have already noted, Tamura in [51] proved that a positive quasi-order has the cm-property if and only if it is
a lower-potent half-congruence, which means that T is the smallest positive quasi-order on S having the cm-
property. Furthermore, T is the smallest positive fuzzy quasi-order on S having the cm-property. Similarly,
the smallest semilattice fuzzy congruence on S is a crisp congruence, and it coincides with the natural equiv-
alence of Tamura’s quasi-order T. These facts will be used in the proof of the following theorem:

Theorem 5.4. The smallest semilattice congruence E on a semigroup S can be represented as follows:
^ ^ ^
E(a, b) = f (a) ↔ f (b) = f (a) ↔ f (b) = f (a) ↔ f (b), for all a, b ∈ S, (64)
f ∈I cs f ∈I cp f ∈C f

where I cs denotes the collection of all completely semiprime fuzzy ideals, I cp the collection of all completely prime
fuzzy ideals, and C f the collection of all fuzzy filters of S.

Proof. Let E be the smallest semilattice congruence on S, and for an arbitrary X ⊆ F (S) let
^
F(a, b) = f (a) ↔ f (b).
f ∈X

Then F is the natural fuzzy equivalence of the fuzzy quasi-order Q defined as in (60) or (61).
When X = I cs , by Theorem 4.4 we obtain that Q is the smallest positive lower-potent fuzzy quasi-order
on S, that is, Q is equal to Tamura’s quasi-order T. This means that F = Q ∧ Q−1 = T ∧ T−1 = E.
If X = I cp , then by Theorem 4.6 we have that Q is the smallest positive fuzzy quasi-order on S having
the cm-property, and according to the above given remark, Q is contained in T, which means that F is
contained in E. In addition, by Theorems 5.1 and 5.2 we obtain that F is a semilattice fuzzy congruence,
and consequently, its crisp part is a semilattice congruence on S, so both the crisp part of F and F contain
E. Hence, we conclude that F = E.
In the same way we prove the case when X = C f .
23
 We also have that the following is true.

Theorem 5.5. Let the considered structure L of membership values be a linearly ordered complete residuated lattice.
Then every completely semiprime fuzzy ideal of a semigroup S is the intersection of some family of completely
prime fuzzy ideals of S.

Proof. As we have already shown above, Tamura’s quasi-order T is the smallest positive lower-potent fuzzy
quasi-order on S, and also, it is the smallest positive fuzzy quasi-order on S having the cm-property. Accord-
ing to Theorem 4.4, IT is the collection of all completely semiprime fuzzy ideals of S, and by Theorem 4.5
we obtain that every completely semiprime fuzzy ideal of S is the intersection of some family of completely
prime fuzzy ideals of S.

Let us note that the above theorem is an immediate generalization of the well-known theorem which
asserts that every completely semiprime ideal of a semigroup is an intersection of some family of completely
prime ideals of this semigroup. This theorem was proved in some special cases by Š. Schwarz [43] and K. Iséki
[31], and in the general case by M. Petrich [35]. The same result, without use of Zorn’s lemma arguments,
was proved by Ćirić and Bogdanović in [13, 14]. The related result in Theory of lattices is known as Prime
ideal theorem, and for the related results for rings we refer to W. Krull [32] and N. H. McCoy [34].

