On distributive join-semilattices

TL;DR

This paper introduces and explores a notion of distributivity in join-semilattices inspired by logical rules, establishing equivalences and extending to structures with implications relevant to intuitionistic logic.

Contribution

It defines a new distributivity concept for join-semilattices, proves its equivalence to existing notions, and extends the concept to structures with arrow, linking algebraic and logical frameworks.

Findings

Distributivity notions are linearly ordered.

The new notion is equivalent to a known concept.

Extension to join-semilattices with arrow connects to intuitionistic logic.

Abstract

Motivated by Gentzen disjunction elimination rule in his Natural Deduction calculus and reading inequalities with meet in a natural way, we conceive a notion of distributivity for join-semilattices. We prove that it is equivalent to a notion present in the literature. In the way, we prove that those notions are linearly ordered. We finally consider the notion of distributivity in join-semilattices with arrow, that is, the algebraic structure corresponding to the disjunction-conditional fragment of intuitionistic logic.

Equations

$$I\vee J=\{a\vee b:a\in I,b\in J\}.$$

$$\{(c,d)\in B:\forall (x,y)\in B,\text{mbox}if(x,y)\le (c,d)\text{mbox}and(x,y)\le (\frac{1}{2},\frac{1}{2})\text{mbox},then(x,y)\le (0,0)\}$$

Full text

1 Introduction

Different notions of distributivity for semilattices have been proposed in the literature as a generalization of the usual distributive property in lattices. As far as we know, notions of distributivity for semilattices have been given, in chronological order, by Grätzer and Schmidt [8] in 1962, by Katriňák [11] in 1968, by Balbes [1] in 1969, by Schein[14] in 1972, by Hickman [10] in 1984, and by Larmerová and Rachůnek [13] in 1988. Following the names of its authors, we will use the terminology GS-, K-, B-, Sn-, H-, and LR-distributivity, respectively.

In this paper, motivated by Gentzen’s disjunction elimination rule in his Natural Deduction calculus, and reading inequalities with meet in a natural way, we conceive another notion of distributivity for join-semilattices, that we call ND-distributivity. We aim to find out whether it is equivalent to any of the notions already present in the literature. In doing so, we also compare the different notions of distributivity for join-semilattices we have found. Namely, we see that the given notions imply each other in the following linear order:

GS $\Rightarrow$ K $\Rightarrow$ (H $\Leftrightarrow$ LR $\Leftrightarrow$ ND) $\Rightarrow$ B $\Rightarrow\cdots$ S${}_{n}\Rightarrow$ S${}_{n-1}\Rightarrow\cdots$ S${}_{3}\Rightarrow$ S2,

and we also provide countermodels for the reciprocals.

Additionally, we show that H-distributivity may be seen as a very natural translation of a way to define distributivity for lattices, fact that will provide more motivation for the use of that notion. Note that Hickman used the term mild distributivity for H-distributivity.

The paper is structured as follows. After this introduction, in Section 2 we provide some notions and notations that will be used in the paper. In Section 3 we show how to arrive to our notion of ND-distributivity for join-semilattices. In Section 4 we compare the different notions of distributivity for join-semilattices that appear in the literature. We prove that one of those is equivalent to the notion of ND-distributivity found in Section 3. Finally, in Section 5 we consider what happens with the different notions of distributivity considered in Section 4 when join-semilattices are expanded with a natural version of the relative meet-complement.

2 Preliminaries

In this section we provide the basic notions and notations that will be used in the paper.

Let ${\bf{J}}=(J;\leq)$ be a poset. For any $S\subseteq J$, we will use the notations $S^{l}$ and $S^{u}$ to denote the set of lower and upper bounds of $S$, respectively. That is,

• $S^{l}=\{x\in J:x\leq s$, for all $s\in S\}$ and

$S^{u}=\{x\in J:s\leq x$, for all $s\in S\}$.

Lemma 1**.**

Let ${\bf{J}}=(J;\leq)$ be a poset. For all $a,b,c\in J$ the following statements are equivalent:

• (i)

for all $x\in J$, if $x\leq a$ and $x\leq b$, then $x\leq c$,

• (ii)

$\{a,b\}^{l}\subseteq\{c\}^{l}$,

• (iii)

$c\in\{a,b\}^{lu}$.

A poset ${\bf{J}}=(J;\leq)$ is a join-semilattice (resp. meet-semilattice) if $\sup\{a,b\}$ (resp. $\inf\{a,b\}$) exists for every $a,b\in J$. A poset ${\bf{J}}=(J;\leq)$ is a lattice if it is both a join- and a meet-semilattice. As usual, the notations $a\vee b$ (resp. $a\wedge b$) shall stand for $\sup\{a,b\}$ (resp. $\inf\{a,b\}$).

Given a join-semilattice ${\bf{J}}=(J;\leq)$, we will use the following notions:

• •

${\bf{J}}$ is downwards directed iff for any $a,b\in J$, there exists $c\in J$ such that $c\leq a$ and $c\leq b$.

• •

A non empty subset $I\subseteq J$ is said to be an ideal iff

(1) if $x,y\in I$, then $x\vee y\in I$ and

(2) If $x\in I$ and $y\leq x$, then $y\in I$.

• •

The principal ideal generated by an element $a\in A$, noted $(a]$, is defined by $(a]=\{x\in A:x\leq a\}$.

• •

$Id({\bf{J}})$ will denote the set of all ideals of ${\bf{J}}$.

• •

$Id_{fp}({\bf{J}})$ will denote the subset of ideals that are intersection of a finite set of principal ideals, that is, $Id_{fp}({\bf{J}})=\{(a_{1}]\cap\dots\cap(a_{k}]:a_{1},...a_{k}\in J\}$.

