Available online at www.sciencedirect.

com

ScienceDirect
Fuzzy Sets and Systems 386 (2020) 60–76
www.elsevier.com/locate/fss

Lattice-based sum of t-norms on bounded lattices

Moataz El-Zekey
Department of Engineering Mathematics, Physics and Science, Faculty of Engineering, Damietta University, 34517 New Damietta, Egypt
Received 7 May 2018; received in revised form 29 November 2018; accepted 9 January 2019
Available online 18 January 2019

Abstract
The concept of ordinal sums in the sense of Clifford have long been blamed for their limitations in constructing new t-norms in-
cluding inability to cope with general bounded lattices. Motivated by this observation, and based on the lattice-based sum of lattices
that has been recently introduced by El-Zekey et al., we propose a new sum-type construction of t-norms, called a lattice-based
sum of t-norms, for building new t-norms on bounded lattices from given ones. The proposed sum is generalizing the well-known
ordinal and horizontal sum constructions of t-norms by allowing for lattice ordered index sets. We demonstrate that, like the ordinal
sum of t-norms, the lattice-based sum of t-norms can be generalized using as summands so-called t-subnorms, still leading to a
t-norm. Subsequently, we apply the results for constructing several new families of t-norms and t-subnorms on bounded lattices. In
the same spirit, by the duality, we will also introduce lattice-based sums of t-conorms and t-subconorms.
© 2019 Elsevier B.V. All rights reserved.

Keywords: Lattice-based sum; Ordinal sum; Horizontal sum; Bounded lattice; Triangular norm and conorm; Triangular subnorm and subconorm

1. Introduction

Triangular norms (t-norms for short) on the unit interval were originally studied by Schweizer and Sklar (see e.g.
[35]) in the framework of probabilistic metric spaces. Later on, they turned out to be interpretations of the conjunction
in fuzzy logic and, subsequently, for the intersection of fuzzy sets. They have also proven to be useful in many other
fields like decision making, statistics as well as in the theories of non-additive measures and cooperative games. For a
comprehensive treatment on t-norms on the unite interval we refer to the monographs [4,24].
One of typical constructions for t-norms on the unit interval is the ordinal sum in the sense of Clifford [5]. It
provided a method to construct a new t-norm from a given system of t-norms indexed by a linearly ordered index
set. As shown in [18], this ordinal sum construction can be generalized using as summands so-called t-subnorms, still
leading to a t-norm. It is worth mentioning that ordinal sums are a construction method but also a representation tool
in the framework of associative conjunctions in many-valued logics and intersections in fuzzy set theory [8,13,18,
24–26,34].

E-mail address: moataz.elzekey@du.edu.eg.

https://doi.org/10.1016/j.fss.2019.01.006
0165-0114/© 2019 Elsevier B.V. All rights reserved.
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 61

Due to the close connection between order theory and fuzzy set theory, the unit interval was replaced by some more
general structure, for example, a bounded lattice (see e.g. [12]). Therefore, recently an increasing interest of t-norms
on bounded lattices can be observed, see e.g. [3,7,10,11,16,17,19,23,24,29,30,32,33,37] and many others.
Considering bounded lattices, the concept of ordinal sums in the sense of Clifford [5] have long been blamed for
their limitations in constructing new t-norms including inability to cope with general bounded lattices. On the one
hand, as observed in [32], in general, an ordinal sum like construction of t-norms may not work on bounded lattices.
In [32] (see also [29,33]) constraints on bounded lattices guaranteeing that the ordinal sum operation yields again a
t-norm on the lattices have been revealed. On the other hand, it is known that the ordinal sum of t-norms in the sense
of Clifford [5] whose carriers are bounded lattices is again a t-norm with a carrier equal to the ordinal sum (in the
sense of Birkhoff [1]) of the summand lattices. However, due to the ordinality (i.e. the linear order of the index set
involved), there exist ordinal sum like construction of t-norms yielding again a t-norm on bounded lattices which are
not an ordinal sum of some of their sublattices, see [32]. In summary, there is a need for a new sum-type construction
generalizing the ordinal sum construction of t-norms and cope very well with general bounded lattices.
Motivated by these observations, in this contribution we propose a new sum-type construction of t-norms for
building new t-norms on bounded lattices from given ones. We will call such a proposed sum-type construction a
lattice-based sum of t-norms. The proposed sum is based on the lattice-based sum of lattices that has been recently
introduced by El-Zekey et al. (see [9]), for building new (bounded) lattice ordered sets from fixed ones. The lattice-
based sum construction of t-norms is generalizing the well-known ordinal sum constructions of t-norms by allowing
for lattice ordered index sets instead of linearly ordered index set. It extends also the horizontal sum based on an un-
structured index set (i.e., any two distinct indices are incomparable), see [9]. We demonstrate that, like the ordinal sum
of t-norms (see [18]), the lattice-based sum of t-norms can be generalized using as summands so-called t-subnorms,
still leading to a t-norm. Subsequently, we apply the results for constructing several new families of t-norms and t-
subnorms on bounded lattices. In the same spirit, by the duality, we will also introduce lattice-based sums of t-conorms
and t-subconorms showing that all results given for t-norms (t-subnorms) remain valid for t-conorms (t-subconorms)
with the obvious changes where necessary.
This paper is organized as follows. The next section is a preliminary that includes the basic concepts and definitions
required in this paper. In Section 3 we briefly recall the lattice-based sum construction of lattice ordered sets. In
Section 4, we introduce the lattice-based sum of t-norms and t-subnorms. In Section 5, we apply the latticed-based sum
theorems, from Section 4, for constructing several new t-norms and t-subnorms on bounded lattices. The lattice-based
sums of t-conorms and t-subconorms are presented in Section 6. We close this contribution by a short summary and
further perspectives.

2. Preliminaries

Recall that [1,6] a lattice-ordered set is a partially ordered set (L, L ) in which each two-element subset {x, y}
has an infimum, denoted x ∧ y, and a supremum, denoted x ∨ y. A bounded lattice (L, L ) is a lattice-ordered set
which has the top and bottom elements which are written as L and ⊥L , respectively, that is, ⊥L L x L L , for
all x ∈ L. If x, y ∈ L such that x L y but x = y, then we will write x ≺L y.

Definition 1 (see [7,24]). Let (L, L , ⊥L , L ) be a bounded lattice. A binary operation T : L2 −→ L is called a
triangular norm (t-norm for short) if it is commutative, associative, non-decreasing in both arguments and it satisfies
the boundary condition T (x, L ) = x for all x ∈ L, i.e. L is the neutral element of T .

Example 1 (see [24]). Given a bounded lattice (L, L , ⊥L , L ), there are at least two t-norms on L, the meet t-norm
TML (x, y) = x ∧ y, and the drastic product t-norm TDL defined by

x ∧ y, if L ∈ {x, y}
TD (x, y) =
L
⊥L , otherwise
which are also the greatest and smallest t-norms on the lattice L, i.e. for any t-norm T on L, we have TDL (x, y) L
T (x, y) L x ∧ y for all x, y ∈ L.

Naturally, the concept of t-subnorm on the unit interval (see [24,31]) may be extended to lattices.
62 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

Definition 2. A binary operation F : L2 −→ L is called a t-subnorm if it is commutative, associative, non-decreasing

in both arguments and it satisfies the range condition F (x, y) L x ∧ y for all x, y ∈ L.

