betaFeedback

WorksMathematics

Uninorm constructions on bounded trellises via closure and interior operators

Abstract

In this paper, we establish necessary and sufficient conditions under which six closure- or interior-based operations on bounded trellises are uninorms. These operations incorporate triangular norms, triangular conorms, or both, and form three order-dual pairs. Since the underlying pseudo-order need not be transitive and meet and join need not be associative, standard lattice arguments do not apply directly. Under a fixed ambient transitivity assumption, the first pair is characterized by image, monotonicity, and compatibility conditions for the associated maps on elements incomparable with the neutral element; an additional comparison condition used in earlier sufficient constructions is shown to be redundant. Complete characterizations are also obtained for the variants with an absorbing endpoint, including both nondegenerate and degenerate interval cases. For the pair involving both a triangular norm and a triangular conorm, we derive exact comparison conditions requiring the triangular norm to take the value~00, or dually the triangular conorm to take the value~11, in the relevant cases. These conditions reduce to the usual incomparability assumptions in the absence of nontrivial zero- or one-divisors. On bounded lattices, the closure and interior conditions simplify to explicit identities or image restrictions. Finite proper-trellis examples clarify the role and limitations of the ambient transitivity assumptions.

INTRODUCTION

Uninorms were introduced by Yager and Rybalov as associative and commutative aggregation operations whose neutral element need not be an endpoint [29]. Their structure on the unit interval has been studied from several viewpoints, including representability and ordinal sums [7, 8, 18]. More broadly, lattice-valued fuzzy-set frameworks have long motivated topological and logical investigations [11, 12].

The extension of uninorms to bounded lattices led to a variety of construction and characterization results [1, 2, 3, 5, 13, 15, 19, 21, 25, 31, 33]. Of particular relevance here are constructions and characterizations based on closure and interior operators [4, 20, 22, 27]. These operators are useful when defining the values of an operation on elements incomparable with its neutral element. A bounded trellis replaces the transitive order of a bounded lattice by a reflexive and antisymmetric pseudo-order while retaining a meet and a join for each pair of elements [9, 10, 23]. In this setting, meet and join need not be associative, and consecutive comparisons cannot be concatenated without a suitable transitivity hypothesis. Zedam and De Baets [30] studied triangular norms on bounded trellises. Kong and Zhao [14] subsequently introduced uninorms on bounded trellises and gave constructions using triangular norms, triangular conorms, and closure or interior operators. Further developments appear in [16, 17, 24, 26, 28, 32].

This paper examines six closure- or interior-based formulas, arranged in three order-dual pairs. Its aim is not to introduce another unrestricted construction scheme, but to determine when these particular formulas define uninorms. Kong and Zhao [14] gave several sufficient constructions on bounded trellises, distinguished by transitivity properties of the neutral element. For the first pair considered here, (C1) is retained as an ambient hypothesis, while (C2) and (C4) are shown to be necessary and sufficient; (C3) then follows from (C1) and (C2). The dual statement holds for the interior-based formula. The variants with an absorbing endpoint are treated separately according to whether the relevant strict interval is empty.

For the pair involving both a triangular norm and a triangular conorm, the usual incomparability assumption remains sufficient, but the exact requirement is the implication in (19) or its order dual (22). These conditions coincide with the incomparability assumptions when there are no nontrivial zero- or one-divisors, and a counterexample shows that the converse fails without this restriction. On bounded lattices, the closure and interior parts reduce to explicit identities or image restrictions. The related lattice constructions in [4, 20, 22, 27] use formulas or input operations that are not identical in general. The lattice results below therefore compare the criteria for the present formulas rather than claim that all of those constructions are special cases.

§ 2 introduces the notation used below. § 3 collects the hypotheses on the closure and interior data. The six operations are studied in § 4, and their bounded-lattice forms are obtained in § 5.

PRELIMINARIES

Definition 2.1. [9] A pseudo-order on a nonempty set XX is a reflexive and antisymmetric binary relation ⊴\unlhd. A pseudo-ordered set is denoted by (X,⊴)(X,\unlhd). We write x⊲yx \lhd y when x⊴yx \unlhd y and x≠yx \ne y, and x∥yx \parallel y when neither x⊲yx \lhd y nor y⊲xy \lhd x holds. A pseudo-ordered set is called bounded if it has a least element 00 and a greatest element 11.

Definition 2.2. [10] A trellis is a pseudo-ordered set (X,⊴)(X,\unlhd) in which every pair x,yx,y has a greatest lower bound x∧yx \wedge y and a least upper bound x∨yx \vee y. It is bounded if it has a least element 00 and a greatest element 11. A bounded trellis is written T=(X,⊴,∧,∨,0,1)\mathbb{T}=(X,\unlhd,\wedge,\vee,0,1).

The relation of a trellis is not assumed to be transitive. Accordingly, all iterated meets and joins in the trellis arguments below are explicitly parenthesized. The order dual of a bounded trellis T=(X,⊴,∧,∨,0,1)\mathbb{T}=(X,\unlhd,\wedge,\vee,0,1) is T∂=(X,⊴∂,∨,∧,1,0)\mathbb{T}^{\partial}=(X,\unlhd^{\partial},\vee,\wedge,1,0), where x⊴∂yx\unlhd^{\partial} y if and only if y⊴xy\unlhd x [6]. It is again a bounded trellis. Under order duality, meet and join, the least and greatest elements, right- and left-transitivity, closure and interior operators, and triangular norms and triangular conorms are interchanged, whereas middle-transitivity is preserved.

Following [14], for a⊴ba \unlhd b put [a,b]={x∈X:a⊴x⊴b}[a,b]=\{x\in X:a\unlhd x\unlhd b\} and [a,b)={x∈[a,b]:x≠b}[a,b)=\{x\in[a,b]:x\ne b\}, and define (a,b](a,b] and (a,b)(a,b) analogously. For a fixed e∈X∖{0,1}e\in X\setminus\{0,1\}, the incomparable set is Ie={x∈X:x∥e}I_e=\{x\in X:x\parallel e\}. Thus XX is the disjoint union of [0,e)[0,e), {e}\{e\}, (e,1](e,1], and IeI_e.

Definition 2.3. [10] A pseudo-ordered set (X,⊴)(X,\unlhd) is a ∧\wedge-semi-trellis (respectively, a ∨\vee-semi-trellis) if every pair of elements has a greatest lower bound (respectively, a least upper bound). A subset AA of a trellis is a sub-trellis if x∧y,x∨y∈Ax\wedge y,x\vee y\in A for all x,y∈Ax,y\in A, with the operations and pseudo-order restricted to AA.

Definition 2.4. [14] An element a∈Xa\in X is right-transitive if a⊴x⊴ya\unlhd x\unlhd y implies a⊴ya\unlhd y; it is left-transitive if x⊴y⊴ax\unlhd y\unlhd a implies x⊴ax\unlhd a; and it is middle-transitive if x⊴a⊴yx\unlhd a\unlhd y implies x⊴yx\unlhd y. The corresponding sets are denoted by XrtrX^{\mathrm{rtr}}, XltrX^{\mathrm{ltr}}, and XmtrX^{\mathrm{mtr}}. Their intersection is denoted by XtrX^{\mathrm{tr}}.

Definition 2.5. [30] A binary operation F:X2→XF:X^2\to X is increasing if

x1⊴x2 and y1⊴y2⟹F(x1,y1)⊴F(x2,y2).x_1\unlhd x_2\text{ and }y_1\unlhd y_2\quad\Longrightarrow\quad F(x_1,y_1)\unlhd F(x_2,y_2).

More generally, for D⊆X2D\subseteq X^2, a map F:D→XF:D\to X is increasing if the same implication holds whenever (x1,y1),(x2,y2)∈D(x_1,y_1),(x_2,y_2)\in D. A unary map f:A→Xf:A\to X, where A⊆XA\subseteq X, is increasing if x⊴yx\unlhd y in AA implies f(x)⊴f(y)f(x)\unlhd f(y). A triangular norm on a bounded pseudo-ordered set is an increasing, associative, and commutative operation with neutral element 1. A triangular conorm is defined dually, with neutral element 0. A uninorm on TT is an increasing, associative, and commutative operation U:X2→XU:X^2\to X having a neutral element e∈Xe\in X.

The four-variable definition of increasingness is essential. In a nontransitive pseudo-order, separate inequalities F(x1,y1)⊴F(x2,y1)F(x_1,y_1)\unlhd F(x_2,y_1) and F(x2,y1)⊴F(x2,y2)F(x_2,y_1)\unlhd F(x_2,y_2) do not by themselves imply the required joint comparison.

Definition 2.6. [20] A map cl⁡:X→X\operatorname{cl}:X\to X is a closure operator if it is extensive, increasing, and idempotent:

x⊴cl⁡(x),x⊴y⇒cl⁡(x)⊴cl⁡(y),cl⁡(cl⁡(x))=cl⁡(x).x\unlhd\operatorname{cl}(x),\qquad x\unlhd y\Rightarrow\operatorname{cl}(x)\unlhd\operatorname{cl}(y),\qquad\operatorname{cl}(\operatorname{cl}(x))=\operatorname{cl}(x).

An interior operator int⁡:X→X\operatorname{int}:X\to X is increasing and idempotent and satisfies int⁡(x)⊴x\operatorname{int}(x)\unlhd x.

We first recall a necessary condition for the neutral element of a uninorm.

Lemma 2.7. [14] If a uninorm on a bounded trellis has neutral element ee, then e∈Xmtre\in X^{\mathrm{mtr}}.

Proposition 2.8. [30] Let TT be a bounded trellis, and let TT be a triangular norm on the sub-trellis [0,e][0,e] with neutral element ee. Then T(0,x)=0T(0,x)=0 for every x∈[0,e]x\in[0,e].

CONDITIONS ON THE CLOSURE AND INTERIOR DATA

Let e∈X∖{0,1}e \in X \setminus\{0,1\}. For a closure operator cl⁡\operatorname{cl}, put

ce(x)=cl⁡(x)∧(x∨e),x∈Ie,(1)c_e(x)=\operatorname{cl}(x)\wedge(x\vee e),\qquad x\in I_e, \tag*{(1)}

and

me(x,y)=(x∧y)∧e,x,y∈Ie.(2)m_e(x,y)=(x\wedge y)\wedge e,\qquad x,y\in I_e. \tag*{(2)}

For the closure-based constructions, we use the following conditions.

(C1) e∈Xmtr∩Xrtre\in X^{\mathrm{mtr}}\cap X^{\mathrm{rtr}} and [0,e)∪Ie⊆Xrtr[0,e)\cup I_e\subseteq X^{\mathrm{rtr}}.

(C2) ce(Ie)⊆Iec_e(I_e)\subseteq I_e, the map cec_e is increasing on IeI_e, and ce(ce(x))=ce(x)c_e(c_e(x))=c_e(x) for every x∈Iex\in I_e.

(C3) If x∈Iex\in I_e, y∈(e,1]y\in(e,1], and x⊴yx\unlhd y, then ce(x)⊴yc_e(x)\unlhd y.

(C4) me:Ie2→Xm_e:I_e^2\to X is increasing in the joint sense, that is,

x1⊴x2, y1⊴y2⟹me(x1,y1)⊴me(x2,y2)(xi,yi∈Ie),x_1\unlhd x_2,\ y_1\unlhd y_2\quad\Longrightarrow\quad m_e(x_1,y_1)\unlhd m_e(x_2,y_2)\qquad(x_i,y_i\in I_e),

and me(ce(x),y)=me(x,y)m_e(c_e(x),y)=m_e(x,y) for all x,y∈Iex,y\in I_e.

The corresponding interior-side conditions are stated explicitly. For an interior operator int⁡\operatorname{int}, put

de(x)=int⁡(x)∨(x∧e),x∈Ie,(3)d_e(x)=\operatorname{int}(x)\vee(x\wedge e),\qquad x\in I_e, \tag*{(3)}

and

ne(x,y)=(x∨y)∨e,x,y∈Ie.(4)n_e(x,y)=(x\vee y)\vee e,\qquad x,y\in I_e. \tag*{(4)}

(I1) e∈Xmtr∩Xltre\in X^{\mathrm{mtr}}\cap X^{\mathrm{ltr}} and (e,1]∪Ie⊆Xltr(e,1]\cup I_e\subseteq X^{\mathrm{ltr}};

(I2) de(Ie)⊆Ied_e(I_e)\subseteq I_e, the map ded_e is increasing on IeI_e, and de(de(x))=de(x)d_e(d_e(x))=d_e(x) for every x∈Iex\in I_e;

(I3) if y∈[0,e)y\in[0,e), x∈Iex\in I_e, and y⊴xy\unlhd x, then y⊴de(x)y\unlhd d_e(x).