In this paper we are concerned with various notions of distributivity for join-semilattices, all of them generalizing the usual notion of distributive lattice, that is, a lattice ${\bf{J}}=(J;\leq)$ is distributive if the following equation holds true for any elements $a,b,c\in J$:

• (D)

$a\land(b\lor c)=(a\land b)\lor(a\land c)$ (equivalently, $a\lor(b\land c)=(a\lor b)\land(a\lor c)$).

There are several equivalent formulations of this property, in particular we mention the following ones that are relevant for this paper:

• •

for all $a,b,c\in J$, if $a\lor b=a\lor c$ and $a\land b=a\land c$ then $b=c$.

• •

for any two ideals $I_{1},I_{2}$ of $\bf J$, the ideal $I_{1}\lor I_{2}$ generated by their union is defined by $I_{1}\lor I_{2}=\{a\lor b:a\in I_{1},b\in I_{2}\}.$

• •

the set $Id({\bf{J}})$ of ideals of $\bf J$ is a distributive lattice.

For the case of semilattices, several non-equivalent generalizations of these conditions can be found in the literature, already mentioned in the introduction. However, as expected, all of them turn to be equivalent to usual distributivity in the case of lattices.

The class of distributive lattices form a variety (that is, an equational class). In contrast, in any sense of distributivity for join-semilattices that coincides with usual distributivity in the case of a lattice, the class of distributive join-semilattices is not even a quasi-variety. Indeed, consider the distributive lattice in Figure 1. Taken as a join-semilattice, the set of black-filled nodes is a sub join-semilattice, that is clearly a non-distributive lattice (a diamond). Thus, it is neither distributive as a join-semilattice. This proves that the class of distributive (in any sense that coincides with usual distributivity in the case of a lattice) join-semilattices is not closed by subalgebras, and hence it is not a quasi-variety.

3 Distributivity and Natural Deduction

Let us consider the disjunction-fragment of intuitionistic logic in the context of Gentzen’s Natural Deduction calculus (see [5, p. 186]). It has the following introduction rule for $\vee$ and an analogous one with $\mathfrak{B}$ as only premiss:

$\mathfrak{A}$

($\vee$I):

$\mathfrak{A\lor B}$

and the following disjunction elimination rule:

$\mathfrak{A\lor B}$       

[$\mathfrak{A}$]

$\mathfrak{C}$

     

[$\mathfrak{B}$]

$\mathfrak{C}$

$\mathfrak{C}$

The last rule may be read as saying that if $\mathfrak{C}$ follows from $\mathfrak{A}$ and $\mathfrak{C}$ follows from $\mathfrak{B}$, then $\mathfrak{C}$ follows from $\mathfrak{A}\vee B$, so reflecting what is usually called “proof by cases”. It is possible to give an algebraic translation in the context of a join-semilattice ${\bf{J}}=(J;\leq)$:

for all $a,b,c\in J$, if $a\leq c$ and $b\leq c$, then $a\vee b\leq c$,

which is easily seen to be one of the conditions stating that $\vee$ is the supremum of $a$ and $b$. Now, the last rule is usually employed in a context with a fourth formula $\mathfrak{H}$:

$\mathfrak{H}$, $\mathfrak{A\vee B}$       

$\mathfrak{H}$, [$\mathfrak{A}$]

$\mathfrak{C}$

     

$\mathfrak{H}$, [$\mathfrak{B}$]

$\mathfrak{C}$

($\vee$E):

$\mathfrak{C}$

In the context of a lattice ${\bf{L}}=(L;\leq)$, we would give the following algebraic translation:

• (D∧∨)**

for all $h,a,b,c\in L$,

if $h\wedge a\leq c$ and $h\wedge b\leq c$, then $h\wedge(a\vee b)\leq c$.

It is easily seen that (D∧∨)** is equivalent to the usual notion of distributivity for lattices. Now, the natural question arises how to give an algebraic translation of ($\vee$E) if only $\vee$ is available, for example, if we are in the context of a join-semilattice.

Considering that an inequality $u\land v\leq w$ in a lattice ${\bf{L}}=(L;\leq)$ is equivalently expressed as the first order statement

for all $x\in L$, if $x\leq u$ and $x\leq v$ then $x\leq w$,

we may write (D∧∨)** in the context of a join-semilattice ${\bf{J}}=(J;\leq)$ as follows:

• (D∨)**

for all $h,a,b,c\in J$,

IF for all $x\in J$ (if $x\leq h$ and $x\leq a$, then $x\leq c)$ and

for all $x\in J$ (if $x\leq h$ and $x\leq b$, then $x\leq c)$,

THEN for all $x\in J$ (if $x\leq h$ and $x\leq a\vee b$, then $x\leq c$).

Alternatively, using the equivalence between parts (i) and (ii) in Lemma 1, we may write

• (D∨)**

for all $h,a,b,c\in J$,

if $\{h,a\}^{l}\cup\{h,b\}^{l}\subseteq\{c\}^{l}$, then $\{h,a\vee b\}^{l}\subseteq\{c\}^{l}$.

Yet, using the equivalence between parts (ii) and (iii) in Lemma 1, we may also alternatively write

• (D∨)**

for all $h,a,b,c\in J$,

if $c\in\{h,a\}^{lu}\cup\{h,b\}^{lu}$, then $c\in\{h,a\vee b\}^{lu}$.

Accordingly, given the above logical motivation, it is natural to consider the following notion of distributivity for join-semilattices.

Definition 1**.**

A join-semilattice ${\bf{J}}=(J;\leq)$ is called ND-distributive (ND for Natural Deduction) if it satisfies **(D∨)**.

Now, it happens that there are many different (and non-equivalent) notions of distributivity for semilattices. This is not new:

“The concept of distributivity permits different non-equivalent generalizations from lattices to semilattices.” (see [14])

So, it is natural to inquire whether the given notion of ND-distributivity for join-semilattices is equivalent to any of the notions already present in the literature and, if so, to which. In what follows we will solve that question. In doing so, we will also compare the different notions of distributivity for join-semilattices that we have found.