It is clear that every t-norm is a t-subnorm. However, in general terms, the reciprocal of this affirmation is not
true (since L need not be the neutral element of a t-subnorm F ). For instance, the trivial t-subnorm defined by
T (x, y) = ⊥L for all x, y ∈ L, is not a t-norm. Note that, like any t-norm, for any t-subnorm F on the bounded lattice
L, ⊥L is the annihilator of F , i.e. F (⊥L , x) = ⊥L for all x ∈ L. Among many others the following are examples of
t-subnorms on arbitrary bounded lattice L:

Example 2. Consider a bounded lattice (L, L , ⊥L , L ) and fix an a ∈ L. It is easy to see that the functions given
by, for all x, y ∈ L

F1 (x, y) = x ∧ y ∧ a (1)

a, if (x, y) ∈ [a, L ]2
F2 (x, y) = (2)
⊥L , otherwise

are t-subnorms on L, where [a, L ] = {x : a L x L L }.

Note that, by putting a = ⊥L in (1) and (2), we have that F1 and F2 coincide with the trivial t-subnorm, i.e.
F1 (x, y) = F2 (x, y) = ⊥L for all x, y ∈ L. The following t-subnorm, by putting a = L , is a particular case of (2):

L , if x = y = L
F3 (x, y) = (3)
⊥L , otherwise

It is always possible to construct a t-norm from a t-subnorm as shown in the following proposition.

Proposition 1 ([24,31]). If F is a t-subnorm on bounded lattice L, then T defined by


F (x, y), if (x, y) ∈ (L \ {L })2
T (x, y) =
x ∧ y, otherwise

is a t-norm on L.

By applying the above proposition on the t-subnorms given by (1) and (2), we get the following t-norm operations
from [3,10].

Example 3. Consider a bounded lattice (L, L , ⊥L , L ) and fix an a ∈ L. Then the functions given by, for all x, y ∈ L

x ∧ y, if L ∈ {x, y}
T1 (x, y) = (4)
x ∧ y ∧ a, otherwise
⎧
⎪
⎨a, if (x, y) ∈ [a, L [2
T2 (x, y) = x ∧ y, if L ∈ {x, y} (5)
⎪
⎩
⊥L , otherwise

are t-norms on L, where [a, L [= {x : a L x ≺L L }.

Note that, by putting a = ⊥L in (4) and (5), we have that T1 and T2 coincide with the drastic t-norm TDL (x, y), i.e.
T1 (x, y) = T2 (x, y) = TDL (x, y) for all x, y ∈ L.
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 63

3. Lattice-based sum of bounded lattices

In this section we recall some definitions and results that will be necessary for our work. Some details will be
omitted, but an appropriate description of them can be found in [9]
In the sequel, (, ) denotes a (bounded) lattice-ordered index set. The top and bottom elements of  will be
denoted by  and ⊥ , respectively. (Lα , ∧α , ∨α ) denotes a bounded lattice-ordered set for some α ∈ , where ∧α
and ∨α are the meet and join operations on Lα , respectively. The top and bottom elements of Lα will be denoted
by α and ⊥α , respectively. Lowercase Latin letters (e.g. “x”, “y” and “z”) will be used as variables ranging over
the elements of Lα , and lowercase Greek letters (e.g. “α”, “β” and “γ ”) will be used as variables ranging over the
elements of . As usual, the partial order relation α on a lattice Lα is defined by x α y if and only if x ∧α y = x.
If α, β ∈  are incomparable elements, then we will write α  β. If α, β ∈  such that α  β but α = β, then we will
write α < β. The cardinality of a set A will be denoted by |A|.

Definition 3 ([9]). Consider a (bounded) lattice-ordered index set (, ). The -sum family is a family of bounded
lattices {(Lα , ∧α , ∨α )}α∈ that satisfies for all α, β ∈  with α = β the sets Lα and Lβ are either disjoint or satisfy
one of the following two conditions:
 
(i) Lα ∩ Lβ = xαβ with α < β, where xαβ is both  thetop element of Lα and the bottom element of Lβ , and where
for each
 ε ∈  with α <
  ε < β we have L ε = xαβ , also for all δ, γ ∈  with δ  γ , δ < β and α < γ we have
Lδ = yδγ or Lγ = zδγ , where yδγ is the top element of Linf{δ,γ } and zδγ is the bottom element of Lsup{δ,γ } .
(ii) 1 ≤ Lα ∩ Lβ ≤ 2 with α  β, and for each xαβ ∈ Lα ∩ Lβ , xαβ is the top element of both Lα and Lβ and the
bottom element of Lsup{α,β} , or xαβ is the bottom element of both Lα and Lβ and the top element of Linf{α,β} .

If necessary, we refer to this kind of a -sum family as the -sum family of bounded lattices.

Definition 4 ([9]). Given a -sum family {(Lα , α )}α∈ and x ∈ α∈ Lα . We say that an element α ∗ ∈  is a
maximal index of x (respectively, α∗ ∈  is a minimal index of x) if α ∗ is a maximal (respectively, minimal) element
of the set Ix = {α ∈  | x ∈ Lα }. Denote by Ixmax and Ixmin the set of all maximal and minimal indices of x, respectively.

Obviously, if {(Lα , α )}α∈ is a -sum family with finite lattice index set , then, for all x ∈ α∈ Lα , the
set Ix = {α ∈  : x ∈ Lα } contains maximal and minimal elements. Note that, in general, the set Ix need not have
maximal or minimal elements (see [9, Example 5]).

Given a -sum family {(Lα , α )}α∈ . Let x, y ∈ α∈ Lα with x = y. If for all α, β ∈  such that x ∈ Lα and
y ∈ Lβ we have α  β, then we will write x  y.

Definition 5 ([9]). A -sum family {(Lα , α )}α∈ is said to be with (M) condition if it satisfies the following (M)
condition:
 
(M): for all x, y ∈ α∈ Lα with x  y, the sets Ix = {δ ∈  : x ∈ Lδ } and Iy = β ∈  : y ∈ Lβ have both maxi-
mal and minimal elements.

The following are examples of -sum families with (M) condition:

(i) A -sum family with index set  that has

no infinite chains, e.g., finite index sets.

(ii) A -sum family satisfying, for all x, y ∈ α∈ Lα with x  y, that both Ix and Iy are finite, e.g., a family of
pairwise disjoint lattices.

(iii) A -sum family satisfying, for all x, y ∈ α∈ Lα with x  y, that every chain in Ix and Iy has an upper and
lower bound in Ix and Iy , respectively.

Definition 6 ([9]). Let (, ) be a (bounded) lattice-ordered

index set and let {(Lα , ∧α , ∨α )}α∈ be a -sum family
of bounded lattices with (M) condition. Put L = α∈ Lα and define the binary operations ∧ and ∨ on L by
64 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

⎧
⎪
⎪ x ∧α y if (x, y) ∈ Lα × Lα ,
⎨
x if (x, y) ∈ Lα × Lβ and α < β,
x∧y = (6)
⎪y
⎪ if (x, y) ∈ Lα × Lβ and β < α,
⎩
inf{α ∗ ,β ∗ } if x  y, α ∗ ∈ Ixmax and β ∗ ∈ Iymax .
⎧
⎪
⎪ x ∨α y if (x, y) ∈ Lα × Lα ,
⎨y if (x, y) ∈ Lα × Lβ and α < β,
x∨y = (7)
⎪
⎪ x if (x, y) ∈ Lα × Lβ and β < α,
⎩
⊥sup{α∗ ,β∗ } if x  y, α∗ ∈ Ixmin and β∗ ∈ Iymin .
Then we say that (L, ∧, ∨) is the lattice-based sums of all {(Lα , ∧α , ∨α )}
α∈ . If necessary, we refer to this type of
lattice-based sum as lattice-based sums of bounded lattices and denoted as α∈ (Lα , ∧α , ∨α ).