(I4) ne:Ie2→Xn_e:I_e^2\to X is increasing in the joint sense, namely,

x1⊴x2, y1⊴y2⟹ne(x1,y1)⊴ne(x2,y2)(xi,yi∈Ie),x_1\unlhd x_2,\ y_1\unlhd y_2\quad\Longrightarrow\quad n_e(x_1,y_1)\unlhd n_e(x_2,y_2)\qquad(x_i,y_i\in I_e),

and ne(de(x),y)=ne(x,y)n_e(d_e(x),y)=n_e(x,y) for all x,y∈Iex,y\in I_e.

Lemma 3.1. For x∈Iex\in I_e, one has x⊴ce(x)x\unlhd c_e(x) and de(x)⊴xd_e(x)\unlhd x. If (C1)–(C4) hold, then me(x,y)∈[0,e)m_e(x,y)\in[0,e) for all x,y∈Iex,y\in I_e. Dually, (I1)–(I4) imply ne(x,y)∈(e,1]n_e(x,y)\in(e,1].

Proof. The element xx is a lower bound of cl⁡(x)\operatorname{cl}(x) and x∨ex\vee e, so the defining property of the meet gives x⊴ce(x)x\unlhd c_e(x). The dual argument gives de(x)⊴xd_e(x)\unlhd x.

Clearly me(x,y)⊴em_e(x,y)\unlhd e. If me(x,y)=em_e(x,y)=e, then e⊴x∧y⊴xe\unlhd x\wedge y\unlhd x. Right-transitivity of ee would give e⊴xe\unlhd x, contrary to x∈Iex\in I_e. Here we use the hypothesis (C1) which says e∈Xrtre\in X^{\mathrm{rtr}}. Thus me(x,y)∈[0,e)m_e(x,y)\in[0,e). The last assertion is dual. □

Lemma 3.2. Assume (C1) and let x⊴yx\unlhd y. Then

  1. x∈(e,1]x \in(e,1] implies y∈(e,1]y \in(e,1];

  1. x=ex=e implies y∈{e}∪(e,1]y \in\{e\} \cup(e,1];

  1. x∈Iex \in I_e implies y∈Ie∪(e,1]y \in I_e \cup(e,1];

  1. y∈[0,e)y \in[0,e) implies x∈[0,e)x \in[0,e).

The order-dual statements hold under (I1).

Proof. For the first two assertions use right-transitivity of ee. If x∈Iex \in I_e and y≤ey \leq e, then x≤y≤ex \leq y \leq e and right-transitivity of xx contradicts x∥ex \parallel e, proving the third assertion. For the fourth assertion, if x∈(e,1]x \in(e,1], then item (1) would force y∈(e,1]y \in(e,1], contrary to y∈[0,e)y \in[0,e); if x=ex=e or x∈Iex \in I_e, items (2) and (3), respectively, give the same contradiction. Hence x∈[0,e)x \in[0,e). The uses of antisymmetry are confined to the endpoint identifications in the interval definitions. □

CONSTRUCTION METHODS FOR UNINORMS ON BOUNDED TRELLISES

A meet-based construction and its dual

Theorem 4.1. Let T\mathbb{T} be a bounded trellis, and let e∈X∖{0,1}e \in X \setminus\{0,1\}. Let cl⁡\operatorname{cl} be a closure operator, and let SS be a triangular conorm on the sub-trellis [e,1][e,1] with neutral element ee. Assume (C1) and define U1:X2→XU_1 : X^2 \to X by

U1(x,y)={S(x,y),x,y∈[e,1],x,(x,y)∈([0,e)×[e,1])∪(Ie×{e}),y,(x,y)∈([e,1]×[0,e))∪({e}×Ie),cl⁡(x)∧(x∨e),x∈Ie, y∈(e,1],cl⁡(y)∧(y∨e),y∈Ie, x∈(e,1],(x∧y)∧e,x,y∈Ie,0,otherwise.(5)U_1(x,y)= \begin{cases} S(x,y), & x,y \in[e,1],\\ x, & (x,y) \in([0,e) \times[e,1]) \cup(I_e \times\{e\}),\\ y, & (x,y) \in([e,1] \times[0,e)) \cup(\{e\} \times I_e),\\ \operatorname{cl}(x) \wedge(x \vee e), & x \in I_e,\ y \in(e,1],\\ \operatorname{cl}(y) \wedge(y \vee e), & y \in I_e,\ x \in(e,1],\\ (x \wedge y) \wedge e, & x,y \in I_e,\\ 0, & \text{otherwise.} \end{cases} \tag*{(5)}

Then the following statements are equivalent:

  1. U1U_1 is a uninorm on T\mathbb{T} with neutral element ee;

  1. conditions (C2) and (C4) hold.

Under these conditions, (C3) follows from (C1) and (C2), and 00 is an absorbing element of U1U_1.

Proof. Assume first that (C2) and (C4) hold. If x∈Iex \in I_e, y∈(e,1]y \in(e,1], and x≤yx \leq y, then yy is an upper bound of xx and ee. Hence x∨e≤yx \vee e \leq y. Since ce(x)≤x∨ec_e(x) \leq x \vee e and ce(x)∈Ie⊆Xrtrc_e(x) \in I_e \subseteq X^{\mathrm{rtr}}, right-transitivity gives ce(x)≤yc_e(x) \leq y. Thus (C3) follows, and (C1)–(C4) are available in the sufficiency proof below.

The displayed clauses assign exactly one value to every pair. The formula is symmetric, and direct substitution gives U1(e,x)=xU_1(e,x)=x for every x∈Xx \in X.

We first verify joint increasingness. Take x⊴x′x \unlhd x' and y⊴y′y \unlhd y'. Simultaneously interchanging (x,x′)(x,x') with (y,y′)(y,y') leaves the required comparison unchanged, so the roles of xx and yy may be interchanged whenever commutativity is used below. By Theorem 3.2, a point of (e,1](e,1] can move only to (e,1](e,1], a point equal to ee can move only to ee or (e,1](e,1], and a point of IeI_e can move only to IeI_e or (e,1](e,1]. We now inspect every possible case for (x,y)(x,y).

If (x,y)(x,y) lies in [0,e)2[0,e)^2, [0,e)×Ie[0,e) \times I_e, or Ie×[0,e)I_e \times[0,e), then U1(x,y)=0U_1(x,y)=0, so the required comparison follows from the minimality of 00.

Suppose x∈[0,e)x \in[0,e) and y∈[e,1]y \in[e,1]. Then y′∈[e,1]y' \in[e,1] and U1(x,y)=xU_1(x,y)=x. If x′∈[0,e)x' \in[0,e), then U1(x′,y′)=x′U_1(x',y')=x' and x⊴x′x \unlhd x'. If x′∈[e,1]x' \in[e,1], increasingness of SS gives x′=S(x′,e)⊴S(x′,y′)=U1(x′,y′)x'=S(x',e) \unlhd S(x',y')=U_1(x',y'). Since x∈[0,e)⊆Xrtrx \in[0,e) \subseteq X^{\mathrm{rtr}} by (C1) and x⊴x′x \unlhd x', it follows that x⊴U1(x′,y′)x \unlhd U_1(x',y'). If x′∈Iex' \in I_e, then

U1(x′,y′)={x′,y′=e,ce(x′),y′∈(e,1],U_1(x',y') = \begin{cases} x', & y'=e,\\ c_e(x'), & y' \in(e,1], \end{cases}

and x′⊴ce(x′)x' \unlhd c_e(x') by Theorem 3.1. Right-transitivity of xx again gives x⊴U1(x′,y′)x \unlhd U_1(x',y').

If x,y∈[e,1]x,y \in[e,1], then x′,y′∈[e,1]x',y' \in[e,1], and the desired inequality is precisely the joint increasingness of SS.

Suppose next that x=ex=e and y∈Iey \in I_e. Here x′∈{e}∪(e,1]x' \in\{e\} \cup(e,1] and y′∈Ie∪(e,1]y' \in I_e \cup(e,1]. If x′=ex'=e, then U1(x′,y′)=y′U_1(x',y')=y', so y⊴U1(x′,y′)y \unlhd U_1(x',y'). If x′∈(e,1]x' \in(e,1] and y′∈Iey' \in I_e, then y⊴y′⊴ce(y′)=U1(x′,y′)y \unlhd y' \unlhd c_e(y')=U_1(x',y'), and right-transitivity of yy yields the conclusion. Finally, if x′,y′∈(e,1]x',y' \in(e,1], increasingness of SS gives y′=S(e,y′)⊴S(x′,y′)=U1(x′,y′)y'=S(e,y') \unlhd S(x',y')=U_1(x',y'). Combining this with y⊴y′y \unlhd y' by right-transitivity of yy proves the remaining comparison.

Suppose x∈(e,1]x \in(e,1] and y∈Iey \in I_e. Then x′∈(e,1]x' \in(e,1] and y′∈Ie∪(e,1]y' \in I_e \cup(e,1]. If y′∈Iey' \in I_e, (C2) gives U1(x,y)=ce(y)⊴ce(y′)=U1(x′,y′)U_1(x,y)=c_e(y) \unlhd c_e(y')=U_1(x',y'). If y′∈(e,1]y' \in(e,1], then (C3) and increasingness of SS give ce(y)⊴y′=S(e,y′)⊴S(x′,y′)c_e(y) \unlhd y'=S(e,y') \unlhd S(x',y'). Because ce(y)∈Ie⊆Xrtrc_e(y) \in I_e \subseteq X^{\mathrm{rtr}}, the two displayed comparisons yield U1(x,y)⊴U1(x′,y′)U_1(x,y) \unlhd U_1(x',y') without invoking transitivity of the ambient relation.

It remains to take x,y∈Iex,y \in I_e. Put r=me(x,y)r=m_e(x,y). By Theorem 3.1, r∈[0,e)⊆Xrtrr \in[0,e) \subseteq X^{\mathrm{rtr}}. If x′,y′∈Iex',y' \in I_e, (C4) directly gives r⊴me(x′,y′)r \unlhd m_e(x',y'). If, say, x′∈Iex' \in I_e and y′∈(e,1]y' \in(e,1], then U1(x′,y′)=ce(x′)U_1(x',y')=c_e(x'). From the defining properties of the meets and repeated applications of right-transitivity of rr, we obtain

r⊴x∧y⊴x⊴x′⊴ce(x′),r \unlhd x \wedge y \unlhd x \unlhd x' \unlhd c_e(x'),

where each adjacent comparison is combined with the preceding one using that the fixed initial element rr is right-transitive. The case x′∈(e,1]x' \in(e,1], y′∈Iey' \in I_e is symmetric. If x′,y′∈(e,1]x',y' \in(e,1], the same argument first gives r⊴x′r \unlhd x', while x′=S(x′,e)⊴S(x′,y′)x'=S(x',e) \unlhd S(x',y'). One more application of right-transitivity of rr yields r⊴S(x′,y′)r \unlhd S(x',y'). The listed cases exhaust all transitions allowed by Theorem 3.2; hence U1U_1 is jointly increasing.

It remains to verify associativity. The element ee is neutral, and inspection of (5) shows that 00 is absorbing. Thus triples containing ee or 00 are immediate. For all remaining triples, the following list checks every possible choice of the inner pair, not merely one ordering for each placement of the three arguments among [0,e)[0,e), (e,1](e,1], and IeI_e.

  1. If all three elements lie in [0,e)[0,e), every inner product is 00, so every bracketing is 00.

  1. If all three lie in (e,1](e,1], the assertion is the associativity of SS.

  1. If all three lie in IeI_e, every inner product belongs to [0,e)[0,e) by Theorem 3.1, and its product with the remaining incomparable element is 00. Hence all three choices of the inner pair give 00.

  1. For two elements of [0,e)[0,e) and one element of (e,1](e,1], pairing the two lower elements first gives 00, whereas pairing a lower and an upper element first returns that lower element, whose product with the other lower element is 00.

  1. For two elements of [0,e)[0,e) and one element of IeI_e, every choice of the inner pair gives 00 either at the first step or at the second step.