In this paper, given our logical motivation, we restrict ourselves to study the distributivity property in join-semilattices, but an analogous path could be followed for meet-semilattices or even for posets.

Remark 1**.**

Let us note that the following rule (reflecting proof by three cases) is equivalent to ($\vee$E):

$\mathfrak{H}$, $\mathfrak{A\vee B\vee C}$       

$\mathfrak{H}$, [$\mathfrak{A}$]

$\mathfrak{D}$

     

$\mathfrak{H}$, [$\mathfrak{B}$]

$\mathfrak{D}$

     

$\mathfrak{H}$, [$\mathfrak{C}$]

$\mathfrak{D}$

$\mathfrak{D}$

Indeed, it implies ($\vee$E) taking $\mathfrak{C}=\mathfrak{B}$. Also, the following derivation shows that it may be derived using ($\vee$E) twice:

$\mathfrak{H}$, $\mathfrak{A\vee B\vee C}$       

$\mathfrak{H}$, [$\mathfrak{A}$]

$\mathfrak{D}$

     

$\mathfrak{H}$, $\mathfrak{B\vee C}$       

$\mathfrak{H}$, [$\mathfrak{B}$]

$\mathfrak{D}$

     

$\mathfrak{H}$, [$\mathfrak{C}$]

$\mathfrak{D}$

$\mathfrak{D}$

$\mathfrak{D}$

4 Different notions of distributivity for join-semilattices

In the following subsections we consider and compare the notions of distributivity for semilattices we have found in the literature. Some authors have presented their notion for the case of meet-semilattices and others for join-semilattices. We will make things uniform and, motivated by the logical considerations in the previous section, we will choose to consider join-semilattices.

We emphasize that all the distributivity notions for semilattices (and posets) proposed in the literature are generalizations of the distributivity property for lattices, in fact, when restricted to lattices all these notions coincide.

4.1 GS-distributivity

The following seems to be the most popular definition of distributivity for join-semilattices.

Definition 2**.**

A join-semilattice ${\bf{J}}=(J;\leq)$ is GS-distributive iff

• (GS)

for all $a,b,x\in J$, if $x\leq a\vee b$, then there exist $a^{\prime},b^{\prime}\in J$ such that $a^{\prime}\leq a$, $b^{\prime}\leq b$, and $x=a^{\prime}\vee b^{\prime}$.

In order to visualize it, see Figure 2. The given definition seems to have appeared for the first time in [8, p. 180, footnote 4]. It also appears in many other places, e.g., in [7, Sect. II.5.1, pp. 167-168].

Next, note that (GS) implies that every pair elements has a lower bound. In fact, we have the following equivalence.

Proposition 1**.**

Let ${\bf{J}}=(J;\leq)$ be a join-semilattice. Then, the following two statements are equivalent:

(i) Every pair of elements has a lower bound.

(ii) for all $a,b,x\in J$, if $x\leq a\vee b$, then there exist $a^{\prime},b^{\prime}\in J$ such that $a^{\prime}\leq a$, $b^{\prime}\leq b$, and $a^{\prime}\vee b^{\prime}\leq x$.

Proof.

(i) $\Rightarrow$ (ii) Suppose $x\leq a\vee b$. Let $a^{\prime}$ be a lower bound of $\{a,x\}$ and $b^{\prime}$ be a lower bound of $\{b,x\}$. Then, $a^{\prime}\leq a$ and $b^{\prime}\leq b$. Also, $a^{\prime}\leq x$ and $b^{\prime}\leq x$, which implies that $a^{\prime}\vee b^{\prime}\leq x$.

(ii) $\Rightarrow$ (i) Let $a,b\in J$. We have $a\leq a\vee b$. Then, by hypothesis, there exist $a^{\prime}\leq a$, $b^{\prime}\leq b$ such that $a^{\prime}\vee b^{\prime}\leq a$. As $b^{\prime}\leq a^{\prime}\vee b^{\prime}$, it follows that $b^{\prime}\leq a$. Then, $b^{\prime}\leq a,b$. That is, $b^{\prime}$ is a lower bound of $\{a,b\}$. ∎

This proposition shows that every GS-distributive join-semilattice is downward directed. This implies, as it is shown in [7], that the ideal $I\lor J$, generated by the union of two ideals $I,J$, is defined as in the case of distributive lattices, namely,

[TABLE]

As a consequence, it follows that the ideals of a (GS)-distributive join-semilattice ${\bf{J}}$ form a lattice that will be denoted by $Id({\bf{J}})$, and Grätzer proves in [7, p. 168] the following characterization result.

Proposition 2**.**

Let ${\bf{J}}$ be a join-semilattice. Then, ${\bf{J}}$ is (GS)-distributive iff $Id({\bf{J}})$ is distributive.

4.2 K-distributivity

The concept given in the following definition is similar to the one in (GS).

Definition 3**.**

A join-semilattice ${\bf{J}}=(J;\leq)$ is K-distributive iff

• (K)

for all $a,b,x\in J$, if $x\leq a\vee b$, $x\nleq a$ and $x\nleq b$, then there exist $a^{\prime},b^{\prime}\in J$ such that $a^{\prime}\leq a$, $b^{\prime}\leq b$, and $x=a^{\prime}\vee b^{\prime}$.

In order to visualize, see again Figure 2. The given definition seems to have appeared for the first time in [11, Definition 4, p. 122]. It also appears, for example, in [10, p. 167].

It turns out that, from the very definition, GS-distributivity implies K-distributivity. In fact, as noted in [11, 1.5, p. 122-123], it is the case that GS-distributivity is equivalent to K-distributivity plus the condition that every pair of elements has a lower bound (that is, downward directed). Therefore, the following proposition makes clear the relationship between GS- and and K-distributivity.

Proposition 3**.**

GS-distributivity implies K-distributivity, but not conversely.