Theorem 2 ([9]). With all the assumptions of Definition 6 the lattice-based sum (L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ) is a
(bounded) lattice.

Remark 1. The lattice-based sum L = α∈ Lα with bounded index set  is bounded where the top and the bottom
elements of L are ⊥L = ⊥⊥ and L =  , respectively.

Remark 2. The assumption of (M) condition in Definition 6 and Theorem 2 above is indispensable for the validity of
the mentioned results as shown in [9] (see also Example 5).

Lemma 3 ([9]). Given a lattice-based sum (L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Then the partial order relation L on
the lattice L obtained by setting x L y in L iff x ∧ y = x coincides with the partial order relation given by
⎧
⎪
⎨∃α ∈  such that x, y ∈ Lα and x α y
x L y if and only if or (8)
⎪
⎩
∃α, β ∈  such that (x, y) ∈ Lα × Lβ and α < β

One obtains the same partial order relation from the given lattice by setting x L y in L if and only if x ∨ y = y.

Remark 3. Notice that the relation  is not equivalent to the incomparability relation in general. Obviously, x and y
are incomparable (denoted by x  y as usual) with respect to L if x  y or x, y ∈ Lα such that x α y and y α x
for some α ∈ . Hence, in general, x  y implies x  y but not vice versa. In case that there is no α ∈  such that
{x, y} ⊆ Lα , x  y if and only if x  y.

We can easily check that, if the index set is linear, then the lattice-based sum reduces to the ordinal sum, i.e., the
ordinal sum in the sense of Birkhoff [1] is a particular case of the more general approach of lattice-based sum. We point
out that, as shown in [9], the lattice-based sum approach extends also the horizontal sum based on an unstructured
index set (i.e., any two distinct indices are incomparable).
Note that the strategy described in this section focuses on the union of the carriers and an order consistent (i.e.,
in agreement) with both the order of the underlying lattices and the order of the lattice-ordered index set (see (8)
in Remark 3). Accordingly, the order relation for elements from different summand carriers is inherited from the
lattice-ordered index set.
We end this section by showing some examples to clarify the ideas (more examples can be found in [9]).

Example 4. Consider the lattice-ordered index set in Fig. 1. The family associated with the structure in Fig. 2 is not
a -sum family because Lα ∩ Lβ = {xαβ }, with xαβ = α = ⊥β , δ < β,  α < γ , but neither Lδ = {inf{δ,γ } } nor
Lγ = {⊥sup{δ,γ } }. Hence the structure in Fig. 2 is not the lattice-based sum α∈ Lα . Note that, its order relation is
not consistent with the order of the index set (since, for x ∈ Lδ and y ∈ Lγ , x L y while the only elements δ and γ
in the index set associated with x and y, respectively, are incomparable elements in ). A slight modification of the
family associated with the structure in Fig. 2 by putting Lδ = {inf{δ,γ } } produces the -sum family associated with
the structure in Fig. 3. Note that in this case the consistency holds, namely for x and y as above, we have x L y,
x ∈ Lδ ∩ L⊥ and y ∈ Lγ , and hence there exists ⊥ , γ ∈  associated with x and y, respectively, such that ⊥ < γ .
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 65

Fig. 1. Lattice (, ).

Fig. 2. Not family.

Fig. 3. Family.

Example 5. Consider the lattice-ordered index set (, ) in Fig. 4. It is easy to check that the family associated with
the structures in Fig. 5 is a -sum family and satisfies the condition (M) (since the index set is finite). Moreover, it
is easy to see that the structures in Fig. 5 is a bounded lattice where the meet and join are defined as in (6) and (7),
respectively. For the reader convenience, we check only the case when we have two incomparable elements. Therefore,
consider x  y where x = β2 and y = α2 . Obviously, Ix = {β2 , δ2 } and Iy = {α2 , α3 , δ3 }. Hence, Ixmax = {δ2 } and
Iymax = {α3 , δ3 }. For computing x ∧ y, the last case in (6) applies. By noting that Lδ1 = {α1 }, we find that, for each
α ∗ ∈ Ixmax and β ∗ ∈ Iymax , x ∧ y = inf{α ∗ ,β ∗ } = α1 and hence it produces the same result (as already proved in [9,
Lemma 4, Corollary 1]). Note that if α ∗ ∈ Ix is not maximal or β ∗ ∈ Iy is not maximal, x ∧ y need not be equal to
inf{α ∗ ,β ∗ } , e.g., for β2 ∈ Ix and α2 ∈ Iy , we have inf{β2 ,δ3 } = β1 = x ∧ y.
66 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

Fig. 4. Lattice (, ).

Fig. 5. Family.

4. Lattice-based sum of t-norms on bounded lattices

Lemma 4. Consider a lattice-ordered index set (, ) and a lattice-based sum of bounded lattices (L, ∧, ∨) =

α∈ (Lα , ∧α , ∨α ). Assume that there exists x1 , x2 ∈ L such that there is no α ∈  such that {x1 , x2 } ⊆ Lα . Then

(i) If x1 ≺L x2 , then there exists α1 , α2 ∈  such that (x1 , x2 ) ∈ (Lα1 , Lα2 ) with α1 < α2 , and for all (z1 , z2 ) ∈
(Lα1 , Lα2 ) we have z1 L z2 .
(ii) If x1  x2 , then for all α1 ∈ Ix1 and α2 ∈ Ix2 , α1  α2 , furthermore for all z1 ∈ Lα1 \{α1 , ⊥α1 } and z2 ∈
Lα2 \{α2 , ⊥α2 } we have z1  z2 .
(iii) If x1  x2 , then necessarily x1 ∧ x2 ∈ {⊥α , α } for some α ∈ .

Proof. Having in mind that the sets Lα \{α , ⊥α } are pairwise disjoint (see Definition 3) and the overlapping is only
allowed in the boundaries (⊥α or α ). Hence, (i) and (ii) are direct consequences from Lemma 3, and (iii) follows
immediately by noting that the last case in (6) applies for computing x1 ∧ x2 .

Note that excluding the boundaries from the sets of incomparable elements, in Lemma 4 (ii) above, is necessary for
the validity of the mentioned result. For example, in the lattice-based sum in Fig. 5, we have that, for all x ∈ Lα1 \{α1 }
and for all y ∈ Lβ2 \{β2 }, x  y while α1 L β2 .
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 67

Definition 7. Consider a (bounded) lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Let (Tα )α∈ be a family of t-subnorms on the corresponding summand lattices
(Lα )α∈ that satisfies for each α, β ∈  with α < β and α = ⊥β we have that Tα is a t-norm. Furthermore, for each
α, β ∈  with α  β we have that Tinf{α,β} is also a t-norm. The lattice-based sum of the family (Tα )α∈ of t-subnorms
is the function T : L2 −→ L given by

Tα (x, y), if (x, y) ∈ Lα × Lα
T (x, y) = (9)
x ∧ y, otherwise,
and denoted as T = {(Lα , Tα )}α∈ . If necessary, we refer to this type of sum as the lattice-based sum of t-subnorms
while the lattice-based sum of t-norms for that whose all underlying operations Tα are t-norms.