  1. Let a∈[0,e)a \in[0,e) and b,c∈(e,1]b,c \in(e,1]. Pairing b,cb,c first gives U1(a,S(b,c))=aU_1(a,S(b,c))=a. Pairing aa with either upper element first also gives aa, and pairing the result with the remaining upper element again gives aa.

  1. For one element of [0,e)[0,e) and two elements of IeI_e, pairing the two incomparable elements first gives an element of [0,e)[0,e) and hence 00 after the second multiplication; pairing the lower element with either incomparable element first gives 00 immediately. Thus all bracketings are 00.

  1. Let a∈(e,1]a \in(e,1] and b,c∈Ieb,c \in I_e. Pairing b,cb,c first gives U1(a,me(b,c))=me(b,c)U_1(a,m_e(b,c))=m_e(b,c). The other two choices of inner pair give, respectively, me(ce(b),c)=me(b,c)m_e(c_e(b),c)=m_e(b,c) and me(ce(c),b)=me(c,b)=me(b,c)m_e(c_e(c),b)=m_e(c,b)=m_e(b,c), where (C4) and the symmetry of mem_e are used.

  1. Let a∈Iea \in I_e and b,c∈(e,1]b,c \in(e,1]. Pairing b,cb,c first gives ce(a)c_e(a). Pairing aa with either upper element first gives ce(a)c_e(a), and the second multiplication gives ce(ce(a))=ce(a)c_e(c_e(a))=c_e(a) by (C2).

  1. Let a∈[0,e)a \in[0,e), b∈(e,1]b \in(e,1], and c∈Iec \in I_e. Pairing a,ba,b first gives aa and then 00; pairing a,ca,c first gives 00; and pairing b,cb,c first gives ce(c)c_e(c), whose product with aa is 00. Hence all three bracketings are $0.

These are the ten possible placements of three arguments in [0,e)[0,e), (e,1](e,1], and IeI_e, and every possible inner pair has been covered. Therefore U1U_1 is associative and hence is a uninorm with neutral element ee.

Conversely, suppose that U1U_1 is a uninorm. We first record two facts that do not use (C2)–(C4). For x∈Iex \in I_e, Theorem 3.1 gives x≤ce(x)x \leq c_e(x). Moreover, for x,y∈Iex,y \in I_e one has me(x,y)≤em_e(x,y) \leq e. Equality me(x,y)=em_e(x,y)=e would give e≤x∧y≤xe \leq x \wedge y \leq x, and right-transitivity of ee would imply e≤xe \leq x, contrary to x∥ex \parallel e. Thus

me(x,y)∈[0,e)(x,y∈Ie).(6)m_e(x,y) \in[0,e) \qquad(x,y \in I_e). \tag*{(6)}

Fix x∈Iex \in I_e. If ce(x)≤ec_e(x) \leq e, then right-transitivity of xx, applied to x≤ce(x)≤ex \leq c_e(x) \leq e, would give x≤ex \leq e, a contradiction. Consequently, the decomposition X=[0,e)∪{e}∪(e,1]∪IeX=[0,e)\cup\{e\}\cup(e,1]\cup I_e shows that ce(x)c_e(x) belongs either to IeI_e or to (e,1](e,1].

Assume temporarily that ce(x)∈(e,1]c_e(x) \in(e,1]. Associativity at the triple (x,1,1)(x,1,1) gives

U1(U1(x,1),1)=U1(x,U1(1,1)).U_1(U_1(x,1),1)=U_1(x,U_1(1,1)).

Since a triangular conorm on [e,1][e,1] satisfies S(t,1)=1S(t,1)=1, the two sides are 11 and ce(x)c_e(x), respectively. Indeed, increasingness and neutrality give 1=S(e,1)⪯S(t,1)⪯11=S(e,1)\preceq S(t,1)\preceq1, so antisymmetry yields S(t,1)=1S(t,1)=1. Hence ce(x)=1c_e(x)=1. Associativity at (x,x,1)(x,x,1) would then give

me(x,x)=U1(U1(x,x),1)=U1(x,U1(x,1))=1,m_e(x,x)=U_1(U_1(x,x),1)=U_1(x,U_1(x,1))=1,

contradicting (6). Therefore

ce(x)∈Ie(x∈Ie).(7)c_e(x)\in I_e \qquad(x\in I_e). \tag*{(7)}

If x,y∈Iex,y\in I_e and x⪯yx\preceq y, increasingness of U1U_1, applied directly to (x,1)⪯(y,1)(x,1)\preceq(y,1), yields ce(x)=U1(x,1)⪯U1(y,1)=ce(y)c_e(x)=U_1(x,1)\preceq U_1(y,1)=c_e(y). Thus cec_e is increasing on IeI_e. Using (7), associativity at (x,1,1)(x,1,1) now reads

ce(ce(x))=U1(U1(x,1),1)=U1(x,U1(1,1))=ce(x).c_e(c_e(x))=U_1(U_1(x,1),1)=U_1(x,U_1(1,1))=c_e(x).

This proves (C2).

On Ie2I_e^2 the operation U1U_1 agrees with mem_e. Hence the joint increasingness of U1U_1 implies the joint increasingness of mem_e. Finally, for x,y∈Iex,y\in I_e, associativity at (1,x,y)(1,x,y) and (6) give

me(ce(x),y)=U1(U1(1,x),y)=U1(1,U1(x,y))=me(x,y).\begin{aligned} m_e(c_e(x),y)&=U_1(U_1(1,x),y)\\ &=U_1(1,U_1(x,y))\\ &=m_e(x,y). \end{aligned}

Thus (C4) also holds. □\square

Kong and Zhao [14] introduced uninorms on bounded trellises and gave four construction methods based on closure or interior operators together with suitable transitivity assumptions. Theorem 4.1 has a more specific purpose: after fixing the formula (5) and the ambient hypothesis (C1), it identifies (C2) and (C4) as the exact remaining conditions and shows that (C3) is redundant. In particular, the necessity proof derives the image, increasingness, idempotence, and compatibility requirements from the uninorm axioms instead of imposing them only as hypotheses. This is the precise sense in which the result is sharper for the fixed formula (5); the necessity is part of the conclusion rather than a restatement of the assumptions. It is not a characterization of all closure-based uninorms on bounded trellises, and the same distinction applies to Theorem 4.3 by order duality. The following example shows both why (C1) is used in Theorem 4.1 and why it need not hold for every individual operation defined by (5).

Example 4.2. Let Z={0,a,b,e,1}Z=\{0,a,b,e,1\}. Besides reflexivity and the boundary comparisons 0⪯z⪯10\preceq z\preceq1, impose only a⪯ba\preceq b and b⪯eb\preceq e. Thus a∥ea\parallel e. Comparable pairs have their usual meet and join, and a∧e=0a\wedge e=0, a∨e=1a\vee e=1. This gives a bounded trellis, which is proper because a⪯b⪯ea\preceq b\preceq e but a⪯̸ea\not\preceq e.

Let cl⁡\operatorname{cl} be the identity map. Then Ie={a}I_e=\{a\}, [0,e)={0,b}[0,e)=\{0,b\}, (e,1]={1}(e,1]=\{1\}, ce(a)=ac_e(a)=a, and me(a,a)=0m_e(a,a)=0. In particular, the lattice-style equality a∧e=cl⁡(a)∧ea\wedge e=\operatorname{cl}(a)\wedge e holds. On [e,1]={e,1}[e,1]=\{e,1\} take the unique triangular conorm. Formula (5), in the order 0,a,b,e,10,a,b,e,1,

becomes

U1U_100aabbee11
000000000000
aa000000aaaa
bb000000bbbb
ee00aabbee11
1100aabb1111

Table 1.

The table is symmetric and has neutral element ee. Associativity follows from the absorbing role of 00, the neutrality of ee, and the fact that the product of two elements of {a,b}\{a,b\} is 00. Direct inspection against the listed comparisons shows joint increasingness. Hence U1U_1 is a uninorm. Nevertheless, a∈Iea \in I_e is not right-transitive, since a⊴b⊴ea \unlhd b \unlhd e but a⋬ea \not\unlhd e; therefore (C1) fails. The same example shows that the lattice-style equality alone does not imply all of (C1)–(C4) on a proper trellis.

Theorem 4.3. Let TT be a bounded trellis, let e∈X∖{0,1}e \in X \setminus\{0,1\}, and let int⁡\operatorname{int} be an interior operator. Let TT be a triangular norm on the sub-trellis [0,e][0,e] with neutral element ee. Assume (I1) and define

U2(x,y)={T(x,y),x,y∈[0,e],x,(x,y)∈((e,1]×[0,e])∪(Ie×{e}),y,(x,y)∈([0,e]×(e,1])∪({e}×Ie),int⁡(x)∨(x∧e),x∈Ie, y∈[0,e),int⁡(y)∨(y∧e),y∈Ie, x∈[0,e),(x∨y)∨e,x,y∈Ie,1,otherwise.(8)U_2(x,y)= \begin{cases} T(x,y), & x,y \in[0,e],\\ x, & (x,y) \in((e,1] \times[0,e]) \cup(I_e \times\{e\}),\\ y, & (x,y) \in([0,e] \times(e,1]) \cup(\{e\} \times I_e),\\ \operatorname{int}(x) \vee(x \wedge e), & x \in I_e,\ y \in[0,e),\\ \operatorname{int}(y) \vee(y \wedge e), & y \in I_e,\ x \in[0,e),\\ (x \vee y) \vee e, & x,y \in I_e,\\ 1, & \text{otherwise}. \end{cases} \tag*{(8)}

Then U2U_2 is a uninorm with neutral element ee if and only if (I2) and (I4) hold. Under these conditions, (I3) follows from (I1) and (I2), and 11 is an absorbing element of U2U_2.

Proof. Consider the order-dual bounded trellis T∂T^{\partial}. With respect to T∂T^{\partial}, the interior operator int⁡\operatorname{int} is a closure operator, the interval [0,e][0,e] in TT becomes [e,1][e,1] in T∂T^{\partial}, and TT becomes a triangular conorm with neutral element ee. Conditions (I1)–(I4) become, respectively, (C1)–(C4), while ded_e and nen_e correspond to cec_e and mem_e. Formula (8) becomes (5). The assertion therefore follows from Theorem 4.1 by order duality. □

Variants with an absorbing endpoint

Theorem 4.4. Let TT be a bounded trellis, let e∈X∖{0,1}e \in X \setminus\{0,1\}, let cl⁡\operatorname{cl} be a closure operator, and let SS be a triangular conorm on [e,1][e,1] with neutral element ee. Assume (C1)–(C4) and (e,1)≠∅(e,1) \ne\varnothing, and define U3:X2→XU_3: X^2 \to X by

U3(x,y)={1,x=1 or y=1,S(x,y),x,y∈[e,1),x,(x,y)∈([0,e)×[e,1))∪(Ie×{e}),y,(x,y)∈([e,1)×[0,e))∪({e}×Ie),cl⁡(x)∧(x∨e),x∈Ie, y∈(e,1),cl⁡(y)∧(y∨e),y∈Ie, x∈(e,1),(x∧y)∧e,x,y∈Ie,0,otherwise.(9)U_3(x,y)= \begin{cases} 1, & x=1\text{ or }y=1,\\ S(x,y), & x,y\in[e,1),\\ x, & (x,y)\in([0,e)\times[e,1))\cup(I_e\times\{e\}),\\ y, & (x,y)\in([e,1)\times[0,e))\cup(\{e\}\times I_e),\\ \operatorname{cl}(x)\wedge(x\vee e), & x\in I_e,\ y\in(e,1),\\ \operatorname{cl}(y)\wedge(y\vee e), & y\in I_e,\ x\in(e,1),\\ (x\wedge y)\wedge e, & x,y\in I_e,\\ 0, & \text{otherwise.} \end{cases} \tag*{(9)}

Then U3U_3 is a uninorm on T\mathbb{T} with neutral element ee if and only if

S(x,y)≠1for every x,y∈(e,1).(10)S(x,y)\ne1\quad\text{for every }x,y\in(e,1). \tag*{(10)}

Proof. We first prove necessity. Suppose S(x,y)=1S(x,y)=1 for some x,y∈(e,1)x,y\in(e,1). Then

U3(0,U3(x,y))=U3(0,1)=1,U3(U3(0,x),y)=U3(0,y)=0,U_3(0,U_3(x,y))=U_3(0,1)=1,\qquad U_3(U_3(0,x),y)=U_3(0,y)=0,

so associativity fails. Hence (10) is necessary.