The most simple counter-example showing that the reciprocal does not hold is the join-semilattice in Figure 3, that is not downward directed. Indeed, the given join-semilattice is K-distributive, as the only way to satisfy the antecedent of (K) is to take $1\leq a\vee b$, but then the consequent is also true. On the other hand, it is not (GS)-distributive, as we have $a\leq a\vee b$ and, however, there are no $a^{\prime}\leq a$, $b^{\prime}\leq b$ such that $a^{\prime}\vee b^{\prime}=a$.

Finally, analogously to Proposition 2, we have the following characterisation of K-distributivity via ideals, a proof of which may be found in [11, p. 123].

Proposition 4**.**

Let ${\bf{J}}$ be a join-semilattice. Then, ${\bf{J}}$ is (K)-distributive iff $Id({\bf{J}})\cup\{\emptyset\}$ is distributive.

4.3 H-distributivity

In [10] Hickman introduces the concept of mildly distributive meet-semilattices as those meet-semilattices whose lattice of their strong ideals is distributive. In [10, Theorem 2.5, p. 290] it is stated that it is equivalent to the following statement: 111Note that the original Hickman’s statement can be misleading since the condition “there exists $(x\wedge a_{1})\vee(x\wedge a_{2})\vee\cdots\vee(x\wedge a_{n})$” is missing.

• (H)

for all $n$ and $x,a_{1},\cdots,a_{n}$,

IF for all $b$ (if $a_{1}\leq b,\cdots,a_{n}\leq b$, then $x\leq b)$,

THEN there exists $(x\wedge a_{1})\vee\cdots\vee(x\wedge a_{n})$ and $x\leq(x\wedge a_{1})\vee\cdots\vee(x\wedge a_{n})$.

The given conditional may be seen as a translation of the following version of distributivity for lattices:

IF $x\leq a_{1}\vee\cdots\vee a_{n}$, THEN $x\leq(x\wedge a_{1})\vee\cdots\vee(x\wedge a_{n})$.

In the case of a join-semilattice ${\bf{J}}=(J;\leq)$ and using quantifiers, (H) may be rendered as follows:

• for all $n$ and $x,a_{1},\dots,a_{n}\in J$,

IF $x\leq a_{1}\vee\cdots\vee a_{n}$,

THEN for all $y$, if for all $i=1,\ldots,n$ (for all $z$, IF $z\leq x$ and $z\leq a_{i}$, THEN $z\leq y$)

then $x\leq y$

that is in turn equivalent to:

• for all $n$ and $x,a_{1},\dots,a_{n}\in J$,

IF $x\leq a_{1}\vee\cdots\vee a_{n}$,

THEN for all $y$, if (for all $z$, IF $z\leq x$ and ($z\leq a_{1}$ or …or $z\leq a_{n}$), THEN $z\leq y$)

then $x\leq y$.

Using set-theoretic notation, (H) may also be rendered as follows:

• (C)

for all $n$ and $x,a_{1},\dots,a_{n}\in J$,

if $x\leq a_{1}\vee\cdots\vee a_{n}$, then $x\in(\{x,a_{1}\}^{l}\cup\cdots\cup\{x,a_{n}\}^{l})^{ul}$.

At this point, the reader may wonder whether the number $n$ of arguments is relevant or whether two arguments are enough. Let us settle this question. Firstly, with that in mind, consider

• (D${}_{\vee_{n}}$)

for all $x,a_{1},\dots,a_{n},c$,

if $\{x,a_{1}\}^{l}\cup\cdots\cup\{x,a_{n}\}^{l}\subseteq\{c\}^{l}$, then $\{x,a_{1}\vee\cdots\vee a_{n}\}^{l}\subseteq\{c\}^{l}$.

Now, let us state the following fact.

Lemma 2**.**

(D${}_{\vee_{n}}$)* is equivalent to (C).*

Proof.

$\Rightarrow$) Suppose $x\leq a_{1}\vee\cdots\vee a_{n}$ and $y\in(\{x,a_{1}\}^{l}\cup\cdots\cup\{x,a_{n}\}^{l})^{u}$. Our goal is to see that $x\leq y$. Take $c=y$ and apply (D${}_{\vee_{n}}$). Then we have $\{x\}^{l}=\{x,a_{1}\vee\cdots\vee a_{n}\}^{l}\subseteq\{y\}^{l}$, and hence $x\leq y$.

$\Leftarrow$) Suppose $\{x,a_{1}\}^{l}\cup\{x,a_{2}\}^{l}\cup\cdots\cup\{x,a_{n}\}^{l}\subseteq\{c\}^{l}$. We have to prove that, if $y\leq x$ and $y\leq a_{1}\vee\cdots\vee a_{n}$ then $y\leq c$. Now, using (C), and the assumptions $y\leq x$ and $y\leq a_{1}\vee\cdots\vee a_{n}$ it follows that $y\in(\{x,a_{1}\}^{l}\cup\cdots\cup\{x,a_{n}\}^{l})^{ul}$. But since $\{x,a_{1}\}^{l}\cup\{x,a_{2}\}^{l}\cup\cdots\cup\{x,a_{n}\}^{l}\subseteq\{c\}^{l}$, we also have $y\in(\{x,a_{1}\}^{l}\cup\cdots\cup\{x,a_{n}\}^{l})^{ul}\subseteq\{c\}^{lul}=\{c\}^{l}$. Hence $y\leq c$. ∎

In turn, let us see that (D${}_{\vee_{n}}$) is equivalent to (D∨), which proves that having more than two arguments does not make any difference.

Lemma 3**.**

(D${}_{\vee_{n}}$)* is equivalent to (D∨).*

Proof.