Remark 4. Observe that, by Definition 6 and Theorem 2, the operation T considered in Definition 7 can be described
alternatively as
⎧
⎪
⎪ Tα (x, y) if (x, y) ∈ Lα × Lα ,
⎨
x if (x, y) ∈ Lα × Lβ and α < β,
T (x, y) =
⎪
⎪ y if (x, y) ∈ Lα × Lβ and β < α,
⎩
inf{α ∗ ,β ∗ } if x  y, α ∗ ∈ Ixmax and β ∗ ∈ Iymax .

The following lemma is a direct consequence from Definition 7, the range condition of t-subnorms and the bound-
ary condition of t-norms (see Definition 1 and Definition 2).

Lemma 5. With all the assumptions of Definition 7 the operation T defined in equation (9) satisfies the following
basic properties:

(i) For all x, y ∈ L, T (x, y) L x ∧ y.

(ii) If x = ⊥α for some α ∈ , then, for all y ∈ L, T (x, y) = x ∧ y.
(iii) If x = α , for some α ∈ , and the operation Tα associated with Lα is a t-norm then, for all y ∈ L, T (x, y) =
x ∧ y.

Remark 5. The first condition in Definition 7 imposed on the family of t-subnorms involved in the sum is required for
the operation T given by (9) to be well-defined (as we will see in the proof of Theorem 9 below). On the other hand,
the second condition in Definition 7 is indispensable for the associativity of the operation T . For example, consider
the lattice-based sum in Fig. 5 and let x ∈ Lα1 \{α1 }, y ∈ Lβ2 \{β2 } and z ∈ L⊥ \{⊥ } with T⊥ (⊥ , z) = z,
then we obtain T (T (x, y), z) = T⊥ (⊥ , z) = z = T (x, T (y, z)). Hence, associativity of T is not fulfilled in this
case.

Theorem 6 (Lattice-based sum theorem for t-subnorms). With all the assumptions of Definition 7 the lattice-based
sum of t-subnorms T = {(Lα , Tα )}α∈ as defined in equation (9) is a t-subnorm on L.

Proof. First it is necessary to check that the operation T is well defined. A problem can only arise if (x, y) ∈ Lα × Lβ
with x ∈ Lα ∩ Lβ . In this situation the two cases in (9) apply. We distinguish the following cases:

(i) If α < β, then (by Definition 3 (i)) we know that x = α = ⊥β and hence (by Lemma 5 (ii)), we get T (x, y) =
x ∧ y, thus producing the same result in either case.
(ii) If β < α, then (from Definition 3 (i) and by using the first condition, in Definition 7, imposed on the family of
t-subnorms involved in the sum) we have that x = β = ⊥α and Tβ is a t-norm. Hence (by Lemma 5 (iii)), we
get T (x, y) = x ∧ y, thus producing the same result in either case.
(iii) If α  β, then (by Definition 3 (ii)) we know that x = α = β = ⊥sup{α,β} or x = ⊥α = ⊥β = inf{α,β} . These
situations are checked in complete analogy by using Lemma 5 (ii), Lemma 5 (iii) and the first condition in
Definition 7.
68 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

We will now prove that the operation T is a t-subnorm on L. The range condition has been already proved in
Lemma 5 (i). Commutativity holds due to the commutativity of ∧ and Tα for each α ∈ .
Monotonicity: By commutativity, it suffices to show that if x L y in L, then for all z ∈ L we have T (x, z) L
T (y, z). The proof is split into the following two cases:

Case 1: Suppose that there exist some α ∈  such that {x, y} ⊆ Lα . If z ∈ Lα , then monotonicity holds due to the
monotonicity of Tα on Lα . If z ∈ / Lα , then T (x, z) = x ∧ z L y ∧ z = T (y, z).
Case 2: Suppose that there is no α ∈  such that {x, y} ⊆ Lα . Then it follows necessarily that x = y and hence
x ≺L y. Hence, by Lemma 4 (i), it follows that there exists α1 , α2 ∈  such that (x, y) ∈ (Lα1 , Lα2 ) with
α1 < α2 . Consequently, we have one of the following subcases:
– If z ∈ Lα1 (i.e., x and z are in the same summand), then (by Lemma 5 (i)) we have T (x, z) L x ∧ z L
y ∧ z = T (y, z).
– If z ∈ Lα2 (i.e., y and z are in the same summand), then T (x, z) = x ∧ z L x ≺L T (y, z) (by Lemma 4 (i)
and since T (y, z) ∈ Lα2 , x ∈ Lα1 and α1 < α2 ).
– If z ∈
/ Lα1 ∪ Lα2 , then T (x, z) = x ∧ z L y ∧ z = T (y, z).

Associativity: We have to show that T (x, T (y, z)) = T (T (x, y), z) for all x, y, z ∈ L. We will prove this with the help
of the following claim.

Claim. If α ∈ Lα ∩ Lβ for some α, β ∈  such that α = β, then necessarily Tα is a t-norm on Lα (i.e. if α is an

intersection point, then Tα is a t-norm).

Proof of claim. We have the following cases:

– If α < β, then (by Definition 3 (i)) we have that α = ⊥β , and hence (by the first condition in Definition 7) we
have that Tα is a t-norm.
– If β < α, then (by Definition 3 (i)) we have that α = β = ⊥α . Therefore | Lα |= 1, i.e. Lα is a singleton, and
consequently Tα is the trivial t-norm.
– If α  β, then (by Definition 3 (ii)) we have that α = β = ⊥sup{α,β} or α = ⊥α = ⊥β . These situations are
checked in complete analogy as before. This completes the proof of the claim.

Associativity holds trivially if all arguments are from the same summand (due to the associativity of the t-subnorm
defined on this summand). For the remaining possibilities we distinguish the following cases:

Case 1: Suppose that there are no two arguments are from the same summand. We assume x ∈ Lα , for some α ∈ ,
and y, z ∈/ Lα (remembering that y and z are also not in the same summand). If y ∧ z ∈ Lα , then necessarily
y  z. Hence (by Lemma 4 (iii)) y ∧ z ∈ {⊥α , α }. In case that y ∧ z = α , then (by applying the last case
in (6)), we have that α = y ∧ z = inf{β,γ } where β ∈ Iymax , γ ∈ Izmax and β  γ . So, if α = inf{β, γ }, then
necessarily Tα is a t-norm (by the second condition, in Definition 7, imposed on the family of t-subnorms
involved in the sum). If α = inf{β, γ }, then α = y ∧ z ∈ Lα ∩ Linf{β,γ } . Hence (by using the claim) Tα is a
t-norm. All this together (using Lemma 5 (ii) and (iii)) implies that the associativity holds trivially due to the
associativity of ∧. Indeed T (x, T (y, z)) = T (x, y ∧ z) = x ∧ (y ∧ z) = (x ∧ y) ∧ z = T (T (x, y), z).
All other cases can be shown analogously or are fulfilled due to the associativity of ∧.
Case 2: Suppose that exactly two arguments involved are from the same summand. We choose x, y ∈ Lα for some
α ∈  and z ∈ / Lα . It can be easily verified (using Lemma 5 (ii)) that in case that either x = ⊥α or y = ⊥α
associativity of T is fulfilled due to the associativity of ∧. In case that x = α and x is an intersection point,
i.e. x ∈ Lα ∩ Lβ for some α, β ∈  such that α = β, we have that (by the claim) Tα is t-norm. Therefore,
(using Lemma 5 (iii)) associativity of T is fulfilled due to the associativity of ∧. The case with y = α and
y is an intersection point can be shown analogously.
Therefore, we assume that {x, y} ⊆ Lα \ {⊥α } and α is not an intersection point (i.e. Iα = {α}). Conse-
quently, we have that for all u ∈ Lα \ {⊥α }, Iu = {α}. We distinguish the following subcases:
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 69