References
[1] W. Bandler, L.J. Kohout, Fuzzy relational products as a tool for analysis and synthesis of the behaviour of complex natural and
artificial systems, in: S.K. Wang, P.P. Chang (eds.), Fuzzy Sets: Theory and Application to Policy Analysis and Information
Systems, Plenum Press, New York, 1980, pp. 341–367.
[2] R. Bělohlávek, Fuzzy Relational Systems: Foundations and Principles, Kluwer, New York, 2002.
[3] R. Bělohlávek, V. Vychodil, Fuzzy Equational Logic, Springer, Berlin/Heidelberg, 2005.
[4] T. S. Blyth, Lattices and Ordered Algebraic Structures, Springer, London, 2005.
[5] U. Bodenhofer, A unified framework of opening and closure operators with respect to arbitrary fuzzy relations, Soft Computing
7 (2003) 220–227.
[6] U. Bodenhofer, B. De Baets, J. Fodor, A compendium of fuzzy weak orders: Representations and constructions, Fuzzy Sets and
Systems 158 (2007) 811–829.
[7] U. Bodenhofer, M. De Cock, E.E. Kerre, Openings and closures of fuzzy preorderings: theoretical basics and applications to fuzzy
rule-based systems, International Journal of General Systems 32 (2003) 343–360.
[8] S. Bogdanović, M. Ćirić, Positive quasi-orders with the common multiple property on a semigroup, in: Lj. D. Kočinac, ed., Proc.
of the Math. Conf. in Priština 1994, Pristina 1995, pp. 1–6.
[9] S. Bogdanović, M. Ćirić, Generation of positive lower-potent half-congruences, Southeast Asian Bulletin of Mathematics 21(6)
(1997) 227–231.
[10] S. Bogdanović, M. Ćirić, Quasi-orders and semilattice decompositions of semigroups, (A survey), in: Semigroups, Proceedings
of the International Conference in Semigroups and its Related Topics, Kunming, China, 1995, (K. P. Shum, Y. Guo, M. Ito and Y.
Fong, eds.) Springer-Verlag, 1998, pp. 27–56.
[11] S. Bogdanović, M. Ćirić, Ž. Popović, Semilattice Decompositions of Semigroups, University of Niš, Faculty of Economics, 2011.
[12] M. Ćirić, S. Bogdanović, Theory of greatest decompositions of semigroups (A survey), Filomat 9(3) (1995) 385–426.
[13] M. Ćirić, S. Bogdanović, Semilattice decompositions of semigroups, Semigroup Forum 52 (1996) 119–132.
[14] M. Ćirić, S. Bogdanović, The lattice of positive quasi-orders on a semigroup, Israel J. Math. 98 (1997) 157–166.
[15] M. Ćirić, S. Bogdanović, Fuzzy social network analysis, Godišnjak Učiteljskog Fakulteta u Vranju 1 (2010) 179–190.
[16] M. Ćirić, J. Ignjatović, S. Bogdanović, Fuzzy equivalence relations and their equivalence classes, Fuzzy Sets and Systems 158
(2007) 1295–1313.
[17] M. Ćirić, A. Stamenković, J. Ignjatović, T. Petković, Fuzzy relation equations and reduction of fuzzy automata, Journal of
Computer and System Sciences 76 (2010) 609–633.
[18] A. H. Clifford, Semigroups admitting relative inverses, Ann. of Math. (2) 42 (1941) 1037–1049.
[19] B. A. Davey, H. A. Priestley, Introduction to Lattices and Order, Cambridge Univ. Press, Cambridge, 1990.
[20] B. De Baets, Analytical solution methods for fuzzy relational equations, in: D. Dubois, H. Prade (eds.), Fundamentals of Fuzzy
Sets, The Handbooks of Fuzzy Sets Series, Vol. 1, Kluwer Academic Publishers, 2000, pp. 291–340.
[21] J. Fodor, M. Roubens, Fuzzy Preference Modelling and Multicriteria Decision Support, Kluwer, Dordrecht, 1994.
[22] J.S. Golan, Semirings and their Applications, Kluwer Academic Publishers, 1999.
[23] M. Gondran, M. Minoux, Graphs, Dioids and Semirings – New Models and Algorithms, Springer, Berlin, 2008.
[24] J. M. Howie, Fundamentals of Semigroup Theory, Clarendon Press, Oxford, 1995.
[25] J. Ignjatović, M. Ćirić, Weakly linear systems of fuzzy relation inequalities and their applications: A brief survey, Filomat 26(2)
(2012) 207–241.