We just prove that (D∨) implies (D${}_{\vee_{3}}$), the reciprocal being immediate. Let us suppose $\{h,a_{1}\}^{l}\cup\{h,a_{2}\}^{l}\cup\{x,a_{3}\}^{l}\subseteq\{c\}^{l}$. Then, we get both $\{h,a_{1}\}^{l}\subseteq\{c\}^{l}$ and $\{h,a_{2}\}^{l}\cup\{x,a_{3}\}^{l}\subseteq\{c\}^{l}$, the last of which, using (D∨), implies that $\{h,a_{2}\vee a_{3}\}^{l}\subseteq\{c\}^{l}$, which, together with the first, using (D∨) again, finally implies that $\{h,a_{1}\vee a_{2}\vee a_{3}\}^{l}\subseteq\{c\}^{l}$. ∎

As a consequence, H-distributivity coincides with the notion of DN-distributivity for join-semilattices introduced in Section 3. Accordingly, we have the following proposition.

Proposition 5**.**

A join-semilattice ${\bf{J}}=(J;\leq)$ is H-distributive iff it is ND-distributive.

Analogously to Propositions 2 and 4, we also have a characterization of H-distributivity for join-semilattices in terms of distributivity of a sublattice of their ideals. This appears as Corollary 2.4 in [10, p. 290]), where $Id_{fp}({\bf{J}})$ denotes the set $\{(a_{1}]\cap\dots\cap(a_{k}]:a_{1},...a_{k}\in J\}$, that is, the set of ideals that are intersection of a finite set of principal ideals of the join-semilattice ${\bf{J}}=(J;\leq)$.

Proposition 6**.**

Let ${\bf{J}}$ be a join-semilattice. Then, ${\bf{J}}$ is H-distributive iff $Id_{fp}({\bf{J}})$ is distributive.

Let us now compare H- with K-distributivity.

Proposition 7**.**

Let ${\bf{J}}=(J;\leq)$ be a join-semilattice. Then, K-distributivity implies H-distributivity.

Proof.

Suppose

• (x1)

for all $x\in J$, if $x\leq h$ and $x\leq a$, then $x\leq c$ and

• (x2)

for all $x\in J$, if $x\leq h$ and $x\leq b$, then $x\leq c$.

Further, suppose both (S1) $x\leq h$ and (S2) $x\leq a\vee b$. The goal is to prove $x\leq c$. Let us suppose that $x\leq a$. Then, using (x1) and (S1), it follows that $x\leq c$. The case $x\leq b$ is analogous using (x2). Finally, suppose both $x\nleq a$ and $x\nleq b$. Using (K) and (S2) it follows that there exist $a^{\prime},b^{\prime}\in J$ such that $a^{\prime}\leq a$, $b^{\prime}\leq b$, and (F) $x=a^{\prime}\vee b^{\prime}$, which implies $a^{\prime}\leq x$, which using (S1) gives $a^{\prime}\leq h$. As we also have $a^{\prime}\leq a$, using (x1) we get $a^{\prime}\leq c$. Reasoning analogously, we get $b^{\prime}\leq c$. So, using (F) it follows that $x\leq c$. ∎

The reciprocal of Proposition 7 does not hold considering the model in Figure 4 (with the understanding that there is no element in the white node). The given model appears as a poset in [6, Figure 2.7, p. 37].222 We thank the author of that paper for communicating this example. We provide a proof using the characterization of K- and H-distributivity by their ideals (Propositions 4 and 6).

Proposition 8**.**

H-distributivity does not imply K-distributivity.

Proof.

Let us characterize the sets $Id_{fp}(J)$ and $Id(J)$, where $(J,\leq)$ is the join-semilattice of Figure 4. An easy computation proves, on the one hand, that $Id_{fp}(J)$ is isomorphic to the ordered set of Figure 4 plus the ideal $I_{\overline{x}}=(f]\wedge(d]$, whose elements are $\{x_{i}:i\in w\}$, that does not exist in the original join-semilattice. On the other hand, $Id(J)$ is the set of ideals in $Id_{fp}(J)$ plus the ideal ${I}_{\overline{y}}$ generated by the set $\{y_{i}:i\in w\}$, that is, the ideal with elements ${I}_{\overline{y}}=\{y_{i}:i\in w\}\cup\{x_{i}:i\in w\}$. Clearly, this ideal is not a finite intersection of principal ideals. Both $Id_{fp}(J)$ and $Id(J)$ are lattices. Moreover, it is obvious that $Id_{fp}(J)$ is a distributive lattice and thus the join-semilattice of the example is H-distributive. But this is not the case for $Id(J)$, since it has a sublattice isomorphic to the pentagon formed by the elements $(a]$, $(d]$, $(c]$, ${I}_{\overline{y}}$, and ${I}_{\overline{x}}$. Thus, the join-semilattice of the example is not K-distributive. ∎

It is natural to ask whether it is possible to find a finite example in order to prove the reciprocal of Proposition 7. Let us see that the answer is negative.

Proposition 9**.**

For finite join-semilattices, H-distributivity and K-distributivity coincide.

Proof.

Consider a finite H-distributive join-semilattice. We want to see that it is K-distributive. Accordingly, suppose $x\leq a\vee b$, $x\nleq a$, and $x\nleq b$. It is natural to consider $\bigvee\{a,x\}^{l}$ and $\bigvee\{b,x\}^{l}$ as candidates for $a^{\prime}$ and $b^{\prime}$ in the definition of K-distributivity. Now, in order to do that, we first need to prove that the sets $\{a,x\}^{l}$ and $\{b,x\}^{l}$ are not empty. Suppose, say, $\{a,x\}^{l}=\emptyset$. Then, we have :

• • for all $y$, if $y\leq x$ and $y\leq a$, then $y\leq b$ (as $\{a,x\}^{l}=\emptyset$),

• for all $y$, if $y\leq x$ and $y\leq b$, then $y\leq b$,

• $x\leq x$, and

• $x\leq a\vee b$.

So, using H-distributivity, it follows that $x\leq b$, a contradiction.