– If x ≺L z or y ≺L z, then (by Lemma 4 (i)) we have that α < δ for some δ ∈  such that z ∈ Lδ , and for
all u ∈ Lα , u ≺L z. Moreover,

T (x, T (y, z)) = T (x, y ∧ z) = T (x, y)

= Tα (x, y) = Tα (x, y) ∧ z
= T (Tα (x, y), z) = T (T (x, y), z).
– If z ≺L x or z ≺L y, then (by Lemma 4 (i)) we have that δ < α for some δ ∈  such that z ∈ Lδ , and for
all u ∈ Lα , z ≺L u. Moreover,

T (x, T (y, z)) = T (x, y ∧ z) = T (x, z)

=x ∧z=z
= Tα (x, y) ∧ z = T (Tα (x, y), z)
= T (T (x, y), z).
– If x  z or y  z, then (by Lemma 4 (ii)) we have that δ  α for all δ ∈ Iz , and for all u ∈ Lα \ {⊥α }, we
have z  u. Moreover, it holds that u ∧ z = ⊥α ∧ z for all u ∈ Lα and hence, we have

T (x, T (y, z)) = T (x, y ∧ z) = x ∧ ⊥α ∧ z = ⊥α ∧ z

= Tα (x, y) ∧ z = T (Tα (x, y), z)
= T (T (x, y), z).

The remaining possibilities can be shown analogously or are fulfilled due to the commutativity of T and using Case 2
such that associativity is proven.

Corollary 1 (Generalized lattice-based sum theorem for t-norms).

 Consider a bounded lattice-ordered index set
(, ) and a lattice-based sum of bounded lattices (L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Suppose that (Tα )α∈ is a fam-
ily of t-subnorms on the corresponding summand lattices (Lα )α∈ where for each α, β ∈  with α < β and α = ⊥β
we have that Tα is a t-norm. Furthermore, suppose that, for each α, β ∈  with α  β we have that Tinf{α,β} is also a
t-norm, and for L we have that T is a t-norm. Let T be a function defined on L by (9). Then T is a t-norm.

Proof. Only the boundary condition of t-norms has to be verified which follows immediately from the last condition.

 Consider a bounded lattice-ordered index set (, ) and a lattice-

Corollary 2 (Lattice sum theorem for t-norms).
based sum of bounded lattices (L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Suppose that (Tα )α∈ is a family of t-norms on the
corresponding summand lattices (Lα )α∈ . Let T be a function defined on L by (9). Then T is a t-norm.

Proof. Since all the operations involved in the sum are t-norms and every t-norm is a t-subnorm, then the result
follows immediately from Corollary 1.

In summary, the lattice-based sum theorems introduced in this section say that the lattice-based sum of t-subnorms
is again a t-subnorm, and it produces a t-norm if we assume that the last summand (i.e. T ) is a t-norm. Then, finally,
the lattice-based sum of t-norms is again a t-norm.

5. Lattice-based sum construction of t-norms on bounded lattices

The aim of this section is to apply the latticed-based sum theorems, from Section 4, for constructing several new
t-norms and t-subnorms on bounded lattices. Perhaps, the most important consequence is that a new construction
method for t-norms on bounded lattices arise.
70 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

Proposition 7. Consider a bounded lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Fix an aα ∈ Lα for each α ∈ . Then the following operation is a t-norm on L.

x ∧ y ∧ aα , if (x, y) ∈ (Lα \{α })2
T (x, y) =
x ∧ y, otherwise.

Proof. With each Lα , α ∈ , associate the t-norm Tα given by (4), i.e. for all x, y ∈ L

x ∧ y, if α ∈ {x, y}
Tα (x, y) =
x ∧ y ∧ aα , otherwise.
Hence the result follows immediately by using Corollary 2 and noting that T is the lattice-based sum of the above
mentioned family (Tα )α∈ of t-norms.

Corollary 3. Consider a bounded lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Then the following operation is a t-norm on L.

⊥α , if (x, y) ∈ (Lα \{α })2
Td (x, y) =
x ∧ y, otherwise.

Proof. It is a direct consequence of Proposition 7 by putting aα = ⊥α . Another way to see this by noting that Td is
the lattice-based sum of the drastic t-norms TDLα , i.e., Td = {(Lα , TDLα )}α∈ .

Since the drastic t-norms TDLα is the smallest t-norm on Lα , then the t-norm Td = {(Lα , TDLα )}α∈ is the smallest
t-norm on L that can be obtained as a lattice-based sum of a family (Tα )α∈ of t-norms. However, in general, Td is
not the smallest t-norm on L that can be obtained as a lattice-based sum of a family (Tα )α∈ of t-subnorms as shown
in the following example.

Example 6. Consider the lattice-ordered index set (, ) in Fig. 4 and the -family of bounded lattices associ-
ated with the bounded lattice in Fig. 5. Consider the following lattice-based sum T = (Lα , Tα )α∈ of t-subnorms
where T is the drastic t-norm on L for each  ∈ {⊥ , α1 , α2 , β2 }, T is the trivial t-subnorm on L for each
 ∈ {δ2 , δ3 , α3 , α4 , β4 }, and the function T is given by
⎧
⎪
⎨⊥ , if (x, y) ∈ (L \ { })2 and  ∈ {⊥ , α1 , α2 , β2 }
T (x, y) = ⊥ , if (x, y) ∈ (L )2 and  ∈ {δ2 , δ3 , α3 , α4 , β4 }
⎪
⎩
x ∧ y, otherwise.
Then, by Corollary 1, T is a t-norm and it is the smallest t-norm obtained as a lattice-based sum of t-subnorms on
the corresponding -sum family (Lα )α∈ associated with the bounded lattice in Fig. 5. Observe that T = Td and
T (x, y) L Td (x, y) for all x, y ∈ L.

Proposition 8. Consider a bounded lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Fix an aα ∈ Lα for each α ∈ . Then the following operation is a t-norm on L.
⎧
⎨ aα , if (x, y) ∈ ([aα , α [)2
T (x, y) = ⊥α , if (x, y) ∈ (Lα \{α })2 \([aα , α [)2
⎩
x ∧ y, otherwise.

Proof. With each Lα , α ∈ , associate the t-norm Tα given by (5), i.e. for all x, y ∈ L
⎧
⎪
⎨aα , if (x, y) ∈ [aα , α [2
Tα (x, y) = x ∧ y, if α ∈ {x, y}
⎪
⎩
⊥α , otherwise
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 71

Fig. 6. The lattice-ordered index set (, ).

Fig. 7. The lattice in Examples 7 and 9.

Hence the result follows immediately by using Corollary 2 and noting that T is the lattice-based sum of the above
mentioned family (Tα )α∈ of t-norms.