Conversely, assume (10). Formula (9) is symmetric and unambiguous, and direct substitution gives U3(e,x)=xU_3(e,x)=x for every x∈Xx\in X.

To prove joint increasingness, let x1⊴x2x_1\unlhd x_2 and y1⊴y2y_1\unlhd y_2. If x2=1x_2=1 or y2=1y_2=1, then U3(x2,y2)=1U_3(x_2,y_2)=1, so the desired comparison follows from the maximality of 11. Otherwise x2,y2≠1x_2,y_2\ne1. Antisymmetry then implies x1,y1≠1x_1,y_1\ne1: for example, x1=1x_1=1 together with 1=x1⊴x2⊴11=x_1\unlhd x_2\unlhd1 would force x2=1x_2=1. On (X∖{1})2(X\setminus\{1\})^2, the operations U3U_3 and U1U_1 agree pointwise. Consequently, the joint increasingness proved in Theorem 4.1 gives

U3(x1,y1)=U1(x1,y1)⊴U1(x2,y2)=U3(x2,y2).U_3(x_1,y_1)=U_1(x_1,y_1)\unlhd U_1(x_2,y_2)=U_3(x_2,y_2).

We next show that X∖{1}X\setminus\{1\} is closed under U3U_3. For inputs in [e,1)[e,1), condition (10) applies when both inputs belong to (e,1)(e,1); when one input equals ee, neutral-element identity S(e,t)=tS(e,t)=t gives a value different from 11. A mixed incomparable–upper clause has value ce(t)∈Iec_e(t)\in I_e by (C2), an incomparable–incomparable clause has value in [0,e)[0,e) by Theorem 3.1, the mixed clauses return one of the inputs, and the remaining clause returns 00. Hence X∖{1}X\setminus\{1\} is closed under U3U_3.

If at least one of x,y,zx,y,z equals 11, both bracketings of their product are 11, because 11 is absorbing for U3U_3. If none equals 11, the closure just proved ensures that both intermediate products also differ from 11. Since U3=U1U_3=U_1 on (X∖{1})2(X\setminus\{1\})^2, associativity of U1U_1 yields

U3(U3(x,y),z)=U1(U1(x,y),z)=U1(x,U1(y,z))=U3(x,U3(y,z)).\begin{aligned} U_3(U_3(x,y),z)&=U_1(U_1(x,y),z)\\ &=U_1(x,U_1(y,z))\\ &=U_3(x,U_3(y,z)). \end{aligned}

Therefore U3U_3 is a uninorm with neutral element ee.

Proposition 4.5. Let T\mathbb{T} be a bounded trellis, let e∈X∖{0,1}e \in X \setminus\{0,1\}, let cl⁡\operatorname{cl} be a closure operator, and let SS be a triangular conorm on [e,1][e,1] with neutral element ee. Assume (C1), and define U3U_3 by (9).

  1. If (e,1)≠∅(e,1) \ne\varnothing, then the following statements are equivalent:

    (a) U3U_3 is a uninorm on T\mathbb{T} with neutral element ee;

    (b) conditions (C2) and (C4) hold, and

S(x,y)≠1(x,y∈(e,1)).(11)S(x,y) \ne1 \qquad(x,y \in(e,1)). \tag*{(11)}

Under these equivalent conditions, (C3) follows from (C1) and (C2) and is therefore redundant.

  1. If (e,1)=∅(e,1) = \varnothing, then U3U_3 is a uninorm if and only if mem_e is jointly increasing on Ie2I_e^2. In this degenerate case no condition involving cec_e is necessary.

Proof. Assume first that (e,1)≠∅(e,1) \ne\varnothing and that U3U_3 is a uninorm. Condition (11) is necessary. Indeed, if p,q∈(e,1)p,q \in(e,1) and S(p,q)=1S(p,q)=1, then

U3(0,U3(p,q))=U3(0,1)=1,U3(U3(0,p),q)=U3(0,q)=0,U_3(0,U_3(p,q))=U_3(0,1)=1,\qquad U_3(U_3(0,p),q)=U_3(0,q)=0,

contrary to associativity. Hence (11) holds.

Condition (C1) also implies

me(x,y)∈[0,e)(x,y∈Ie).(12)m_e(x,y) \in[0,e) \qquad(x,y \in I_e). \tag*{(12)}

Indeed, me(x,y)⊴em_e(x,y) \unlhd e. If me(x,y)=em_e(x,y)=e, then e⊴x∧y⊴xe \unlhd x \wedge y \unlhd x, and right-transitivity of ee would give e⊴xe \unlhd x, contrary to x∥ex \parallel e.

Fix a∈(e,1)a \in(e,1) and x∈Iex \in I_e. Since x⊴ce(x)x \unlhd c_e(x), if ce(x)⊴ec_e(x) \unlhd e, right-transitivity of xx would imply x⊴ex \unlhd e, a contradiction. Thus ce(x)∈Ie∪(e,1]c_e(x) \in I_e \cup(e,1]. Associativity at the triple (x,x,a)(x,x,a) gives

me(x,x)=U3(U3(x,x),a)=U3(x,U3(x,a))=U3(x,ce(x)).(13)\begin{aligned} m_e(x,x) &= U_3(U_3(x,x),a)\\ &=U_3(x,U_3(x,a))=U_3(x,c_e(x)). \tag*{(13)} \end{aligned}

If ce(x)∈(e,1]c_e(x) \in(e,1], then the last member of (13) equals ce(x)c_e(x), whereas its first member belongs to [0,e)[0,e) by (12). This is impossible. Consequently,

ce(Ie)⊆Ie.(14)c_e(I_e) \subseteq I_e. \tag*{(14)}

If x,y∈Iex,y \in I_e and x⊴yx \unlhd y, increasingness of U3U_3, applied to (x,a)⊴(y,a)(x,a) \unlhd(y,a), gives ce(x)=U3(x,a)⊴U3(y,a)=ce(y)c_e(x)=U_3(x,a) \unlhd U_3(y,a)=c_e(y). Hence cec_e is increasing on IeI_e.

Increasingness of SS and neutrality of ee yield a=S(e,a)⊴S(a,a)a=S(e,a) \unlhd S(a,a). Since a≠ea \ne e, antisymmetry shows that S(a,a)≠eS(a,a) \ne e. Together with (11), this gives S(a,a)∈(e,1)S(a,a) \in(e,1).

Associativity at (x,a,a)(x,a,a) and (14) now yield

ce(ce(x))=U3(U3(x,a),a)=U3(x,U3(a,a))=U3(x,S(a,a))=ce(x).\begin{aligned} c_e(c_e(x)) &= U_3(U_3(x,a),a)\\ &=U_3(x,U_3(a,a))\\ &=U_3(x,S(a,a))\\ &=c_e(x). \end{aligned}

Thus (C2) holds. The restriction of U3U_3 to Ie2I_e^2 is mem_e. Therefore the joint increasingness of U3U_3 implies that mem_e is jointly increasing on Ie2I_e^2. Finally, for x,y∈Iex,y\in I_e, associativity at (a,x,y)(a,x,y) gives

me(ce(x),y)=U3(U3(a,x),y)=U3(a,U3(x,y))=U3(a,me(x,y))=me(x,y),\begin{aligned} m_e(c_e(x),y) &= U_3(U_3(a,x),y)\\ &= U_3(a,U_3(x,y))\\ &= U_3(a,m_e(x,y))\\ &= m_e(x,y), \end{aligned}

where (12) is used in the last equality. Hence (C4) holds.

Conversely, suppose that (C2), (C4), and (11) hold. As in Theorem 4.1, conditions (C1) and (C2) imply (C3). Therefore (C1)–(C4) hold, and Theorem 4.4 shows that U3U_3 is a uninorm.

It remains to consider (e,1)=∅(e,1)=\varnothing. In this case the definition of U3U_3 reduces to

U3(x,y)={1,x=1 or y=1,y,x=e,x,y=e,me(x,y),x,y∈Ie,0,otherwise.U_3(x,y)= \begin{cases} 1, & x=1\text{ or }y=1,\\ y, & x=e,\\ x, & y=e,\\ m_e(x,y), & x,y\in I_e,\\ 0, & \text{otherwise}. \end{cases}

In particular, neither cec_e nor a nontrivial value of SS occurs in the formula.

If U3U_3 is a uninorm, its restriction to Ie2I_e^2 shows immediately that mem_e is jointly increasing. Conversely, assume that mem_e is jointly increasing. Commutativity and neutrality of ee follow directly from the displayed formula.

For increasingness, let x⪯x′x\preceq x' and y⪯y′y\preceq y'. If x′=1x'=1 or y′=1y'=1, the required comparison follows from the maximality of 1. Otherwise, Theorem 3.2 and (e,1)=∅(e,1)=\varnothing imply that x=ex=e forces x′=ex'=e, and x∈Iex\in I_e forces x′∈Iex'\in I_e; the same statements hold for yy and y′y'. If x=ex=e or y=ey=e, neutrality reduces the desired inequality to the comparison involving the other variable. If x,y∈Iex,y\in I_e, the assertion is the joint increasingness of mem_e. In every remaining case U3(x,y)=0U_3(x,y)=0. Thus U3U_3 is jointly increasing.

For associativity, a triple containing 1 gives the value 1 under both bracketings, and a triple containing ee reduces by neutrality. Every remaining entry belongs to [0,e)∪Ie[0,e)\cup I_e. If all three entries belong to IeI_e, each inner product belongs to [0,e)[0,e) by (12), and the second product is 0. If exactly two entries belong to IeI_e, pairing them first gives an element of [0,e)[0,e) and hence gives 0 at the second step, while pairing an incomparable and a lower element first gives 0 immediately. If at most one entry belongs to IeI_e, both bracketings are again 0. Therefore U3U_3 is associative, completing the proof. □

Theorem 4.6. Let T\mathbb{T} be a bounded trellis, let e∈X∖{0,1}e\in X\setminus\{0,1\}, let int⁡\operatorname{int} be an interior operator, and let TT be a triangular norm on [0,e][0,e] with neutral element ee. Assume (I1)–(I4) and (0,e)≠∅(0,e) \ne\varnothing, and define

U4(x,y)={0,x=0 or y=0,T(x,y),x,y∈(0,e],x,(x,y)∈((e,1]×(0,e])∪(Ie×{e}),y,(x,y)∈((0,e]×(e,1])∪({e}×Ie),int⁡(x)∨(x∧e),x∈Ie, y∈(0,e),int⁡(y)∨(y∧e),y∈Ie, x∈(0,e),(x∨y)∨e,x,y∈Ie,1,otherwise.(15)U_4(x,y)= \begin{cases} 0, & x=0\text{ or }y=0,\\ T(x,y), & x,y\in(0,e],\\ x, & (x,y)\in((e,1]\times(0,e])\cup(I_e\times\{e\}),\\ y, & (x,y)\in((0,e]\times(e,1])\cup(\{e\}\times I_e),\\ \operatorname{int}(x)\vee(x\wedge e), & x\in I_e,\ y\in(0,e),\\ \operatorname{int}(y)\vee(y\wedge e), & y\in I_e,\ x\in(0,e),\\ (x\vee y)\vee e, & x,y\in I_e,\\ 1, & \text{otherwise.} \end{cases} \tag*{(15)}

Then U4U_4 is a uninorm if and only if

T(x,y)≠0for every x,y∈(0,e).(16)T(x,y)\ne0\qquad\text{for every }x,y\in(0,e). \tag*{(16)}

Proof. We first prove necessity. If T(x,y)=0T(x,y)=0 for some x,y∈(0,e)x,y\in(0,e), then

U4(1,U4(x,y))=U4(1,0)=0,U4(U4(1,x),y)=U4(1,y)=1.U_4(1,U_4(x,y))=U_4(1,0)=0,\qquad U_4(U_4(1,x),y)=U_4(1,y)=1.

Thus associativity implies (16).

Conversely, assume (16). The formula is the order dual of the definition in Theorem 4.4, so it is well defined and symmetric, and ee is neutral. We give the remaining dual argument explicitly.