Having proved that both $\{a,x\}^{l}\neq\emptyset$ and $\{b,x\}^{l}\neq\emptyset$, let us note that both $\bigvee\{a,x\}^{l}$ and $\bigvee\{b,x\}^{l}$ exist, due to having a finite structure. Next, let us see that $\bigvee\{a,x\}^{l}=$ inf $\{a,x\}$ (analogously, $\bigvee\{b,x\}^{l}=$ inf $\{b,x\}$). It is clear that both $\bigvee\{a,x\}^{l}\leq a$ and $\bigvee\{a,x\}^{l}\leq x$. Now, suppose $y\leq a,x$. Then, $y\in\{a,x\}^{l}$, and so, $y\leq\bigvee\{a,x\}^{l}$, as desired.

It remains to be seen that 1) inf $\{a,x\}\leq a$, 2) inf $\{b,x\}\leq b$, and 3) $x=$ inf$\{a,x\}\vee$ inf$\{b,x\}$. Now, 1) and 2) are easy to see. Regarding 3), as we have both that inf$\{a,x\}\leq x$ and sup $\{b,x\}\leq x$, it follows that inf$\{a,x\}\ \vee$  inf$\{b,x\}\leq x$. Finally, observe that the inequality $x\leq$ inf$\{a,x\}\ \vee$ inf$\{b,x\}$ follows from

• • for all $y$, if $y\leq x$ and $y\leq a$, then $y\leq$ inf$\{a,x\}\ \vee$ inf$\{b,x\}$,

• for all $y$, if $y\leq x$ and $y\leq b$, then $y\leq$ inf$\{a,x\}\ \vee$ inf$\{b,x\}$,

• $x\leq x$, and

• $x\leq a\vee b$,

using H-distributivity. ∎

In fact, it is easy to observe that in the case of a finite join-semilattice $J$, the sets of ideals $Id(J)$ and $Id_{fp}(J)$ coincide since, for any two elements $a,b$, either there is no lower bound, that is, $\{a,b\}^{l}=\emptyset$, or there exists their meet $a\land b=\bigvee\{a,b\}^{l}$.

4.4 LR-distributivity

Larmerová-Rachůnek version of distributivity (see [13]) was given for posets, as we next see.

Definition 4**.**

A poset ${\bf{P}}=(P;\leq)$ is LR-distributive iff

• (LRP)

for all $a,b,c\in P$, $(\{c,a\}^{l}\cup\{c,b\}^{l})^{ul}=(\{c\}\cup\{a,b\}^{u})^{l}$.

Remark 2**.**

In the given definition, it is enough to take one inclusion. Indeed, given a poset ${\bf{P}}=(P;\leq)$ and $a,b,c\in P$, it is always the case that $(\{c,a\}^{l}\cup\{c,b\}^{l})^{ul}\subseteq(\{c\}\cup\{a,b\}^{u})^{l}$.

It is natural to ask for LR-distributivity in the case of a join-semilattice. The following definition follows from the fact that in a join-semilattice ${\bf{J}}=(J,\leq)$ it holds that $(\{c\}\cup\{a,b\}^{u})^{l}=\{c,a\vee b\}^{l}$.

Definition 5**.**

A join-semilattice ${\bf{J}}=(J,\leq)$ is LR-distributive iff

• (LR)

for all $a,b,c\in J$, $\{c,a\vee b\}^{l}\subseteq(\{c,a\}^{l}\cup\{c,b\}^{l})^{ul}$.

Now, it can be seen that LR-distributivity is equivalent to H-distributivity, and hence to the condition (D∨)** as well.

Proposition 10**.**

Let ${\bf{J}}=(J;\leq)$ be a join-semilattice. Then the following conditions are equivalent:

• (i)

${\bf{J}}$* satisfies (LR),*

• (ii)

${\bf{J}}$* satisfies (H),*

• (iii)

${\bf{J}}$* satifies (D∨).*

Proof.

The equivalence between (ii) and (iii) is Prop. 5. Let us prove that (LR) implies (H). Suppose

• (x1) for all $x\in J$, if $x\leq h$ and $x\leq a$, then $x\leq c$,

(x2) for all $x\in J$, if $x\leq h$ and $x\leq b$, then $x\leq c$, $x\leq h$ and $x\leq a\vee b$.

Then, the last two inequalities imply $x\in\{c,a\vee b\}^{l}$. So, using (LR) we get that $x\in(\{c,a\}^{l}\cup\{c,b\}^{l})^{ul}$. That is, for all $y\in J$, if $y\in(\{c,a\}^{l}\cup\{c,b\}^{l})^{u}$, then $x\leq y$. Now, it should be clear that (x1) and (x2) imply that $c\in(\{c,a\}^{l}\cup\{c,b\}^{l})^{u}$. So, $x\leq c$, as desired.

Now, let us see that (H) implies (LR). Suppose $x\in\{h,a\vee b\}^{l}$, that is, (H1) $x\leq h$ and (H2) $x\leq a\vee b$. In order to get our goal, that is, $x\in(\{h,a\}^{l}\cup\{h,b\}^{l})^{ul}$, let us suppose that (S) $y\in(\{h,a\}^{l}\cup\{h,b\}^{l})^{u}$ and try to derive $x\leq y$. Now, (S) means that for all $z\in J$, if $z\in(\{h,a\}^{l}\cup\{h,b\}^{l}$, then $z\in y$, that is,

• (y1) for all $z\in J$, if $z\leq h$ and $z\leq a$, then $z\leq y$ and

(y2) for all $z\in J$, if $z\leq h$ and $z\leq b$, then $z\leq y$.

Now, using (H), (y1), (y2), (H1), and (H2), we get our goal, that is, $x\leq y$. ∎

4.5 B-distributivity

The following definition seems to have appeared for the first time in [1, Theorem 2.2. (i), p. 261].

Definition 6**.**

A join-semilattice ${\bf J}=(J;\leq)$ is B-distributive iff

• (B)

for all $n,a_{1},a_{2},\dots,a_{n},x\in J$, if $a_{1}\wedge a_{2}\wedge\cdots\wedge a_{n}$ exists, then also $(x\vee a_{1})\wedge(x\vee a_{2})\wedge\cdots\wedge(x\vee a_{n})$ exists and equals $x\vee(a_{1}\wedge a_{2}\wedge\cdots\wedge a_{n})$.