Example 7. Consider the lattice-ordered index set (, ) in Fig. 6. It is routine to check that the family of bounded
lattices associated with the structures in Fig. 7 is -sum family since it satisfies all the conditions in Definition 3. It
also satisfies the condition (M) since the index set is finite. Accordingly, by Theorem 2, the structure in Fig. 7 is a
bounded lattice. Now we consider the following lattice-based sum T = (Lα , Tα )α∈ of t-subnorms where Tδ is the
trivial t-subnorm on Lδ (i.e. Tδ (x, y) ≡ ⊥δ ), T is the drastic t-norm on L for each  ∈ {⊥ , α, β}, and (by using the
formula (9) in Definition 7) the function T is given by
⎧
⎪
⎪a , if (x, y) ∈ (Lα \ {aμ })2 ∪ (Lδ )2
⎪
⎨⊥ ,
β if (x, y) ∈ (Lβ \ {aμ })2
T (x, y) =
⎪ ⊥L ,
⎪ if (x, y) ∈ (L⊥ \ {⊥ })2
⎪
⎩
x ∧ y, otherwise
where a = ⊥α = ⊥δ and aμ = α = β . Then, by Corollary 1, T is a t-norm on L, and it is the smallest t-norm
obtained as a lattice-based sum of t-subnorms on the corresponding -sum family (Lα )α∈ associated with the
bounded lattice L in Fig. 7.

In the following example, we consider a finite instance of the bounded lattice in Fig. 7 where each summand is a
finite sublattice.
72 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

Fig. 8. The lattice in Example 8.

Table 1
The t-norm T in Example 8.
T ⊥ a0 c1 c2 c3 c4 b4 b0 b1 b2 b3 a1 a2 a3 a4 
⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥
a0 ⊥ a0 a0 a0 a0 a0 a0 ⊥ ⊥ ⊥ ⊥ a0 a0 a0 a0 a0
c1 ⊥ a0 a0 a0 c1 a0 c1 ⊥ ⊥ ⊥ ⊥ a0 a0 a0 a0 c1
c2 ⊥ a0 a0 c2 c2 c2 c2 ⊥ ⊥ ⊥ ⊥ a0 a0 a0 a0 c2
c3 ⊥ a0 c1 c2 c3 c2 c3 ⊥ ⊥ ⊥ ⊥ a0 a0 a0 a0 c3
c4 ⊥ a0 a0 c2 c2 c4 c4 ⊥ ⊥ ⊥ ⊥ a0 a0 a0 a0 c4
b4 ⊥ a0 c1 c2 c3 c4 b4 b0 b1 b2 b3 a0 a0 a0 a0 b4
b0 ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ b0 b0 b0 b0 b0 ⊥ ⊥ ⊥ ⊥ b0
b1 ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ b1 b0 b0 b0 b0 ⊥ ⊥ ⊥ ⊥ b1
b2 ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ b2 b0 b0 b0 b1 ⊥ ⊥ ⊥ ⊥ b2
b3 ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ b3 b0 b0 b1 b2 ⊥ ⊥ ⊥ ⊥ b3
a1 ⊥ a0 a0 a0 a0 a0 a0 ⊥ ⊥ ⊥ ⊥ a0 a0 a0 a0 a1
a2 ⊥ a0 a0 a0 a0 a0 a0 ⊥ ⊥ ⊥ ⊥ a0 a0 a0 a1 a2
a3 ⊥ a2 a0 a0 a0 a0 a0 ⊥ ⊥ ⊥ ⊥ a0 a0 a1 a2 a3
a4 ⊥ a0 a0 a0 a0 a0 a0 ⊥ ⊥ ⊥ ⊥ a0 a1 a2 a3 a4
 ⊥ a0 c1 c2 c3 c4 b4 b0 b1 b2 b3 a1 a2 a3 a4 

Example 8. Consider the lattice-ordered index set (, ) in Fig. 6, further the finite lattice with order given in
Fig. 8. Let Lδ = {a0 , a1 , a2 , a3 , a4 } and Lβ = {b0 , b1 , b2 , b3 , b4 } be finite chains in L such that a0 ≺L a1 ≺L
a2 ≺L a3 ≺L a4 and b0 ≺L b1 ≺L b2 ≺L b3 ≺L b4 . Further, consider the following sublattices of the lattice L:
Lα = {a0 , c1 , c2 , c3 , c4 , b4 }, L = {a0 }, Lμ = {b4 } L⊥ = Lγ = {⊥}, and L = Lν = {}. It can be easily checked
that the given family of bounded sublattices of the lattice L in Fig. 8 is a -sum family since it is an instance of
the -sum family from Example 7, and hence the lattice L = α∈ Lα , i.e. it is the lattice-based sum of its (given)
sublattices. Now we consider the operation T on the lattice L as defined by Table 1. It can also be easily checked
that T is the t-norm defined by formula (9) in Definition 7 and is, moreover, the lattice-based sum T = (Lα , Tα )α∈
of t-subnorms where both Tβ and Tα are t-norms on Lβ and Lα , respectively, while Tδ is a t-subnorm on Lδ , which
can be easily checked from their definitions on the corresponding sub-tables of Table 1. Observe that the t-norm Tβ is
given by Tβ (bi , bj ) = bk where k = max(0, i + j − 4), and it is usually known as the Łukasiewicz t-norm (see [28]),
while Tδ is the proper t-subnorm (from [27]) given by Tδ (ai , aj ) = ak where k = max(0, i + j − 5).

Example 9. Consider the lattice-ordered index set (, ) and the -sum family from Example 7. Now we consider
the following lattice-based sum T = (Lα , Tα )α∈ of t-subnorms where Tδ is the t-subnorm on Lδ given by (2), Tα
is the drastic t-norm on Lα , Tβ is the t-norm on Lβ given by (5), T⊥ is the t-norm on L⊥ given by (4), and the
function T is given by
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 73

Fig. 9. The bounded lattice in Example 10.

⎧
⎪
⎪ x ∧ y ∧ a⊥ , if (x, y) ∈ (L⊥ \ {⊥ })2
⎪
⎪
⎪
⎪ aδ , if (x, y) ∈ [aδ , δ ]2
⎪
⎪
⎨a , if (x, y) ∈ (Lα \ {aμ })2 ∪ ((Lδ )2 \ [aδ , δ ]2 )

T (x, y) =
⎪
⎪ a β, if (x, y) ∈ [aβ , aμ [2
⎪
⎪
⎪
⎪ ⊥β , if (x, y) ∈ (Lβ \ {aμ })2 \ [aβ , aμ [2
⎪
⎪
⎩
x ∧ y, otherwise

where aδ ∈ Lδ , aβ ∈ Lβ and a⊥ ∈ L⊥ . Then, by Corollary 1, T is a t-norm on the bounded lattice in Fig. 7.