Let x1⊑x2x_1\sqsubseteq x_2 and y1⊑y2y_1\sqsubseteq y_2. If x1=0x_1=0 or y1=0y_1=0, then U4(x1,y1)=0⊑U4(x2,y2)U_4(x_1,y_1)=0\sqsubseteq U_4(x_2,y_2). Otherwise x1,y1≠0x_1,y_1\ne0, and antisymmetry implies x2,y2≠0x_2,y_2\ne0; for example, 0⊑x1⊑x2=00\sqsubseteq x_1\sqsubseteq x_2=0 would force x1=0x_1=0. On (X∖{0})2(X\setminus\{0\})^2, the operations U4U_4 and U2U_2 agree pointwise. The joint increasingness of U2U_2 therefore gives

U4(x1,y1)=U2(x1,y1)⊑U2(x2,y2)=U4(x2,y2).U_4(x_1,y_1)=U_2(x_1,y_1)\sqsubseteq U_2(x_2,y_2)=U_4(x_2,y_2).

The set X∖{0}X\setminus\{0\} is closed under U4U_4. Indeed, for inputs in (0,e](0,e], condition (16) treats the case in which both inputs belong to (0,e)(0,e), while T(e,t)=tT(e,t)=t treats the case in which one input is ee. The mixed incomparable–lower clauses take values in IeI_e by (I2), the incomparable–incomparable clause takes values in (e,1](e,1] by Theorem 3.1, the mixed clauses return a nonzero input, and the remaining clause returns 1.

If one of x,y,zx,y,z is 0, both bracketings are 0, because 0 is absorbing for U4U_4. If none is 0, closure keeps both intermediate values in X∖{0}X\setminus\{0\}, where U4=U2U_4=U_2. Hence

U4(U4(x,y),z)=U2(U2(x,y),z)=U2(x,U2(y,z))=U4(x,U4(y,z)).\begin{aligned} U_4(U_4(x,y),z)&=U_2(U_2(x,y),z)\\ &=U_2(x,U_2(y,z))\\ &=U_4(x,U_4(y,z)). \end{aligned}

by associativity of U2U_2. Thus U4U_4 is a uninorm. □

Combining the preceding result with the order-dual necessity argument gives the criterion for the interior construction with absorbing element 0.

Corollary 4.7. Let T\mathbb{T} be a bounded trellis, e∈X∖{0,1}e \in X \setminus\{0,1\}, int⁡\operatorname{int} be an interior operator, and TT be a triangular norm on [0,e][0,e] with neutral element ee. Assume (I1), and define U4U_4 by (15).

  1. If (0,e)≠∅(0,e) \ne\varnothing, then U4U_4 is a uninorm if and only if (I2), (I4), and T(x,y)≠0T(x,y) \ne0 for all x,y∈(0,e)x,y \in(0,e) hold. Under these conditions, (I3) is redundant.

  1. If (0,e)=∅(0,e) = \varnothing, then U4U_4 is a uninorm if and only if nen_e is jointly increasing on Ie2I_e^2; no condition involving ded_e is necessary.

Proof. Apply Theorem 4.5 to the order-dual bounded trellis. Under this duality, (C1)–(C4), cec_e, mem_e, and condition (11) become (I1)–(I4), ded_e, nen_e, and condition (16), respectively. The nondegenerate and degenerate criteria therefore dualize exactly as stated. □

Constructions using both a triangular norm and a triangular conorm

Let T\mathbb{T} be a bounded trellis, let e∈X∖{0,1}e \in X \setminus\{0,1\}, let cl⁡\operatorname{cl} be a closure operator, let TT be a triangular norm on [0,e][0,e] with neutral element ee, and let SS be a triangular conorm on [e,1][e,1] with neutral element ee. We shall use the following sufficient condition:

x∥yfor every x∈(0,e), y∈Ie.(17)x \parallel y \qquad\text{for every } x \in(0,e),\ y \in I_e. \tag*{(17)}

Define

U5(x,y)={T(x,y),x,y∈[0,e),S(x,y),x,y∈[e,1],x,(x,y)∈([0,e)×[e,1])∪(Ie×{e}),y,(x,y)∈([e,1]×[0,e))∪({e}×Ie),cl⁡(x)∧(x∨e),x∈Ie, y∈(e,1],cl⁡(y)∧(y∨e),y∈Ie, x∈(e,1],0,otherwise.(18)U_5(x,y) = \begin{cases} T(x,y), & x,y \in[0,e),\\ S(x,y), & x,y \in[e,1],\\ x, & (x,y) \in([0,e) \times[e,1]) \cup(I_e \times\{e\}),\\ y, & (x,y) \in([e,1] \times[0,e)) \cup(\{e\} \times I_e),\\ \operatorname{cl}(x) \wedge(x \vee e), & x \in I_e,\ y \in(e,1],\\ \operatorname{cl}(y) \wedge(y \vee e), & y \in I_e,\ x \in(e,1],\\ 0, & \text{otherwise.} \end{cases} \tag*{(18)}

The following result gives the necessary and sufficient condition under (C1).

Theorem 4.8. Let T\mathbb{T} be a bounded trellis, let e∈X∖{0,1}e \in X \setminus\{0,1\}, let cl⁡\operatorname{cl} be a closure operator, let TT be a triangular norm on [0,e][0,e] with neutral element ee, and let SS be a triangular conorm on [e,1][e,1] with neutral element ee. Assume (C1), and define U5U_5 by (18). Then the following statements are equivalent:

  1. U5U_5 is a uninorm on T\mathbb{T} with neutral element ee;

  1. condition (C2) holds and

p⊴z⟹T(p,q)=0(p,q∈[0,e), z∈Ie).(19)p \unlhd z \quad\Longrightarrow\quad T(p,q)=0 \qquad(p,q \in[0,e),\ z \in I_e). \tag*{(19)}

Under these assumptions, (C3) follows from (C2), whereas (C4) is not required for this construction.

Proof. Suppose first that U5U_5 is a uninorm. For x∈Iex \in I_e, put c=ce(x)c=c_e(x). By Theorem 3.1, x⊴cx \unlhd c. If c⊴ec \unlhd e, then right-transitivity of xx would imply x⊴ex \unlhd e, contrary to x∥ex \parallel e. Hence c∈Ie∪(e,1]c \in I_e \cup(e,1].

Assume that c∈(e,1]c \in(e,1]. Since 11 is absorbing for the triangular conorm SS, associativity at (x,1,1)(x,1,1) gives

1=U5(c,1)=U5(U5(x,1),1)=U5(x,U5(1,1))=c.1=U_5(c,1)=U_5(U_5(x,1),1)=U_5(x,U_5(1,1))=c.

Thus c=1c=1. Associativity at (x,x,1)(x,x,1) would then give

0=U5(U5(x,x),1)=U5(x,U5(x,1))=1,0=U_5(U_5(x,x),1)=U_5(x,U_5(x,1))=1,

a contradiction. Consequently ce(Ie)⊆Iec_e(I_e)\subseteq I_e. If x⊴yx\unlhd y in IeI_e, joint increasingness applied to (x,1)⊴(y,1)(x,1)\unlhd(y,1) gives ce(x)=U5(x,1)⊴U5(y,1)=ce(y)c_e(x)=U_5(x,1)\unlhd U_5(y,1)=c_e(y). Moreover, associativity at (x,1,1)(x,1,1), now using ce(x)∈Iec_e(x)\in I_e, yields

ce(ce(x))=U5(U5(x,1),1)=U5(x,U5(1,1))=ce(x).c_e(c_e(x))=U_5(U_5(x,1),1)=U_5(x,U_5(1,1))=c_e(x).

This proves (C2).

Let p,q∈[0,e)p,q\in[0,e) and z∈Iez\in I_e satisfy p⊴zp\unlhd z. Then (p,q)⊴(z,q)(p,q)\unlhd(z,q) componentwise, by the definition of U5U_5, U5(z,q)=0U_5(z,q)=0 for z∈Iez\in I_e and q∈[0,e)q\in[0,e), and joint increasingness gives T(p,q)=U5(p,q)⊴U5(z,q)=0T(p,q)=U_5(p,q)\unlhd U_5(z,q)=0. Since 00 is the least element, (19) follows.

Conversely, assume (C2) and (19). We first recover (C3). If x∈Iex\in I_e, y∈(e,1]y\in(e,1], and x⊴yx\unlhd y, then yy is an upper bound of both xx and ee, and hence ce(x)⊴x∨e⊴yc_e(x)\unlhd x\vee e\unlhd y. Condition (C2) gives ce(x)∈Iec_e(x)\in I_e, so (C1) makes ce(x)c_e(x) right-transitive. Therefore ce(x)⊴yc_e(x)\unlhd y, which is (C3).

We next prove joint increasingness. If p,q∈[0,e)p,q\in[0,e), monotonicity and neutrality of TT give

t=T(p,q)⊴T(p,e)=p,t⊴T(e,q)=q,t=T(p,q)\unlhd T(p,e)=p,\qquad t\unlhd T(e,q)=q,

and t∈[0,e)t\in[0,e). Similarly, for r,s∈(e,1]r,s\in(e,1] one has r,s⊴S(r,s)r,s\unlhd S(r,s) and S(r,s)∈(e,1]S(r,s)\in(e,1]. Let x⊴x′x\unlhd x' and y⊴y′y\unlhd y'. By commutativity, we may simultaneously interchange xx with yy and x′x' with y′y' in the cases below.

Suppose first that x,y∈[0,e)x,y\in[0,e) and put t=T(x,y)t=T(x,y). If x′,y′∈[0,e)x',y'\in[0,e), the assertion follows from joint increasingness of TT. If one of x′,y′x',y' lies in [0,e)[0,e) and the other in IeI_e, or if both lie in IeI_e, then x⊴x′x\unlhd x' or y⊴y′y\unlhd y' has an element of IeI_e on the right. Condition (19), together with commutativity of TT, gives t=0=U5(x′,y′)t=0=U_5(x',y'). If, say, x′∈[0,e)x'\in[0,e) and y′∈[e,1]y'\in[e,1], then U5(x′,y′)=x′U_5(x',y')=x' and t⊴x⊴x′t\unlhd x\unlhd x'; right-transitivity of tt completes the comparison. If x′,y′∈[e,1]x',y'\in[e,1], then t⊴e⊴S(x′,y′)t\unlhd e\unlhd S(x',y'), and middle-transitivity of ee applies. If {x′,y′}={e,z}\{x',y'\}=\{e,z\} with z∈Iez\in I_e, right-transitivity of tt gives t⊴z=U5(x′,y′)t\unlhd z=U_5(x',y'). If one of x′,y′x',y' is z∈Iez\in I_e and the other lies in (e,1](e,1], the same argument gives t⊴z⊴ce(z)=U5(x′,y′)t\unlhd z\unlhd c_e(z)=U_5(x',y').

Suppose next that x=p∈[0,e)x=p\in[0,e) and y∈[e,1]y\in[e,1]. Then U5(x,y)=pU_5(x,y)=p and Theorem 3.2 gives y′∈[e,1]y'\in[e,1]. If x′∈[0,e)x'\in[0,e), then U5(x′,y′)=x′U_5(x',y')=x' and p⊴x′p\unlhd x'. If x′∈[e,1]x'\in[e,1], then p⊴e⊴U5(x′,y′)p\unlhd e\unlhd U_5(x',y'), and middle-transitivity of ee applies. If x′=z∈Iex'=z\in I_e, then U5(x′,y′)U_5(x',y') is zz when y′=ey'=e and is ce(z)c_e(z) when y′∈(e,1]y'\in(e,1]; the required comparison follows from p⊴z⊴ce(z)p\unlhd z\unlhd c_e(z) and right-transitivity of pp. If x∈[0,e)x\in[0,e) and y∈Iey\in I_e, then U5(x,y)=0U_5(x,y)=0. If x,y∈[e,1]x,y \in[e,1], then x′,y′∈[e,1]x',y' \in[e,1] and joint increasingness of SS applies. Suppose next that x=ex=e and y=z∈Iey=z \in I_e. If y′=z′∈Iey'=z' \in I_e, then U5(x′,y′)U_5(x',y') is either z′z' or ce(z′)c_e(z'), and z⊴z′⊴ce(z′)z \unlhd z' \unlhd c_e(z') together with right-transitivity of zz gives the result. If y′∈(e,1]y' \in(e,1], then U5(x′,y′)U_5(x',y') is either y′y' or an SS-value above y′y', and the same argument applies.