We have the following fact.

Proposition 11**.**

Let ${\bf{J}}=(J;\leq)$ be a join-semilattice. Then, H-distributivity implies B-distributivity.

Proof.

Let us have a H-distributive join semilattice $J$ and let us take $a,b,x\in J$ (the general case follows by induction). Let us suppose that $a\wedge b$ exists in $J$. Then, also $x\vee(a\wedge b)$ exists in $J$. Our goal is to see that $x\vee(a\wedge b)=$ inf $\{x\vee a,x\vee b\}$. It is clear that $x\vee(a\wedge b)\leq x\vee a,x\vee b$. Now, suppose both (F1) $y\leq x\vee b$ and (F2) $y\leq x\vee a$. We have to see that $y\leq x\vee(a\wedge b)$. It immediately follows that

• (x1) for all $w\in J$, if $w\leq x\vee b$ and $w\leq x$, then $w\leq x\vee(a\wedge b)$.

Now, suppose (F3) $w\leq x\vee b$ and (F4) $w\leq a$. Then, we have both

• (x1’) for all $y\in J$, if $y\leq a$ and $y\leq x$, then $y\leq x\vee(a\wedge b)$, and

(x2’) for all $y\in J$, if $y\leq a$ and $y\leq b$, then $y\leq x\vee(a\wedge b)$.

So, applying H-distributivity to (F3), (F4), (z1’), and (x2’), we have $w\leq x\vee(a\wedge b)$. That is, we have proved

• (x2) for all $w\in J$, if $w\leq x\vee b$ and $w\leq a$, then $w\leq x\vee(a\wedge b)$.

Using H-distributivity, (F1), (F2), (x1) and (x2), it finally follows that $y\leq x\vee(a\wedge b)$, as desired. ∎

The reciprocal of Proposition 11 does not hold as may be seen in Figure 5.

Observe also that the lattice $Id_{fp}(J)$, for $J$ being the join-semilattice of Figure 5, is not distributive since it is a diamond.

4.6 Sn-distributivity

The following definition seems to have appeared for the first time in [14].

Definition 7**.**

A join-semilattice $(J;\leq)$ is said to be Sn-distributive* for $n$ a natural number, $2\leq n$, iff*

• (Sn)

for all $a_{1},a_{2},\dots,a_{n},x\in J$, if $a_{1}\wedge a_{2}\wedge\cdots\wedge a_{n}$ exists, then also $(x\vee a_{1})\wedge(x\vee a_{2})\wedge\cdots\wedge(x\vee a_{n})$ exists and equals $x\vee(a_{1}\wedge a_{2}\wedge\cdots\wedge a_{n})$.

It is easy to see that B-distributivity implies $S_{n}$-distributivity, for any $n\geq 2$. It is also clear that for any $n\geq 2$, $S_{n+1}$ implies $S_{n}$. On the other hand, we have that for no natural $n\geq 2$ it holds that $S_{n}$-distributivity implies B-distributivity. In fact, it was proved that for any $n\geq 2$, $S_{n}$ does not imply $S_{n+1}$ (see [12]), where infinite models using the real numbers are provided. As in the case of GS- and H-distributivity, it is natural to ask whether, for example, finite models are possible. As in the cases just mentioned, the answer is negative as already proved in [16, Theorem 7.1, p. 1071]. In [15, Theorem, p. 26] it is also proved that it is not possible to find infinite wellfounded models.

Therefore, so far we have seen that, in the case of a join-semilattice, we have the following chain of implications:

(GS) $\Rightarrow$ (K) $\Rightarrow$ (H) $\Leftrightarrow$ (LR) $\Leftrightarrow$ (ND) $\Rightarrow$ (B) $\Rightarrow\cdots$ (Sn) $\Rightarrow$ (Sn-1) $\Rightarrow\cdots$ (S2).

5 Join-semilattices with arrow

The expansion of semilattices with an arrow operation has been well studied in the literature in the case of meet-semilattices under the name of relatively pseudo-complemented semilattices (see, for example, [7]). However, as far as we know, the expansion of join-semilattices with an arrow has not received much attention, see, for instance, [3, 4]. In this section we deal with distributivity of join-semilattices expanded with an arrow operation.

A join-semilattice with arrow is a structure $(J;\leq,\to)$ where $(J;\leq)$ is a join-semilattice and the arrow $\to$ is a binary operation such that for all $a,b\in J$:

$a\to b=\max\{c\in J:$ for all $x\in J$, if $x\leq a$ and $x\leq c$, then $x\leq b\}$.

The existence of the $\to$ operation is clearly equivalent to the requirement that $\to$ satisfies the following two conditions:

($\to$E) for all $x\in J$, if $x\leq a$ and $x\leq a\to b$, then $x\leq b$,

($\to$I) for all $c\in J$, IF for all $x\in J$, if $x\leq a$ and $x\leq c$, then $x\leq b$, THEN $c\leq a\to b$.

Remark 3**.**

The idea of defining arrow in a poset was already present in [9] (see Definition 4, where the author uses the terminology of Brouwer poset and also proves that a poset with arrow is LR-distributive). Moreover, the author, using LR-notation, defines $a\to b=$ max $\{c\in J:\{a,c\}^{l}\subseteq\{b\}^{l}\}$.

Remark 4**.**

In a lattice, or even in a meet-semilattice, arrow coincides with the usual relative meet-complement. This follows from the fact that, as previously mentioned, the inequality $a\wedge x\leq b$ is equivalent to the following universal quantification: for all $y$, if $y\leq a$ and $y\leq x$, then $y\leq b$. By the way, we prefer to use “arrow” instead of “relative meet-complement”, because the meet is not present.

As is well known, a lattice with a relative meet-complement (that is in fact a Heyting algebra) is distributive (see [17] or [18]). The natural question arises whether a join-semilattice with arrow is distributive in any of the senses considered in Section 4. The answer is negative in the case of (GS)-distributivity, as the join-semilattice in Figure 6 has arrow and is not GS-distributive.

A similar question in the case of K-distributivity has also a negative answer, as the the join-semilattice in Figure 7, already given in Figure 4, has arrow and is not K-distributive.

The case of H-distributivity is different, as we see next.

Proposition 12**.**

Every join-semilattice expanded with arrow is H-distributive.

Proof.

Let ${\bf J}=(J;\leq)$ be a join-semilattice with arrow. Take $a,b,c,h\in J$. Suppose

• (x1) for all $x\in J$, if $x\leq h$ and $h\leq a$, then $h\leq c$ and

(x2) for all $x\in J$, if $x\leq h$ and $x\leq b$, then $x\leq c$.

Take $y\in J$ and suppose

• (F1) $y\leq h$ and

(F2) $y\leq a\vee b$.

Now, using ($\to$I), (x1) implies $a\leq h\to c$ and (x2) implies $b\leq h\to c$. These inequalities together with (F2) imply $y\leq h\to c$, which, using (F1) and ($\to$E), gives $y\leq c$. ∎

Analogously to what happens when considering lattices, in the finite case we have the following fact.

Proposition 13**.**

Every finite H-distributive join-semilattice has arrow.

Proof.

Let ${\bf J}=(J;\leq)$ be a finite H-distributive join-semilattice. Due to finiteness, $c_{1}\vee c_{2}\vee\cdots\vee c_{n}=\bigvee\{c\in J:$ for all $x\in J$, if $x\leq a$ and $x\leq c$, then $x\leq b\}$ exists, for any $a,b\in J$. It is clear that for any $c_{i},1\leq i\leq n$, it holds that

• (F) for all $x$, if $x\leq a$ and $x\leq c_{i}$, then $x\leq b$.

Now, let us see that $c_{1}\vee c_{2}\vee\cdots\vee c_{n}$ is in fact $a\to b$.

First, let us see that $c_{1}\vee c_{2}\vee\cdots\vee c_{n}\in\{c\in J:$ for all $x\in J$, if $x\leq a$ and $x\leq c$, then $x\leq b\}$. That is, we have to see that

• (T) for all $x\in J$, if $x\leq a$ and $x\leq c_{1}\vee c_{2}\vee c_{n}$, then $x\leq b$.

Now, (T) clearly follows from (F) by H-distributivity.

Secondly, let us take $c\in J$ such that for all $x\in J$, if $x\leq a$ and $x\leq c$, then $x\leq b$. Then, obviously, $c\in\{c\in J:$ for all $x\in J$, if $x\leq a$ and $x\leq c$, then $x\leq b\}$. Then, $c\leq c_{1}\vee c_{2}\cdots\vee c_{n}$, as $c_{1}\vee c_{2}\vee\cdots\vee c_{n}=\bigvee\{c\in J:$ for all $x\in J$, if $x\leq a$ and $x\leq c$, then $x\leq b\}$. ∎

Finally, the natural question arises whether the class of join-semilattices expanded with arrow forms a variety or at least a quasi variety. The following example proves that the answer is negative. Indeed, consider the distributive lattice in Figure 8, which is the direct product ${\bf J}=(L\times L;\leq)$ where $L=\{0,\tfrac{1}{2},1\}$. It is clear that we can define in ${\bf J}$ an arrow $\to$, in fact, ${\bf J}^{*}=(L\times L;\leq,\to)$ becomes a Heyting algebra. Now, consider ${\bf J}^{*}$ as a join-semilattice with arrow, and observe that the set $B$ of elements represented by black nodes in the figure is the domain of a subalgebra $(B;\leq,\to)$ of ${\bf J}^{*}$, since both $\lor$ and $\to$ are closed on $B$. However, the join-semilattice $(B,\leq)$ is not distributive (it contains a pentagon), and moreover the arrow operation is not defined for all pairs of elements. In particular, $(\tfrac{1}{2},\tfrac{1}{2})\Rightarrow(0,0)$ is not defined since the set

[TABLE]

has no maximum.

6 Conclusions

In this paper we have proposed a notion of distributivity for join-semilattices with logical motivations related to Gentzen’s disjunction elimination rule in the $\{\lor,\to\}$-fragment of intuitionistic logic, and we have compared it to other notions of distributivity for join-semilattices proposed in the literature.

There are a number of open problems that we plan to address as future research. In particular we can mention the following ones:

• •

As for the logical motivation, similar to the $\bf(\lor E)$ rule in Section 3, one can consider the following rule with two contexts:

$\mathfrak{H}_{1}$, $\mathfrak{H}_{2}$, $\mathfrak{A\vee B}$       

$\mathfrak{H}_{1}$, $\mathfrak{H}_{2}$, [$\mathfrak{A}$]

$\mathfrak{C}$

     

$\mathfrak{H}_{1}$, $\mathfrak{H}_{2}$, [$\mathfrak{B}$]

$\mathfrak{C}$

$\mathfrak{C}$

This rule also has a natural algebraic translation in the case of join-semilattices. The question arises whether it is equivalent to the condition (D∨)** or if it leads to a different one.

• •

Distributive lattices are charecterized by their lattice of ideals. In the case of join-semillatices, there are similar characterizations for GS-, K- and H-distributivity, but not for B- and Sndistributivity. The question is whether B- and Sn-distributive join-semilattices can be characterized by means of their ideals.

• •

In [2] the authors generalize the well-known characterisation of distributive lattices in terms of forbidden sublattices (diamond and pentagon) to distributive posets, also identifying the set of forbidden subposets. A similar study for distributive join-semilattices is an open question.

Acknowledgments

The authors acknowledge partial support by the H2020 MSCA-RISE-2015 project SYSMICS. Esteva and Godo also acknowledge the FEDER/MINECO project TIN2015-71799-C2-1-P.