Example 10. Consider the lattice-ordered index set (, ) in Fig. 6. It is also easy to see that the family of bounded
lattices associated with the structure in Fig. 9 is -sum families with (M) condition. Accordingly, by Theorem 2,
the structures in Fig. 9 is a bounded lattice. Now we consider the following lattice-based sum T = (Lα , Tα )α∈ of
t-subnorms, on the bounded lattice in Fig. 9, where Tδ and T are the t-subnorms given by (3) and (1), respectively,
T is the drastic t-norm on L for each  ∈ {⊥ , β}, Tα is the t-norm on Lα given by (5), and the function T is given
by
⎧
⎪
⎪ ⊥L , if (x, y) ∈ (L⊥ \ {⊥ })2
⎪
⎪
⎪
⎪ ⊥ , if (x, y) ∈ (Lβ \ {aμ })2
⎪
⎪ β
⎪
⎪
⎪
⎪ aα , if (x, y) ∈ [aα , aμ [2
⎪
⎨a ,
γ if (x, y) ∈ ((Lα \ {aμ })2 \ [aα , aμ [2 )
T (x, y) =
⎪
⎪ ∪((Lδ )2 \ {(δ , δ )})
⎪
⎪
⎪
⎪ δ , if x = y = δ
⎪
⎪
⎪
⎪
⎪
⎪ x ∧ y ∧ a , if (x, y) ∈ L
⎪
⎩
x ∧ y, otherwise

where aα ∈ Lα , a ∈ L , aγ  = ⊥α = ⊥δ = ⊥ , and aμ = α = β . Then, by Theorem 9, T is a t-subnorm

on the bounded lattice in Fig. 9. Observe that T is not a t-norm but a t-subnorm since the last summand T is not
a t-norm (assuming that a ∈ / {⊥ ,  }). A slight modification of the above lattice-based sum of t-subnorms by
replacing the t-subnorm T on L by a t-norm, say the t-norm given by (4), produces a t-norm on the bounded
lattice in Fig. 9. In this case, replacing the case before the last one in the above definition of the operation T by
(x ∧ y ∧ a , if (x, y) ∈ L \ { }), and the other cases remain the same, produces a t-norm on L.
74 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

6. Lattice-based sum of t-conorms on bounded lattices

In the same spirit as in Definition 7 we will also introduce in this section the lattice-based sum of t-subconorms
and then the lattice-based sum of t-conorms.
Recall that, from an axiomatical point of view, t-norms and t-conorms differ only with respect to their respective
boundary conditions, i.e. t-norms have the top element as unit, while t-conorms have the bottom element as unit. Anal-
ogously, t-subnorms and t-subconorms differ only with respect to their respective range conditions, i.e. t-subnorms
satisfy the range condition T (x, y) L x ∧ y, while t-subconorms satisfy x ∨ y L S(x, y).
In fact, the concepts t-norms and t-conorms, and, analogously, t-subnorms and t-subconorm, are dual in the follow-
ing sense: A t-conorm (t-subconorm) on L is a binary operator S such that S is a t-norm (t-subnorm) on Lop , where
Lop , the opposite of ordered set L, is defined by reversing the order on L.
The duality between t-norms and t-conorms allows us to translate the lattice-based sum construction methods of
t-norms (t-subnorms) into the corresponding ones of t-conorms (t-subconorms), as we shall see in this section. There-
fore, all results in Sections 4 and 5 given for t-norms (t-subnorms) remain valid for t-conorms (t-subconorms) with
the obvious changes where necessary. To be precise, we have the following results as corollaries of the corresponding
ones from the previous two sections:

Definition 8.Consider a (bounded) lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Let (Sα )α∈ be a family of t-subconorms on the corresponding summand lattices
(Lα )α∈ that satisfies for each α, β ∈  with β < α and ⊥α = β we have that Sα is a t-conorm. Furthermore, for
each α, β ∈  with α  β we have that Ssup{α,β} is also a t-conorm. The lattice-based sum of the family (Sα )α∈ of
t-subconorms is the function S : L2 −→ L given by

Sα (x, y), if (x, y) ∈ Lα × Lα
S(x, y) = (10)
x ∨ y, otherwise,
and denoted as S = {(Lα , Sα )}α∈ . If necessary, we refer to this type of sum as the lattice-based sum of t-subconorms
while the lattice-based sum of t-conorms for that whose all underlying operations Sα are t-conorms.

Remark 6. Observe that, by Definition 6 and Theorem 2, the operation S considered in Definition 8 can be described
alternatively as
⎧
⎪
⎪ Sα (x, y) if (x, y) ∈ Lα × Lα ,
⎨
y if (x, y) ∈ Lα × Lβ and α < β,
S(x, y) =
⎪
⎪ x if (x, y) ∈ Lα × Lβ and β < α,
⎩
⊥sup{α ∗ ,β ∗ } if x  y, α ∗ ∈ Ixmax and β ∗ ∈ Iymax .

Theorem 9. With all the assumptions of Definition 8 the lattice-based sum of t-subconorms S = {(Lα , Sα )}α∈ as
defined in equation (10) is a t-subconorm on L. If we assume further that for L⊥ we have that S⊥ is a t-conorm,
then S is a t-conorm on L.

Theorem 10.Consider a bounded lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Suppose that (Sα )α∈ is a family of t-conorms on the corresponding summand
lattices (Lα )α∈ . Let S be a function defined on L by (10). Then S is a t-conorm.

By using the latticed-based sum theorems from this section, and in complete analogy with Section 5, we can
construct several new t-conorms on bounded lattices. For examples, the following results perfectly dual to the corre-
sponding ones from Section 5 and hence they have dual proofs.

11. Consider a bounded lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
Proposition 
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Fix an aα ∈ Lα for each α ∈ . Then the following operations are t-conorms on L.

x ∨ y ∨ aα , if (x, y) ∈ (Lα \{⊥α })2
S(x, y) =
x ∨ y, otherwise.
 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76 75

⎧
⎨ aα , if (x, y) ∈ (]⊥α , aα ])2
S(x, y) = α , if (x, y) ∈ (Lα \{⊥α })2 \(]⊥α , aα ])2
⎩
x ∨ y, otherwise.

Corollary 4. Consider a bounded lattice-ordered index set (, ) and a lattice-based sum of bounded lattices
(L, ∧, ∨) = α∈ (Lα , ∧α , ∨α ). Then the following operation is a t-conorm on L.

α , if (x, y) ∈ (Lα \{⊥α })2
Sd (x, y) =
x ∨ y, otherwise.

7. Conclusions and future work

In this contribution, we submit to the reader a proposal of a new construction method for t-norms which is promising
for the further development of the theory of t-norms on bounded lattices, generalizing both ordinal and horizontal sums
of t-norms. We generalized the well-known ordinal sum technique of t-norms to what we call lattice-based sum of
t-norms by allowing for lattice ordered index set instead of linearly ordered index set, showing that the lattice-based
sum of t-norms is again a t-norm. We demonstrated that t-subnorms can be used (with a little restriction) instead
of t-norms as summands in the lattice-based sum theorem of t-norms, still leading to a t-norm. Subsequently, we
applied the results for constructing several new families of t-norms and t-subnorms on bounded lattices. We ended
by showing that the duality allows us to translate the lattice-based sum construction methods of t-norms (t-subnorms)
into the corresponding ones of t-conorms (t-subconorms). Perhaps, the most important consequence is that a new
construction method for t-norms (t-conorms) on bounded lattices arise.
Note that though a consecutive repetition of standard ordinal and horizontal sum constructions is covered by the
lattice-based sum approach, the opposite is not true. Indeed, using results from [14,15,36] about the relationship be-
tween series-parallel lattices and N-free lattices, it is easy to see that the consecutive repetition of mentioned classical
construction using N-free summand lattices yields again an N-free lattice what is not the case of the lattice-based sum,
see e.g. the lattice-based sum in Fig. 5.
On the other hand, it is well known that (see [32]) the structure of the underlying lattice has quite an influence on
how the ordinal sum construction principle for building t-norms on bounded lattices can be determined. It turned out
that lattices built as ordinal and horizontal sums of chains are the most important and appropriate ones. Further that
there exist no t-norms on product lattices constructed through the ordinal sum construction principle except for some
trivial cases regarding the t-norms or the lattices involved.
These considerations would inevitably lead one into studying the expressive power of lattice-based sums. Thus, we
may ask whether there are lattice-based sum t-norms on product lattices with the order on the product lattice being
defined coordinate-wisely. Also, it might be worthwhile looking for a characterization of all lattice-based sum t-norms
that can be obtained as ordinal and horizontal sums of the given t-norms on the corresponding summands. Further that
it might be worthwhile looking for a characterization of t-norms that are not decomposable into a lattice-based sum
in a nontrivial way. But first one should define precisely what is meant by a trivial or a nontrivial decomposition. For
this end, articles [7,19,25,29,32,33] may be relevant. We suggest these questions as a topic of future research.
Clearly, inspired by ideas of Clifford [5] (in the context of ordinal sums of abstract semigroups), further develop-
ment of this approach could deal with the lattice-based sums of semigroups and their consequences for the construction
of associative aggregation operators on bounded lattices, see e.g. [2,4,8,13,20–23,30]. This also allows us to study the
theory of t-norms on bounded lattices from the point of view of lattice sums in analogy to what has already been done
in the context of ordinal sums of t-norms (see, e.g., [18,24–26,32,34]). These topics will be investigated in a future
sequel to the present article.