24
[26] J. Ignjatović, M. Ćirić, Moore-Penrose equations in involutive residuated semigroups and involutive quantales, Filomat 31(2)
(2017) 183–196.
[27] J. Ignjatović, M. Ćirić, S. Bogdanović, Fuzzy homomorphisms of algebras, Fuzzy Sets and Systems 160 (2009) 2345–2365.
[28] J. Ignjatović, M. Ćirić, S. Bogdanović, On the greatest solutions to certain systems of fuzzy relation inequalities and equations,
Fuzzy Sets and Systems 161 (2010) 3081–3113.
[29] J. Ignjatović, M. Ćirić, B. Šešelja, A. Tepavčević, Fuzzy relational inequalities and equations, fuzzy quasi-orders, closures and
openings of fuzzy sets, Fuzzy Sets and Systems 260 (2015) 1–24.
[30] J. Ignjatović, M. Ćirić, V. Simović, Fuzzy relation equations and subsystems of fuzzy transition systems, Knowl.-Based Syst. 38
(2013) 48–61.
[31] K. Iséki, Contribution to the theory of semigroups IV, Proc. Japan Acad. Ser. A Math. Sci. 32 (1956) 430–435.
[32] W. Krull, Idealtheorie in Ringen ohne Endlichkeitsbedingung, Math. Ann. 101 (1929) 729–744.
[33] T.J. Li, Y. Leung, W.X. Zhang, Generalized fuzzy rough approximation operators based on fuzzy coverings, International Journal
of Approximate Reasoning 48 (2008) 836–856.
[34] N. H. McCoy, Theory of rings, McMillan, New York, 1970 (7th printing).
[35] M. Petrich, The maximal semilattice decomposition of a semigroup, Math. Z. 85 (1964) 68–82.
[36] M. Petrich, Introduction to semigroups, Merill, Ohio, 1973.
[37] M. S. Putcha, Semilattice decompositions of semigroups, Semigroup Forum 6 (1973) 12–34.
[38] M. S. Putcha, Positive quasi-orders on semigroups, Duke Math. J. 40 (1973) 857–869.
[39] M. S. Putcha, Minimal sequences in semigroups, Trans. Amer. Math. Soc. 189 (1974) 93–106.
[40] M. S. Putcha, Paths in graphs and minimal π-sequences in semigroups, Discrete Math. 11 (1975) 173–185.
[41] S. Roman, Lattices and Ordered Sets, Springer, New York, 2008.
[42] B. M. Schein, On certain classes of semigroups of binary relations, Siberian Math. J. 6 (1965) 616–635 (in Russian).
[43] S. Schwarz, On a Galois connections in the theory of characters of commutative semigroups, Czechoslovak Math. J. 4 (1956)
296–313 (in Russian).
[44] A. Stamenković, M. Ćirić, J. Ignjatović, Reduction of fuzzy automata by means of fuzzy quasi-orders, Inf. Sci. 275 (2014) 168–198.
[45] I. Stanković, M. Ćirić, J. Ignjatović, Fuzzy relation inequalities and equations with two unknowns and their applications, Fuzzy
Sets and Systems 322 (2017) 86–105.
[46] T. Tamura, The theory of construction of finite semigroups I, Osaka Math. J. 8 (1956) 243-–261.
[47] T. Tamura, On Putcha’s theorem concerning semilattice of Archimedean semigroups, Semigroup Forum 4 (1972) 83–86.
[48] T. Tamura, Note on the greatest semilattice decomposition of semigroups, Semigroup Forum 4 (1972) 255–261.
[49] T. Tamura, Semilattice congruences viewed from quasi-orders, Proc. Amer. Math. Soc. 41 (1973) 75–79.
[50] T. Tamura, Remark on the smallest semilattice congruence, Semigroup Forum 5 (1973) 277–282.
[51] T. Tamura, Quasi-orders, generalized archimedeaness and semilattice decompositions, Mathematische Nachrichten 68 (1975)
201–220.
[52] T. Tamura, N. Kimura, On decompositions of a commutative semigroup, Kodai Math. Sem. Rep. 4 (1954) 109-–112.
[53] L. Valverde, On the structure of F-indistinguishability operators, Fuzzy Sets Syst. 17 (3) (1985) 313–328.

25