Finally, suppose that x∈(e,1]x \in(e,1] and y=z∈Iey=z \in I_e. If y′∈Iey' \in I_e, increasingness of cec_e applies. If y′∈(e,1]y' \in(e,1], condition (C3) gives ce(z)⊴y′⊴S(x′,y′)c_e(z) \unlhd y' \unlhd S(x',y'), after a possible simultaneous interchange of the coordinates; since ce(z)∈Iec_e(z) \in I_e is right-transitive, the comparison follows. If x,y∈Iex,y \in I_e, then U5(x,y)=0U_5(x,y)=0. These cases prove joint increasingness of U5U_5 without using (17) or (C4).

It remains to prove associativity. Triples containing ee reduce by neutrality. On [0,e)3[0,e)^3 and (e,1]3(e,1]^3, associativity follows from that of TT and SS, respectively, while every pairing on Ie3I_e^3 is 0. For two lower and one upper input, every pairing gives their TT-value; for two lower and one incomparable input, every pairing gives 0. For one lower and two upper inputs, every pairing gives the lower input, and for one lower and two incomparable inputs every pairing gives 0, using T(0,x)=0T(0,x)=0. For one upper and two incomparable inputs, (C2) places the two possible cec_e-values in IeI_e, so all pairings are 0. For one incomparable and two upper inputs, all pairings give ce(x)c_e(x) by the idempotence in (C2). For one input from each of [0,e)[0,e), (e,1](e,1], and IeI_e, every pairing is 0. Thus U5U_5 is associative. Commutativity and neutrality of ee are immediate from (18), completing the proof. □

Corollary 4.9. Assume (C1)–(C3) and (17). Then the operation U5U_5 in (18) is a uninorm with neutral element ee.

Proof. Let p,q∈[0,e)p,q \in[0,e) and z∈Iez \in I_e with p⊴zp \unlhd z. Condition (17) excludes p∈(0,e)p \in(0,e), so p=0p=0 and T(p,q)=0T(p,q)=0. Hence (19) holds, and the result follows from Theorem 4.8. □

The incomparability restrictions used in the closure-based trellis constructions of [14] are sufficient for controlling pairs consisting of a lower and an incomparable element. For the fixed formula (18), Theorem 4.9 gives the corresponding sufficient statement, whereas Theorem 4.8 shows that the exact requirement is (19). The two requirements are equivalent when TT has no zero divisors by Theorem 4.11; Theorem 4.12 shows that they can differ when zero divisors are present. Thus the improvement concerns the criterion for (18), rather than an inclusion between all classes constructed in the two papers.

The construction in Theorem 4.9 occurs on the following small proper bounded trellis.

Example 4.10. Let X={0,e,d,a,b,c,1}X=\{0,e,d,a,b,c,1\}. Besides reflexivity and 0≤x≤10 \leq x \leq1, impose e≺ae \prec a, e≺be \prec b, e≺ce \prec c and a≺b≺c≺aa \prec b \prec c \prec a, with no further comparisons. Comparable pairs have the indicated lower and upper elements, whereas every incomparable pair has meet 0 and join 1. Hence XX is a bounded trellis. It is proper because a≤b≤ca \leq b \leq c but a≰ca \not\leq c.

Define cl⁡(a)=cl⁡(b)=cl⁡(c)=1\operatorname{cl}(a)=\operatorname{cl}(b)=\operatorname{cl}(c)=1 and let cl⁡\operatorname{cl} fix 0,e,d,10,e,d,1. This map is extensive and idempotent. It is also increasing: the only nonboundary comparisons not fixed by cl⁡\operatorname{cl} have an element of {a,b,c}\{a,b,c\} on the right, and all three of these elements have image 1. Thus cl⁡\operatorname{cl} is a closure operator. Moreover, Ie={d}I_e=\{d\}, [0,e)={0}[0,e)=\{0\}, and (e,1]={a,b,c,1}(e,1]=\{a,b,c,1\}. The element ee is middle- and right-transitive, while 0 and dd are right-transitive, so (C1) holds. Moreover, ce(d)=dc_e(d)=d and me(d,d)=0m_e(d,d)=0; since IeI_e is a singleton, conditions

(C2)–(C4) follow immediately. Condition (17) is vacuous because (0,e)=∅(0,e)=\varnothing; hence (19) holds.

On [0,e]={0,e}[0,e]=\{0,e\} take the unique triangular norm. On [e,1]={e,a,b,c,1}[e,1]=\{e,a,b,c,1\} define S(e,x)=S(x,e)=xS(e,x)=S(x,e)=x and put S(x,y)=1S(x,y)=1 whenever x,y≠ex,y\ne e. The operation is commutative and has neutral element ee. If at least two of three inputs differ from ee, both associative products are 11; otherwise associativity follows from neutrality. For increasingness, a value different from 11 can occur only when one input is ee, in which case the required comparison is one of the assumed coordinate comparisons. Hence SS is a triangular conorm, and Theorem 4.8 yields a uninorm on this proper bounded trellis.

Corollary 4.11. Under the hypotheses of Theorem 4.8, suppose that TT has no zero divisors, that is,

T(p,q)=0⟹p=0 or q=0(p,q∈[0,e]).T(p,q)=0 \quad\Longrightarrow\quad p=0\text{ or }q=0 \qquad(p,q\in[0,e]).

Then (19) is equivalent to (17). Consequently, under (C1) and the no-zero-divisor assumption, condition (C2) together with (17) is necessary and sufficient for U5U_5 to be a uninorm.

Proof. Assume first (17). If p∈[0,e)p\in[0,e) and p⊴z∈Iep\unlhd z\in I_e, then pp cannot belong to (0,e)(0,e), so p=0p=0; hence T(p,q)=T(0,q)=0T(p,q)=T(0,q)=0. Thus (19) holds. Conversely, assume (19). If p∈(0,e)p\in(0,e) and p⊴z∈Iep\unlhd z\in I_e, taking q=pq=p gives T(p,p)=0T(p,p)=0, which contradicts the absence of zero divisors. The reverse comparison z⊴pz\unlhd p is also impossible: since z∈Ie⊆Xrtrz\in I_e\subseteq X^{\mathrm{rtr}}, the chain z⊴p⊴ez\unlhd p\unlhd e would imply z⊴ez\unlhd e. Hence p∥zp\parallel z, as required. □

The converse relation between (19) and (17) is clarified by the following counterexample.

Remark 4.12. Condition (17) always implies (19), but the converse may fail when TT has zero divisors. Indeed, let X={0,p,e,z,1}X=\{0,p,e,z,1\} be the bounded lattice whose nontrivial covering relations are

0◃p◃e◃1,p◃z◃1,e∥z.0\triangleleft p\triangleleft e\triangleleft1,\qquad p\triangleleft z\triangleleft1,\qquad e\parallel z.

Take clc_l to be the identity map. Then Ie={z}I_e=\{z\} and ce(z)=z∧(z∨e)=z∧1=zc_e(z)=z\wedge(z\vee e)=z\wedge1=z, so (C1) and (C2) hold. On the three-element chain [0,e]={0,p,e}[0,e]=\{0,p,e\} define the drastic triangular norm

T(u,v)={u∧v,u=e or v=e,0,u,v∈{0,p},T(u,v)= \begin{cases} u\wedge v, & u=e\text{ or }v=e,\\ 0, & u,v\in\{0,p\}, \end{cases}

and take the triangular conorm on [e,1]={e,1}[e,1]=\{e,1\} determined by S(e,e)=eS(e,e)=e and S(u,v)=1S(u,v)=1 otherwise. Since [0,e)={0,p}[0,e)=\{0,p\}, one has T(u,v)=0T(u,v)=0 for all u,v∈[0,e)u,v\in[0,e), so (19) holds. Hence Theorem 4.8 shows that U5U_5 is a uninorm. Nevertheless, p⊴zp\unlhd z, and therefore (17) fails. Thus condition (17) is sufficient but is not necessary in general.

Theorem 4.13. Let int⁡\operatorname{int} be an interior operator, TT be a triangular norm on [0,e][0,e], and SS be a triangular conorm on [e,1][e,1], both with neutral element ee. Assume (I1)–(I3) and the condition

x∥yfor every x∈(e,1), y∈Ie.(20)x\parallel y \qquad\text{for every }x\in(e,1),\ y\in I_e. \tag*{(20)}

Define

U6(x,y)={T(x,y),x,y∈[0,e],S(x,y),x,y∈(e,1],x,(x,y)∈((e,1]×[0,e])∪(Ie×{e}),y,(x,y)∈([0,e]×(e,1])∪({e}×Ie),int⁡(x)∨(x∧e),x∈Ie, y∈[0,e),int⁡(y)∨(y∧e),y∈Ie, x∈[0,e),1,otherwise.(21)U_6(x,y)= \begin{cases} T(x,y), & x,y\in[0,e],\\ S(x,y), & x,y\in(e,1],\\ x, & (x,y)\in((e,1]\times[0,e])\cup(I_e\times\{e\}),\\ y, & (x,y)\in([0,e]\times(e,1])\cup(\{e\}\times I_e),\\ \operatorname{int}(x)\vee(x\wedge e), & x\in I_e,\ y\in[0,e),\\ \operatorname{int}(y)\vee(y\wedge e), & y\in I_e,\ x\in[0,e),\\ 1, & \text{otherwise}. \end{cases} \tag*{(21)}

Then U6U_6 is a uninorm with neutral element ee.

Proof. Consider the order-dual bounded trellis T∂T^{\partial}. The incomparable set of ee is unchanged.

With respect to T∂T^{\partial}, the map int⁡\operatorname{int} is a closure operator. Indeed, int⁡\operatorname{int} is extensive in the dual order because int⁡(x)⊑x\operatorname{int}(x)\sqsubseteq x in the original order implies x⊑∂int⁡(x)x\sqsubseteq^{\partial}\operatorname{int}(x). Conditions (I1)–(I3) are exactly conditions (C1)–(C3) in the dual pseudo-order, and the expression corresponding to cec_e is int⁡(x)∨(x∧e)=de(x)\operatorname{int}(x)\vee(x\wedge e)=d_e(x). The triangular conorm SS on the original interval [e,1][e,1] is a triangular norm on the dual interval from 11 to ee, whereas TT is the corresponding triangular conorm on the dual interval from ee to 00. Moreover, (20) is precisely (17) in the dual pseudo-order.

Applying Theorem 4.9 to these dual data gives exactly the cases displayed in (21): the lower and upper intervals are interchanged, 00 and 11 are interchanged, and cec_e is replaced by ded_e. Hence the joint increasingness and all three associativity pairings proved for U5U_5 dualize without any additional assumption. Therefore U6U_6 is a uninorm with neutral element ee. □

We next state the exact criterion for the two-sided interior construction.

Corollary 4.14. Let T\mathbb{T} be a bounded trellis, e∈X∖{0,1}e\in X\setminus\{0,1\}, int⁡\operatorname{int} be an interior operator, TT be a triangular norm on [0,e][0,e] with neutral element ee, and SS be a triangular conorm on [e,1][e,1] with neutral element ee. Assume (I1), and define U6U_6 by (21). Then U6U_6 is a uninorm if and only if (I2) holds and

z⊑p ⟹ S(p,q)=1(p,q∈(e,1], z∈Ie).(22)z\sqsubseteq p\ \Longrightarrow\ S(p,q)=1\qquad(p,q\in(e,1],\ z\in I_e). \tag*{(22)}

Under these assumptions, (I3) follows from (I2), whereas (I4) is not required. If SS has no nontrivial one-divisors, in the sense that

S(p,q)=1 ⟹ p=1 or q=1(p,q∈[e,1]),S(p,q)=1\ \Longrightarrow\ p=1\ \text{or}\ q=1\qquad(p,q\in[e,1]),

then (22) is equivalent to (20). Consequently, under (I1) and the absence of nontrivial one-divisors, condition (I2) together with (20) is necessary and sufficient for U6U_6 to be a uninorm.

Proof. Apply Theorem 4.8 and Theorem 4.11 to the order-dual bounded trellis. Under this duality, (C1)–(C3), condition (19), and the absence of zero divisors become (I1)–(I3), condition (22), and the absence of nontrivial one-divisors, respectively. This gives all the stated equivalences. □

BOUNDED-LATTICE REDUCTIONS AND SCOPE

Conditions (C1)–(C4) and (I1)–(I4) are formulated for trellises because ordinary lattice manipulations cannot be used without transitivity and associativity. On a bounded lattice, these conditions reduce to explicit identities and restrictions on the operator images. Combined with the characterizations in § 4, these reductions yield exact lattice criteria for all six constructions.