References

[1] G. Birkhoff, Lattice Theory, third edition, American Mathematical Society, Providence, Rhode Island, 1967.
[2] G.D. Çayli, F. Karaçal, R. Mesiar, On a new class of uninorms on bounded lattices, Inf. Sci. 367–368 (2016) 221–231.
[3] G.D. Çayli, On a new class of t-norms and t-conorms on bounded lattices, Fuzzy Sets Syst. 332 (2018) 129–143.
[4] T. Cavlo, G. Mayor, R. Mesiar, Aggregation Operators: New Trends and Applications, Physica-Verlag, Heidelberg, 2002.
[5] A. Clifford, Naturally totally ordered commutative semigroups, Am. J. Math. 76 (1954) 631–646.
[6] B. Davey, H. Priestley, Introduction to Lattices and Order, second edition, Cambridge University Press, 2002.
76 M. El-Zekey / Fuzzy Sets and Systems 386 (2020) 60–76

[7] B. De Baets, R. Mesiar, Triangular norms on product lattices, Fuzzy Sets Syst. 104 (1999) 61–75.
[8] B. De Baets, R. Mesiar, Ordinal sums of aggregation operators, in: B. Bouchon-Meunier, J. Gutiérrez-Ríos, L. Magdalena, R.R. Yager (Eds.),
Technologies for Constructing Intelligent Systems, Tools, 2, Physica-Verlag, Heidelberg, 2002, pp. 137–148.
[9] M. El-Zekey, J. Medina, R. Mesiar, Lattice-based sums, Inf. Sci. 223 (2013) 270–284.
[10] Ü. Ertuǧrul, F. Karaçal, R. Mesiar, Modified ordinal sums of triangular norms and triangular conorms on bounded lattices, Int. J. Intell. Syst.
30 (2015) 807–817.
[11] F. Esteva, L. Godo, Monoidal t-norm based logic: towards a logic for left-continuous t-norms, Fuzzy Sets Syst. 124 (2001) 271–288.
[12] J. Goguen, L-fuzzy sets, J. Math. Anal. Appl. 18 (1967) 145–174.
[13] M. Grabisch, J.L. Marichal, R. Mesiar, E. Pap, Aggregation functions: construction methods, conjunctive, disjunctive and mixed classes, Inf.
Sci. 181 (1) (2011) 23–43.
[14] J.L. Gischer, The equational theory of pomsets, Theor. Comput. Sci. 61 (1988) 199–224.
[15] J. Grabowski, On partial languages, Ann. Soc. Math. Polon. Ser. IV Fund. Inform. 4 (2) (1981) 427–498.
[16] P. Hájek, Metamathematics of Fuzzy Logic, Kluwer, Dordrecht, 1998.
[17] U. Höhle, Commutative, residuated
-monoids, in: U. Höhle, E.P. Klement (Eds.), Non-classical Logics and Their Applications to Fuzzy
Subsets, Kluwer Academic Publishers, Dordrecht, 1995, pp. 53–106.
[18] S. Jenei, A note on the ordinal sum theorem and its consequence for the construction of triangular norms, Fuzzy Sets Syst. 126 (2002) 199–205.
[19] S. Jenei, B. De Baets, On the direct decomposability of t-norms on product lattices, Fuzzy Sets Syst. 139 (2003) 699–707.
[20] F. Karaçal, R. Mesiar, Uninorms on bounded lattices, Fuzzy Sets Syst. 261 (2015) 33–43.
[21] F. Karaçal, M.A. Ince, R. Mesiar, Nullnorms on bounded lattices, Inf. Sci. 325 (2015) 227–236.
[22] F. Karaçal, Ü. Ertuǧrul, R. Mesiar, Characterization of uninorms on bounded lattices, Fuzzy Sets Syst. 308 (2017) 54–71.
[23] F. Karaçal, R. Mesiar, Aggregation functions on bounded lattices, Int. J. Gen. Syst. 46 (2017) 37–51.
[24] E. Klement, R. Mesiar, E. Pap, Triangular Norms, Kluwer Academic Publ., 2000.
[25] E. Klement, R. Mesiar, E. Pap, Triangular norms as ordinal sums of semigroups in the sense of A, H. Clifford, Semigroup Forum 65 (2002)
71–82.
[26] C. Ling, Representation of associative functions, Publ. Math. (Debr.) 12 (1965) 189–212.
[27] M. Mas, M. Monserrat, J. Torrens, Smooth t-subnorms on finite scales, Fuzzy Sets Syst. 167 (2011) 82–91.
[28] G. Mayor, J. Torrens, Triangular norms in discrete settings, in: E.P. Klement, R. Mesiar (Eds.), Logical, Algebraic, Analytic, and Probabilistic
Aspects of Triangular Norms, Elsevier, Amsterdam, 2005, pp. 189–230.
[29] J. Medina, Characterizing when an ordinal sum of t-norms is a t-norm on bounded lattices, Fuzzy Sets Syst. 202 (2012) 75–88.
[30] R. Mesiar, Magda Komorníková, Aggregation functions on bounded partially ordered sets and their classification, Fuzzy Sets Syst. 175 (2011)
48–56.
[31] E.S. Palmeira, B.C. Bedregal, Extension of fuzzy logic operators defined on bounded lattices via retractions, Comput. Math. Appl. 63 (2012)
1026–1038.
[32] S. Saminger-Platz, On ordinal sums of triangular norms on bounded lattices, Fuzzy Sets Syst. 157 (2006) 1403–1416.
[33] S. Saminger-Platz, E. Klement, R. Mesiar, On extension of triangular norms on bounded lattices, Indag. Math. 19 (1) (2008) 135–150.
[34] B. Schweizer, A. Sklar, Associative functions and abstract semigroups, Publ. Math. (Debr.) 10 (1963) 69–81.
[35] B. Schweizer, A. Sklar, Probabilistic Metric Spaces, North Holland, New York, 1983, reprinted, Dover, Mineola NY, 2005.
[36] J. Valdes, R.E. Tarjan, E.L. Lawler, The recognition of series parallel digraphs, SIAM J. Comput. 11 (2) (1982) 298–313.
[37] D. Zhang, Triangular norms on bounded partially ordered sets, Fuzzy Sets Syst. 153 (2005) 195–209.