Proposition 5.1. Let LL be a bounded lattice and let e∈L∖{0,1}e \in L \setminus\{0,1\}.

  1. If cl⁡\operatorname{cl} is a closure operator, then (C1)–(C4) hold if and only if x∧e=cl⁡(x)∧ex \wedge e = \operatorname{cl}(x) \wedge e for all x∈Iex \in I_e.

  1. If int⁡\operatorname{int} is an interior operator, then (I1)–(I4) hold if and only if x∨e=int⁡(x)∨ex \vee e = \operatorname{int}(x) \vee e for all x∈Iex \in I_e.

  1. If cl⁡\operatorname{cl} is a closure operator, then the following statements are equivalent:

(a) conditions (C1)–(C3) hold;

(b) cl⁡(x)∥e\operatorname{cl}(x) \parallel e for every x∈Iex \in I_e;

(c) cl⁡(x)∉[e,1]\operatorname{cl}(x) \notin[e,1] for every x∈Iex \in I_e.

Dually, for an interior operator int⁡\operatorname{int}, one has

(I1)–(I3) hold⟺int⁡(x)∥e(x∈Ie)⟺int⁡(x)∉[0,e](x∈Ie).\begin{aligned} \text{(I1)–(I3) hold} &\Longleftrightarrow\operatorname{int}(x) \parallel e && (x \in I_e)\\ &\Longleftrightarrow\operatorname{int}(x) \notin[0,e] && (x \in I_e). \end{aligned}

Proof. Every element of a lattice is right-, left-, and middle-transitive, and lattice meet and join are associative and jointly increasing. Thus (C1) and its order dual hold automatically.

For (1), let x∈Iex \in I_e and put ce(x)=cl⁡(x)∧(x∨e)c_e(x) = \operatorname{cl}(x) \wedge(x \vee e). Since x≤ce(x)x \leq c_e(x), the inequality ce(x)≤ec_e(x) \leq e would imply x≤ex \leq e. If e≤ce(x)e \leq c_e(x), then

e=ce(x)∧e=cl⁡(x)∧(x∨e)∧e=cl⁡(x)∧e=x∧e,e = c_e(x) \wedge e = \operatorname{cl}(x) \wedge(x \vee e) \wedge e = \operatorname{cl}(x) \wedge e = x \wedge e,

which would imply e≤xe \leq x. Both alternatives contradict x∥ex \parallel e, so ce(x)∈Iec_e(x) \in I_e. Monotonicity of cec_e follows from that of cl⁡\operatorname{cl}, meet, and join. Moreover, x≤ce(x)≤cl⁡(x)x \leq c_e(x) \leq\operatorname{cl}(x) and x≤ce(x)≤x∨ex \leq c_e(x) \leq x \vee e, whence cl⁡(ce(x))=cl⁡(x)\operatorname{cl}(c_e(x)) = \operatorname{cl}(x) and ce(x)∨e=x∨ec_e(x) \vee e = x \vee e. Therefore ce(ce(x))=ce(x)c_e(c_e(x)) = c_e(x), proving (C2). Joint increasingness of mem_e follows from that of meet, and

me(ce(x),y)=cl⁡(x)∧(x∨e)∧y∧e=x∧y∧e=me(x,y),m_e(c_e(x),y) = \operatorname{cl}(x) \wedge(x \vee e) \wedge y \wedge e = x \wedge y \wedge e = m_e(x,y),

so (C4) holds. Finally, if x≤yx \leq y and y>ey > e, then ce(x)≤x∨e≤yc_e(x) \leq x \vee e \leq y, which is (C3). This proves the sufficiency direction in (1).

Conversely, suppose that (C1)–(C4) hold. For x∈Iex \in I_e put c=ce(x)c = c_e(x). Condition (C2) gives c∈Iec \in I_e, so (C4) may be applied with y=cy = c. Since x≤cx \leq c, lattice absorption and associativity of meet give c∧e=me(c,c)=me(x,c)=x∧ec \wedge e = m_e(c,c) = m_e(x,c) = x \wedge e. On the other hand,

c∧e=[cl⁡(x)∧(x∨e)]∧e=cl⁡(x)∧e,c \wedge e = [\operatorname{cl}(x) \wedge(x \vee e)] \wedge e = \operatorname{cl}(x) \wedge e,

because e≤x∨ee \leq x \vee e. Hence x∧e=cl⁡(x)∧ex \wedge e = \operatorname{cl}(x) \wedge e, proving the necessity direction. Assertion (2), including both directions, is its order dual.

For (3), fix x∈Iex \in I_e. Extensivity gives x≤cl⁡(x)x \leq\operatorname{cl}(x), so cl⁡(x)≤e\operatorname{cl}(x) \leq e would imply x≤ex \leq e, which is impossible. Consequently, cl⁡(x)\operatorname{cl}(x) can be comparable with ee only in the direction e≤cl⁡(x)e \leq\operatorname{cl}(x). Since every element is below 11, this proves the equivalence cl⁡(x)∥e⟺cl⁡(x)∉[e,1]\operatorname{cl}(x) \parallel e \Longleftrightarrow\operatorname{cl}(x) \notin[e,1].

Suppose that cl⁡(x)∥e\operatorname{cl}(x) \parallel e for every x∈Iex \in I_e, and put c=ce(x)c = c_e(x). As before, x≤cx \leq c, so c≤ec \leq e is impossible. If e≤ce \leq c, then c≤cl⁡(x)c \leq\operatorname{cl}(x) would give e≤cl⁡(x)e \leq\operatorname{cl}(x), again a contradiction. Hence c∈Iec \in I_e. The map cec_e is increasing because closure, meet, and join are increasing on a lattice. Moreover, x≤ce(x)≤cl⁡(x)x \leq c_e(x) \leq\operatorname{cl}(x) and x≤ce(x)≤x∨ex \leq c_e(x) \leq x \vee e, so closure idempotence and lattice absorption give cl⁡(ce(x))=cl⁡(x)\operatorname{cl}(c_e(x)) = \operatorname{cl}(x) and ce(x)∨e=x∨ec_e(x) \vee e = x \vee e. It follows that ce(ce(x))=ce(x)c_e(c_e(x)) = c_e(x), proving (C2). If x≤yx \leq y with y>ey > e, then yy is an upper bound of both xx and ee; thus ce(x)≤x∨e≤yc_e(x) \leq x \vee e \leq y, which is (C3). Condition (C1) is automatic on a lattice, and hence (C1)–(C3) hold.

Conversely, suppose that (C1)–(C3) hold. By (C2), ce(x)∈Iec_e(x) \in I_e. If e≤cl⁡(x)e \leq\operatorname{cl}(x), then ee is a lower bound of both cl⁡(x)\operatorname{cl}(x) and x∨ex \vee e, whence e≤cl⁡(x)∧(x∨e)=ce(x)e \leq\operatorname{cl}(x) \wedge(x \vee e) = c_e(x), contrary to ce(x)∥ec_e(x) \parallel e. Therefore e≰cl⁡(x)e \nleq\operatorname{cl}(x), and the first paragraph shows that cl⁡(x)∥e\operatorname{cl}(x) \parallel e, or equivalently cl⁡(x)∉[e,1]\operatorname{cl}(x) \notin[e,1]. This completes the closure part of (3). The interior statement follows by order duality. □

Corollary 5.2. Let LL be a bounded lattice, let e∈L∖{0,1}e \in L \setminus\{0,1\}, let SS be a triangular conorm on [e,1][e,1] with neutral element ee, and let cl⁡\operatorname{cl} be a closure operator on LL. Define U1U_1 by (5). Then the following statements are equivalent:

  1. x∧e=cl⁡(x)∧ex \wedge e = \operatorname{cl}(x) \wedge e for every x∈Iex \in I_e;

  2. conditions (C1)–(C4) hold;

  3. U1U_1 is a uninorm on LL with neutral element ee;

  4. U1U_1 is associative.

Thus the closure equality is necessary and sufficient for this bounded-lattice construction.

Proof. The equivalence of (1) and (2) is Theorem 5.1, and (2) implies (3) by Theorem 4.1. Clearly (3) implies (4). It remains to show that (4) implies (1).

Fix x∈Iex \in I_e and write c=ce(x)c = c_e(x). Since x≤cx \leq c, one cannot have c≤ec \leq e. Suppose that c>ec > e. Associativity at (x,1,1)(x,1,1) gives S(c,1)=cS(c,1) = c, and hence c=1c = 1. Associativity at (x,x,1)(x,x,1) then gives

x∧e=U1(U1(x,x),1)=U1(x,U1(x,1))=1.x \wedge e = U_1(U_1(x,x),1) = U_1(x,U_1(x,1)) = 1.

which is impossible because x∥ex \parallel e. Thus c∈Iec \in I_e. Applying associativity to (1,x,c)(1,x,c) gives

c∧e=U1(U1(1,x),c)=U1(1,U1(x,c))=x∧e,c \wedge e = U_1(U_1(1,x),c) = U_1(1,U_1(x,c)) = x \wedge e,

where x≤cx \leq c was used in the last equality. Finally,

c∧e=[cl⁡(x)∧(x∨e)]∧e=cl⁡(x)∧e.c \wedge e = [\operatorname{cl}(x) \wedge(x \vee e)] \wedge e = \operatorname{cl}(x) \wedge e.

Therefore x∧e=cl⁡(x)∧ex \wedge e = \operatorname{cl}(x) \wedge e, as required. □

Remark 5.3. Ouyang and Zhang [20, Theorem 4.1] gave a sufficient construction on a bounded lattice from a closure operator and a triangular norm. Çaylı [4, Theorems 3.1 and 3.11] studied two further lattice formulas and obtained necessary and sufficient conditions. Qi and Xiu [22] instead start from a uninorm on a subinterval and extend it by means of a closure operator and a triangular superconorm, while Xiu and Si [27] consider closure operators together with triangular conorms. These papers are close in method, but their input data and defining clauses are not identical in general.

The preceding corollary concerns only the operation in (5). For this formula, it shows that the displayed closure equality is not merely sufficient but also necessary. It neither classifies all uninorms on a bounded lattice nor subsumes all the constructions cited above. On a proper trellis, moreover, the ambient hypothesis in Theorem 4.1 cannot in general be recovered from the lattice equality, as shown by the five-element example.

Corollary 5.4. Let LL be a bounded lattice, let e∈L∖{0,1}e \in L \setminus\{0,1\}, let TT be a triangular norm on [0,e][0,e] with neutral element ee, and let int⁡\operatorname{int} be an interior operator on LL. Define U2U_2 by (8). Then the following statements are equivalent:

  1. x∨e=int⁡(x)∨ex \vee e = \operatorname{int}(x) \vee e for every x∈Iex \in I_e;

  2. conditions (I1)–(I4) hold;

  3. U2U_2 is a uninorm on LL with neutral element ee;

  4. U2U_2 is associative.

Proof. Apply Theorem 5.2 to the order-dual lattice. Closure, meet, triangular conorm, cec_e, mem_e, and (5) become, respectively, interior, join, triangular norm, ded_e, nen_e, and (8). This gives all four equivalences. □

The constructions with an absorbing endpoint have the following corresponding lattice criteria.

Corollary 5.5. Let LL be a bounded lattice and let e∈L∖{0,1}e \in L \setminus\{0,1\}.

  1. Let cl⁡\operatorname{cl} be a closure operator, let SS be a triangular conorm on [e,1][e,1] with neutral element ee, and define U3U_3 by (9). If (e,1)≠∅(e,1) \ne\varnothing, then U3U_3 is a uninorm on LL with neutral element ee if and only if cl⁡(x)∧e=x∧e\operatorname{cl}(x) \wedge e = x \wedge e for every x∈Iex \in I_e and S(x,y)≠1S(x,y) \ne1 for all x,y∈(e,1)x,y \in(e,1). If (e,1)=∅(e,1) = \varnothing, then U3U_3 is a uninorm for every closure operator cl⁡\operatorname{cl}.

  2. Dually, let int⁡\operatorname{int} be an interior operator, let TT be a triangular norm on [0,e][0,e] with neutral element ee, and define U4U_4 by (15). If (0,e)≠∅(0,e) \ne\varnothing, then U4U_4 is a uninorm on LL with neutral element ee if and only if int⁡(x)∨e=x∨e\operatorname{int}(x) \vee e = x \vee e for every x∈Iex \in I_e and T(x,y)≠0T(x,y) \ne0 for all x,y∈(0,e)x,y \in(0,e). If (0,e)=∅(0,e) = \varnothing, then U4U_4 is a uninorm for every interior operator int⁡\operatorname{int}.

Proof. Every element of LL satisfies (C1), and mem_e is jointly increasing. Suppose first that (e,1)≠∅(e,1) \ne\varnothing. By Theorem 4.5, the uninorm property of U3U_3 is equivalent to (C2), (C4), and condition (11). Under (C1), condition (C3) follows from (C2), while Theorem 5.1(1) identifies the conjunction of (C2) and (C4) with the displayed closure equality. This proves the first equivalence. If (e,1)=∅(e,1)=\varnothing, Theorem 4.5 reduces the criterion to joint increasingness of mem_e, which is automatic on a lattice. The second assertion is the order dual, using Theorem 4.7 and Theorem 5.1[2]. □

The following example treats the case in which the upper interval is degenerate.

Example 5.6. Let L=P({1,2,3})L=\mathcal{P}(\{1,2,3\}) be ordered by inclusion, with 0=∅0=\varnothing, 1={1,2,3}1=\{1,2,3\}, and e={1,2}e=\{1,2\}. Then (e,1)=∅(e,1)=\varnothing and Ie={{3},{1,3},{2,3}}I_e=\{\{3\},\{1,3\},\{2,3\}\}. Define the closure operator by cl⁡(A)=1\operatorname{cl}(A)=1 for every A∈LA\in L. For each x∈Iex\in I_e one has ce(x)=cl⁡(x)∩(x∪e)=1c_e(x)=\operatorname{cl}(x)\cap(x\cup e)=1, so the image requirement in (C2) fails. Nevertheless, Theorem 5.5 shows that U3U_3 is a uninorm. Thus the nonemptiness of (e,1)(e,1) is essential when the uninorm property of U3U_3 is used to recover conditions (C1)–(C4); its order dual gives the corresponding boundary for U4U_4.

The two-sided constructions reduce to the following lattice criteria.

Corollary 5.7. Let LL be a bounded lattice, let e∈L∖{0,1}e\in L\setminus\{0,1\}, let TT be a triangular norm on [0,e][0,e], and let SS be a triangular conorm on [e,1][e,1], both with neutral element ee.

  1. Let cl⁡\operatorname{cl} be a closure operator and define U5U_5 by (18). Then U5U_5 is a uninorm on LL with neutral element ee if and only if cl⁡(x)∥e\operatorname{cl}(x)\parallel e for every x∈Iex\in I_e and condition (19) holds; explicitly,

p≤z,p,q∈[0,e),z∈Ie⟹T(p,q)=0.p\le z,\quad p,q\in[0,e),\quad z\in I_e \quad\Longrightarrow\quad T(p,q)=0.

Equivalently, the first displayed condition may be replaced by cl⁡(x)∉[e,1]\operatorname{cl}(x)\notin[e,1] for every x∈Iex\in I_e. If TT has no zero divisors, then condition (19) may equivalently be replaced by (17).

  1. Dually, let int⁡\operatorname{int} be an interior operator and define U6U_6 by (21). Then U6U_6 is a uninorm on LL with neutral element ee if and only if int⁡(x)∥e\operatorname{int}(x)\parallel e for every x∈Iex\in I_e and condition (22) holds; explicitly,

z≤p,p,q∈(e,1],z∈Ie⟹S(p,q)=1.z\le p,\quad p,q\in(e,1],\quad z\in I_e \quad\Longrightarrow\quad S(p,q)=1.

Equivalently, the first displayed condition may be replaced by int⁡(x)∉[0,e]\operatorname{int}(x)\notin[0,e] for every x∈Iex\in I_e. If SS has no nontrivial one-divisors, then condition (22) may equivalently be replaced by (20).

Proof. Condition (C1) is automatic on LL. By Theorem 4.8, U5U_5 is a uninorm if and only if (C2) and (19) hold. Under (C1), condition (C2) implies (C3); hence (C2) is equivalent here to (C1)–(C3). Part (3) of Theorem 5.1 identifies these conditions with cl⁡(x)∥e\operatorname{cl}(x)\parallel e, equivalently cl⁡(x)∉[e,1]\operatorname{cl}(x)\notin[e,1], on IeI_e. This proves the first assertion. The second assertion is its order dual and follows from Theorem 4.14. The final equivalences follow from Theorem 4.11 and its stated order-dual counterpart. □

CONCLUSIONS

We have obtained necessary and sufficient conditions for six constructions on bounded trellises. Under the ambient transitivity condition, the first closure-based construction is characterized by (C2) and (C4), while (C3) follows from the other assumptions; the interior-based result is its order dual. For the variants with an absorbing endpoint, the nondegenerate cases require (10) or (16), whereas in the degenerate cases the closure or interior map no longer occurs in the effective formula.

For the constructions involving both a triangular norm and a triangular conorm, the necessary and sufficient requirements are (19) and its order dual (22). The stronger incomparability assumptions (17) and (20) are equivalent to these requirements when the triangular norm has no zero divisors or the triangular conorm has no nontrivial one-divisors. On bounded lattices, the closure and interior parts of the criteria reduce to explicit identities or image restrictions.

The examples also show the limitation of these reductions: a lattice identity need not imply the ambient transitivity condition on a proper trellis, and that condition need not be necessary for a particular operation to be a uninorm. Further work may determine weaker ambient hypotheses and compare the resulting classes with other uninorm constructions on bounded trellises.

DECLARATION OF COMPETING INTEREST

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

DATA AVAILABILITY

No datasets were generated or analysed in this theoretical study.

ACKNOWLEDGEMENT

This work was supported by the National Natural Science Foundation of China (Nos. 12671553 and 12271036).

Zhenyu Xiu, Corresponding author. School of Mathematical Science, Guangxi Minzu University, Nanning 530006, Guangxi, China

e-mail: [email protected]

References

  1. [1]S. Bodjanova, M. Kalina, Construction of uninorms on bounded lattices, in: IEEE 12th International Symposium on Intelligent Systems and Informatics, SISY 2014, 2014, pp. 61–66.DOI
  2. [2]G.D. Çaylı, F. Karaçal, R. Mesiar, On a new class of uninorms on bounded lattices, Inf. Sci. 367–368 (2016) 221–231.DOI
  3. [3]G.D. Çaylı, New methods to construct uninorms on bounded lattices, Int. J. Approx. Reason. 115 (2019) 254–264.DOI
  4. [4]G.D. Çaylı, A characterization of uninorms on bounded lattices via closure and interior operators, Kybernetika 59 (2023) 768–790. doi:10.14736/kyb-2023-5-0768.DOI
  5. [5]Y. Dan, B.Q. Hu, J. Qiao, New constructions of uninorms on bounded lattices, Int. J. Approx. Reason. 110 (2019) 185–209.DOI
  6. [6]B.A. Davey, H.A. Priestley, Introduction to Lattices and Order, 2nd ed., Cambridge University Press, Cambridge, 2002. doi:10.1017/CBO9780511809088.DOI
  7. [7]P. Drygás, D. Ruiz-Aguilera, J. Torrens, A characterization of a class of uninorms with continuous underlying operators, Fuzzy Sets Syst. 287 (2016) 137–153.DOI
  8. [8]J.C. Fodor, R.R. Yager, A. Rybakov, Structure of uninorms, Int. J. Uncertain. Fuzziness Knowl.-Based Syst. 5 (1997) 411–427.DOI
  9. [9]E. Fried, Tournaments and non-associative lattices, Ann. Univ. Sci. Bp. Rolando Eötvös Nomin., Sect. Math. 13 (1970) 151–164.
  10. [10]K. Gladstien, A characterization of complete trellises of finite length, Algebra Univers. 3 (1973) 341–344.DOI
  11. [11]J.J.A. Goguen, L-fuzzy sets, J. Math. Anal. Appl. 18 (1967) 145–174.DOI
  12. [12]J.J.A. Goguen, The fuzzy Tychonoff theorem, J. Math. Anal. Appl. 43 (1973) 734–742.DOI
  13. [13]F. Karaçal, R. Mesiar, Uninorms on bounded lattices, Fuzzy Sets and Systems 261 (2015) 33–43.DOI
  14. [14]Y. Kong, B. Zhao, Uninorms on bounded trellises, Fuzzy Sets and Systems 481 (2024) 108898. doi:10.1016/j.fss.2024.108898.DOI
  15. [15]G. Li, H.W. Liu, On a characterization of representable uninorms, Fuzzy Sets Syst. 408 (2021) 57–64.DOI
  16. [16]Y. Luo, B. Pang, J.-C. Yao, Constructions of uninorms on bounded trellises, Fuzzy Sets and Systems 521 (2025) 109576. doi:10.1016/j.fss.2025.109576.DOI
  17. [17]A. Mehenni, L. Zedam, Pseudo triangular norms on bounded trellises, arXiv:2212.03605 (2022).arxiv.org/abs/2212.03605
  18. [18]A. Mesiarová-Zemánková, Ordinal sums of representable uninorms, Fuzzy Sets Syst. 308 (2017) 42–53.
  19. [19]A. Mesiarová-Zemánková, Uninorms internal on one or more non-trivial cuts, Inf. Sci. 653 (2024) 119793.
  20. [20]Y. Ouyang, H.-P. Zhang, Constructing uninorms via closure operators on a bounded lattice, Fuzzy Sets and Systems 395 (2020) 93–106. doi:10.1016/j.fss.2019.05.006.
  21. [21]F. Qin, L. Fu, A characterization of uninorms not internal on the boundary, Fuzzy Sets Syst. 469 (2023) 108641.DOI
  22. [22]J. Qi, Z.-Y. Xiu, New methods to construct uninorms by extending uninorms with closure operators and t-superconorms, Kybernetika 61 (2025) 855–871. doi:10.14736/kyb-2025-6-0855.DOI
  23. [23]H.L. Skala, Trellis theory, Algebra Universalis 1 (1971) 218–233. doi:10.1007/BF02944982.DOI
  24. [24]H. Song, B. Zhao, New constructions of uninorms on bounded trellises, Fuzzy Sets and Systems 537 (2026) 109902. doi:10.1016/j.fss.2026.109902.DOI
  25. [25]X.R. Sun, H.W. Liu, Further characterization of uninorms on bounded lattices, Fuzzy Sets Syst. 427 (2022) 96–108.DOI
  26. [26]Y.-M. Wang, Y. Dan, B. De Baets, On the classes of uninorms U_min and U_max on bounded trellises, International Journal of Approximate Reasoning 191 (2026) 109627. doi:10.1016/j.ijar.2026.109627.
  27. [27]Z.-Y. Xiu, Z.-Y. Si, New methods to construct uninorms on bounded lattices via closure operators and triangular conorms, Computational and Applied Mathematics 45 (2026) 12. doi:10.1007/s40314-025-03375-9.DOI
  28. [28]Z. Xiu, X. Zheng, Nullnorms on bounded trellises, Fuzzy Sets Syst. 521 (2025) 109599.arxiv.org/abs/2408.09321
  29. [29]R.R. Yager, A. Rybalov, Uninorm aggregation operators, Fuzzy Sets and Systems 80 (1996) 111–120.DOI
  30. [30]L. Zedam, B. De Baets, Triangular norms on bounded trellises, Fuzzy Sets and Systems 462 (2023) 108468. doi:10.1016/j.fss.2023.01.003.DOI
  31. [31]H.P. Zhang, M. Wu, Z. Wang, Y. Ouyang, B. De Baets, A characterization of the classes U_min and U_max of uninorms on a bounded lattice, Fuzzy Sets Syst. 423 (2021) 107–121.
  32. [32]Y.F. Zhao, H.W. Liu, New constructions of t-norms and t-conorms on bounded lattices and trellises, Int. J. Gen. Syst. 55 (2026) 70–90. doi:10.1080/03081079.2025.2457559.
  33. [33]B. Zhao, T. Wu, Some further results about uninorms on bounded lattices, Int. J. Approx. Reason. 130 (2021) 22–49.

Paper details

vlnc:W_w6hbt3pj
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY-SA 4.0
Usage

views · 9

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents