arXiv:2006.00642v3 [math.LO] 26 Jun 2021 lentvl,w a Σ say we Alternatively, with variable nΣwithout Σ in implication fcascllgcfi o h eeatlgc,icuigthe including 1.1. logics, relevant Definition a the for logic [1 fail such Kerr logic any (see classical call logic of we classical [7], to collapse Restall not and does and implication material than stronger fawd wt frlvn ois pnigfo h ai logic the basic to the lead from work spanning This logics, relevant logic. of classical swath in wide betwe formalization a mismatch its of a and from implication result the of that avoid inferences to counterintuitive introduced are first were logics Relevant Introduction 1 eea a efudi n f[,2 ,1,5.I hspprw ilfoc will we paper this In 5]. 16, 7, 2, [1, of any logic in found be can general entailment q KR iecaso eainagba.Ti ls nldsteBoo the includes class This algebras. relation of class wide atcscntutdb reet hwta pmrhssne epimorphisms that Ur su show of the to suggestion Freese of a by counterpart Following constructed algebraic lattices monoids. the Boolean to so-called extend the not does logics ehDfiaiiyPoet al for fails Property Definability Beth iu,Uqhr n19 hwdta h ehDfiaiiyPro Definability Beth including the logics, that relevant su showed of is class 1999 algebras lo in of algebraizable Urquhart category an nique, corresponding for the holds in Property epimorphism Definability Beth The esyΣ say We . p hc ossso digteprdxclaim( axiom paradoxical the adding of consists which , o e variable new a For . E R nbten opeesv ecito fteelgc n releva and logics these of description comprehensive A between. in ) p ihmn te eeatlgc sc stce entailment ticket as (such logics relevant other many with , R uhthat such , n hsntaprl eeatlogic, relevant purely a not thus and Let ehDfiaiiyi h Logic the in Definability Beth mlctydefines implicitly L xlctydefines explicitly eapooiinllgcadΣasto omlsfrom formulas of set a Σ and logic propositional a be q e Σ[ let , Σ p T ue2,2021 29, June Garber Jacob if ∪ , KR p/q E p Σ Σ[ Abstract ue relevant super and , fteei formula a is there if ⊢ p/q eoetersl frpaigalisacsof instances all replacing of result the denote ] . p 1 ] ↔ R ⊢ oee,tecutrxml o those for counterexample the However, . p A. aaoe fmtra implication material of paradoxes ↔ KR q. oi.Svrlipratproperties important Several logic. ehDfiaiiyProperty Definability Beth A visvrosohrprdxsof paradoxes other various avoids ∧¬ enmnis n hsthe thus and monoids, lean A jcie sn hstech- this Using rjective. dntb ujciei a in surjective be not ed uat euemodular use we quhart, otiigol h variables the only containing A i fadol fevery if only and if gic e eeatlogic relevant per ) B et al o wide a for fails perty nteitiiemeaning intuitive the en → KR otelgco relevant of logic the to soratnino the on attention our us B ].FloigDunn Following 4]). to R T ept being Despite . L development n relevant and otiiga containing tlgcin logic nt KR . which , , p A logic L is said to have the Beth Definability Property if for any set of formulas Σ and variable p, if Σ implicitly defines p, then Σ also explicitly defines p. The well-known Beth Definability Theorem states that the Beth Definability Property holds for classical propositional logic. However, as shown by Urquhart [17] in 1999 and extended by Blok and Hoogland [4, Corollary 4.15] in 2006, the Beth Definability Property fails for all relevant logics between B and R (inclusive). The techniques of those papers rely on the fact that Boolean negation is implicitly (but not explicitly) definable in those logics. This approach does not extend to KR, where Boolean negation is identified with relevant negation and is thus explicitly definable. However, Urquhart conjectured that the Beth Definability Property fails for KR as well, and outlined a possible method of attack using algebraic logic. In algebraic logic, every algebraizable logic L has a corresponding category of algebras Alg L. For instance, the algebras of classical logic are the Boolean algebras, and those of intuitionistic logic are the Heyting algebras. Using this correspondence, it is often possible to translate properties of a logic into properties of its corresponding algebra. Definition 1.2. For objects A, B in a category C, we say a map f : A → B is an epimorphism if for any other object C and maps g, h : B → C, if g ◦ f = h ◦ f, then g = h. The Beth Definability Property holds in an algebraizable logic L iff in the corresponding algebra Alg L all epimorphisms are surjective (the ES property). This correspondence was first proven by N´emeti in [13, Section 5.6], and further developed by Blok and Hoogland in [4]. Intuitively, one can think of epimorphisms as being implicit definitions, and surjections as being explicit ones. For algebras A ⊆ B, we say A is an epic subalgebra of B if the inclusion map i : A → B is an epimorphism. Of course, this inclusion map is surjective iff A = B. To disprove the Beth Definability Property for a logic, it thus suffices to find a proper epic subalgebra in its corresponding category. To apply this to KR, we will analyze epimorphisms in the category of algebras for KR, the Boolean monoids. Boolean monoids are closely related to relation algebras, and can be equivalently defined as dense symmetric relation algebras. We will tackle this problem using the approach described in Urquhart [18, Problem 5.3]. As discussed in that paper, there is a general correspondence between Boolean monoids and modular lattices. Every Boolean monoid contains a modular lattice, and given a modular lattice L, one can construct a corresponding Boolean monoid A(L) that contains an isomor- phic copy of L as a sublattice. As shown by Freese in [8, Theorem 3.3], there exist modular lattices A and B such that A is a proper epic sublattice of B. Using the above correspon- dence, we extend this to the construction of a proper epic sub-Boolean monoid, which shows the Beth Definability Property fails for KR. This construction is rather general, and in fact shows that ES fails for a wide class of relation algebras that includes the Boolean monoids.

2 Boolean Monoids and Modular Lattices

As shown by Anderson and Belnap in [1, Section 28.2.3], the algebraic counterpart of the t logic R with truth constant t (sometimes denoted R ) is the variety of De Morgan monoids.

2 These are De Morgan lattices with a commutative monoid operation. The addition of the axiom (A∧¬A) → B to R corresponds to adding the axiom a∧¬a = 0 to the algebra, which reduces the De Morgan lattice to a Boolean algebra. Such objects, which we call Boolean monoids, are the algebraic counterpart of KR. A particularly useful description of Boolean monoids is in terms of relation algebras. The following axiomatization of relation algebras is taken from Givant [11, Definition 2.1]. ` Definition 2.1. A relation algebra is an algebra A, ∨, ¬, ◦, , t such that for all a, b, c ∈ A, 1. a ∨ b = b ∨ a 2. a ∨ (b ∨ c)=(a ∨ b) ∨ c 3. ¬(¬a ∨ b) ∨ ¬(¬a ∨ ¬b)= a 4. a ◦ (b ◦ c)=(a ◦ b) ◦ c 5. a ◦ t = a 6. a`` = a 7. (a ◦ b)` = b` ◦ a` 8. (a ∨ b) ◦ c =(a ◦ c) ∨ (b ◦ c) ` 9. (a ∨ b) = a` ∨ b`

` 10. a ◦ ¬(a ◦ b)
∨ ¬b = ¬b With the standard definition of a ∧ b = ¬(¬a ∨ ¬b), axioms 1–3 imply that hA, ∨, ∧, ¬i ` is a Boolean algebra, axioms 4–7 imply that A, ◦, , t is a monoid with involution, and axioms 8–10 relate the Boolean and monoid operations to each other. For all a, b ∈ A, a relation algebra A is called 1. abelian if a ◦ b = b ◦ a, 2. symmetric if a` = a, 3. dense if a ≤ a ◦ a. In particular, a Boolean monoid can be equivalently defined as a dense symmetric relation algebra. One important result from the theory of relation algebras is that every abelian relation algebra contains a special set of elements that form a modular lattice. Definition 2.2. A lattice L is called modular if for all x, y, z ∈ L, x ≤ z =⇒ x ∨ (y ∧ z)=(x ∨ y) ∧ z. This implication is equivalent to the following dual identities: (x ∧ y) ∨ (x ∧ z)= x ∧ (y ∨ (x ∧ z)), (x ∨ y) ∧ (x ∨ z)= x ∨ (y ∧ (x ∨ z)).

3 Definition 2.3. For a relation algebra A, an element a ∈ A is called reflexive if t ≤ a, symmetric if a` = a, and transitive if a◦a ≤ a. An element with all three of these properties is a reflexive equivalence element. Define L(A) to be the set of all reflexive equivalence elements of A.

Theorem 2.4. For an abelian relation algebra A, the set of reflexive equivalence elements L(A) is closed under fusion and meet, and forms a bounded modular lattice. Join is given by a ◦ b, meet by a ∧ b, t is the bottom element, and 1 is the top.

Proof. See Givant [11, Corollary 5.17]. We can also in some sense reverse the above theorem, and use a modular lattice to construct a relation algebra.

Definition 2.5. A KR frame or model structure is a triple F = hS, R, 0i of a set S with ternary relation R and distinguished element 0, satisfying:

1. R0ab iff a = b

2. Raaa

3. Rabc implies Rbac and Racb (total symmetry)

4. Rabc and Rcde implies ∃f ∈ S such that Radf and Rfbe (Pasch’s Postulate)

The last condition has close ties to projective geometry and is explored by Urquhart in [18].

Definition 2.6. For a KR frame F = hS, R, 0i, the complex algebra of F is the algebra A(F )= hP (S), ∪, ∩, c, ◦, ti, where

1. hP (S), ∪, ∩, ci is the Boolean algebra on the power set of S.

2. t = {0} is the monoid identity.

3. For A, B ⊆ S, fusion is defined as

A ◦ B = { c ∈ S | Rabc for some a ∈ A, b ∈ B }.

We often write A(S) for the complex algebra when the ternary relation and distinguished element are understood from context.

Theorem 2.7. For a KR frame F , the complex algebra A(F ) is a Boolean monoid.

Proof. See Urquhart [18, Section 2].

Definition 2.8. For a lattice L with least element 0, we define the following ternary relation on the elements of L: Rabc ⇐⇒ a ∨ b = a ∨ c = b ∨ c. Then with this relation, hL, R, 0i is a KR frame iff L is modular.

4 Proof. The first three properties of a KR frame follow immediately from the lattice structure of L. The last property, Pasch’s Postulate, is equivalent to the modular law on L, which is shown in Urquhart [18, Theorem 2.7]. Thus for a modular lattice L with 0, the lattice complex algebra A(L) is a Boolean monoid. An alternative more direct proof of this can be found in Givant [11, Section 3.7]. The definition of a KR frame is also an instance of the more general notion of a relational structure, where the construction of a complex algebra can be repeated in the context of Boolean algebras with operators. Givant [10, Chapter 19] and [9, Chapter 1] go into more detail. For a lattice complex algebra A(L), a particularly simple description of its reflexive equivalence elements can be given in terms of the ideals of L. Definition 2.9. For a lattice L, an ideal of L is a non-empty subset J ⊆ L such that 1. If a, b ∈ J, then a ∨ b ∈ J.

2. If a ∈ J and b ≤ a, then b ∈ J. The set of all ideals of L is denoted Id L, which forms a lattice with respect to set inclusion. A special class of ideals are the principal ideals, which are of the form (a]= { b ∈ L | b ≤ a }. We will sometimes use the notation (a]L to emphasize that this is the principal ideal of a inside L. Proposition 2.10. For a lattice L, the principal ideal map

I : L → Id L a 7→ (a]

is an injective homomorphism of lattices. Proof. See Gr¨atzer [12, Corollary 4, p. 24]. Theorem 2.11. For a modular lattice L with least element 0, the set of reflexive equivalence elements L(A(L)) and the set of ideals Id L are identical as lattices. That is, L(A(L)) = Id L, and for all ideals J, K ∈ Id L

J ∨ K = J ◦ K J ∧ K = J ∩ K.

Proof. See Maddux [15, p. 243].

3 Embeddings of Lattice Complex Algebras

The purpose of this section is to prove Theorem 3.2, which states that for all complete sublattices K of a modular lattice L, there is a corresponding complete embedding φ : A(K) →A(L) of Boolean monoids. The construction of this map will rely on the following result from the theory of relation algebras.

5 Theorem 3.1. Let A be a complete and atomic Boolean monoid, U the set of atoms of A, and B a complete Boolean monoid. Suppose φ : U → B is a map with the following properties:

1. The elements φ(u) for u ∈ U are non-zero, mutually disjoint, and have a join of 1 in B.

2. t = W{ φ(u) | u ∈ U and u ≤ t }

3. φ(u) ◦ φ(v)= W{ φ(w) | w ∈ U and w ≤ u ◦ v } for all u, v ∈ U. Then φ extends in a unique way to a complete embedding φ : A → B of Boolean monoids, given by φ(r)= _{ φ(u) | u ∈ U and u ≤ r }, where r is any element of A.

Proof. This is a specialization of Givant [11, Theorem 7.13 and Corollary 7.14] to Boolean monoids. We apply this to lattice complex algebras as follows. Recall that for a complete lattice L, a subset K ⊆ L is a complete sublattice iff for all S ⊆ K, V S ∈ K and W S ∈ K, where these infima and suprema are calculated in L.

Theorem 3.2. Let L be a complete modular lattice, K ⊆ L a complete sublattice, and IK and IL their respective principal ideal maps. Then there is a unique complete embedding of Boolean monoids φ : A(K) →A(L) such that φ ◦ IK = IL.

Proof. We will first show uniqueness to determine what the map φ should be, and then use that definition to show it is a complete embedding. Suppose that φ : A(K) →A(L) is a complete embedding with φ ◦ IK = IL. Since φ is a complete homomorphism, it is determined by its values on the singleton subsets of K, which are the atoms of A(K). For all a ∈ K, we can write (a]K as the disjoint union

(a]K = {a} ∪ [ (b]K b < a b ∈ K

=⇒ φ((a]K)= φ({a}) ∪ [ φ((b]K ) φ is a complete homomorphism b < a b ∈ K

=⇒ (a]L = φ({a}) ∪ [ (b]L φ ◦ IK = IL b < a b ∈ K

=⇒ φ({a})=(a]L  [ (b]L φ preserves disjoint unions b < a b ∈ K

Thus φ is uniquely determined.

6 So then, let U be the set of singletons in A(K), and define the map φ : U →A(L) by

φ({a})=(a]L  [ (b]L b < a b ∈ K We will verify the three conditions of Theorem 3.1 to show that this can be extended to a complete embedding of Boolean monoids. 1. It suffices to show that the sets φ({a}) for a ∈ K are non-empty, mutually disjoint, and cover L. • All sets are non-empty, since a ∈ φ({a}) for any a ∈ K. • Let a, b ∈ K be distinct elements. Then a∧b ≤ a, and a∧b ≤ b. Since a and b are distinct, at least one of the previous inequalities must be strict, so without loss of generality suppose a ∧ b < a. Since K is a sublattice, a ∧ b ∈ K. Now suppose x ∈ φ({a}) ∩ φ({b}). Then x ≤ a and x ≤ b, so x ∈ (a ∧ b]L. Since a ∧ b < a, this implies x∈ / φ({a}), which is a contradiction. Thus φ({a}) and φ({b}) are disjoint. • For an arbitrary x ∈ L, let a = ^ Fx

where Fx = { b ∈ K | x ≤ b }. Since K is a complete sublattice, this infimum exists and is an element of K. By definition, x is a lower bound for Fx, so x ≤ a, and thus x ∈ (a]L. Since a ∈ Fx, we in fact have a = min Fx. Furthermore, for any other b ∈ K with b < a, it cannot be that x ∈ (b]L, since then we would have a ≤ b, which is impossible. Thus

x ∈ (a]L  [ (b]L = φ({a}). b < a b ∈ K So the images of φ cover L. 2. The monoid identity t = {0} is itself a singleton, and from the definition of φ we trivially have φ(t)= φ({0}) = (0]L = {0} = t. 3. From left to right, let a, b ∈ K, and suppose that z ∈ φ({a}) ◦ φ({b}). We wish to show z ∈ φ({c}), for some c ∈ K with {c}⊆{a}◦{b}. By assumption Rxyz for some x ∈ φ({a}) and y ∈ φ({b}). From the first condition, we know a = min Fx, b = min Fy, and z ∈ φ({c}), where c = min Fz. Since x ≤ a and y ≤ b, we have x ∨ y ≤ a ∨ b =⇒ x ∨ z ≤ a ∨ b since Rxyz =⇒ z ≤ a ∨ b =⇒ c ≤ a ∨ b minimality of c =⇒ a ∨ c ≤ a ∨ b.

7 Symmetrically, we conclude a ∨ b ≤ a ∨ c, and so a ∨ b = a ∨ c. A similar argument shows a ∨ c = b ∨ c. Thus Rabc, and so c ∈{a}◦{b} as desired. From right to left, let a, b, c ∈ K and suppose {c}⊆{a}◦{b}. We wish to show φ({c}) ⊆ φ({a}) ◦ φ({b}). That is, for all z ∈ φ({c}), there exists x ∈ φ({a}) and y ∈ φ({b}) such that Rxyz. To do this, we use an approach similar to the one of Maddux in [15, p. 244]. For a given z, let

x =(b ∨ z) ∧ a y =(a ∨ z) ∧ b.

We first show that a = min Fx. From the definition of x we have a ∧ b ≤ x ≤ a, so a ∈ Fx. Now let d ∈ Fx be any other element. Then d ∈ K with x ≤ d, so x ≤ a ∧ d. Furthermore,

x ∨ b = ((b ∨ z) ∧ a) ∨ b definition of x =(b ∨ z) ∧ (a ∨ b) modularity =(b ∨ z) ∧ (b ∨ c) since Rabc = b ∨ z since z ≤ c.

Since z ≤ b ∨ z = x ∨ b, we then have

z ≤ (a ∧ d) ∨ b since x ≤ a ∧ d =⇒ c ≤ (a ∧ d) ∨ b minimality of c =⇒ b ∨ c ≤ (a ∧ d) ∨ b =⇒ a ∨ b ≤ (a ∧ d) ∨ b since Rabc.

Using absorption, this implies

a ≤ ((a ∧ d) ∨ b) ∧ a =(a ∧ d) ∨ (a ∧ b) modularity = a ∧ d since a ∧ b ≤ x ≤ a ∧ d.

Therefore a ≤ d, so a = min Fx as wanted. Thus x ∈ φ({a}), and a symmetric argument shows that y ∈ φ({b}). Now we show Rxyz. Using modularity,

x ∨ z = ((b ∨ z) ∧ a) ∨ z =(b ∨ z) ∧ (a ∨ z) =(a ∨ z) ∧ (b ∨ z) = ((a ∨ z) ∧ b) ∨ z = y ∨ z.

8 Since x ≤ a and z ≤ c, we have x ∨ z ≤ a ∨ c = a ∨ b. Thus

x ∨ z =(a ∨ b) ∧ (x ∨ z) =(a ∨ b) ∧ (a ∨ z) ∧ (b ∨ z) fromabove =(a ∨ (b ∧ (a ∨ z))) ∧ (b ∨ z) modularity =(b ∨ z) ∧ (a ∨ ((a ∨ z) ∧ b)).

Using that (a ∨ z) ∧ b ≤ b ≤ b ∨ z and a final application of the modular law, we thus have

x ∨ z = ((b ∨ z) ∧ a) ∨ ((a ∨ z) ∧ b) = x ∨ y.

Thus Rxyz, and the condition is shown. Thus by Theorem 3.1, φ extends uniquely to a complete embedding φ : A(K) → A(L) of Boolean monoids, where for all S ⊆ K,

φ(S)= [ φ({a}). a ∈ S

We use this definition to show that φ◦IK = IL. For any a ∈ K,(a]K is a reflexive equivalence element of A(K) by Theorem 2.11. Since φ preserves equational properties, the image φ((a]K) is also a reflexive equivalence element of A(L), and thus an ideal of L by the same theorem. From the definition of φ, a ∈ φ({a}) ⊆ φ((a]K ),

and so (a]L ⊆ φ((a]K ) from the definition of an ideal. On the other hand,

φ((a]K )= [ φ({b}) ⊆ [ (b]L =(a]L, b ≤ a b ≤ a b ∈ K b ∈ K

and so φ((a]K )=(a]L.

4 An Epimorphism That is Not Surjective

In this section, let LRA be the class of all subalgebras of lattice complex algebras, ARA the variety of abelian relation algebras, and R any class of relation algebras with LRA ⊆ R ⊆ ARA. We will now use the following general construction and modular lattices constructed by Freese to show that ES fails for R. Let L be a complete modular lattice, and K ⊆ L a complete sublattice. The principal ideal map IL : L → Id L is an embedding of lattices, and from Theorem 2.11 we know ′ ′ Id L = L(A(L)). Thus, let K = IL(K) and L = IL(L) be the isomorphic images of L and K contained in L(A(L)). In A(L), let U be the subalgebra generated by K′, and V the subalgebra generated by L′.

9 Theorem 4.1. In the above situation, if K is a proper epic sublattice of L, then U is a proper R-epic subalgebra of V . Proof. Since K′ ⊂ L′ we have U ⊆ V . Let W be any other algebra of R, and f,g : V → W two homomorphisms that agree on U. The image of a reflexive equivalence element is a reflexive equivalence element, so f and g restrict to maps ′ f|L′ , g|L′ : L →L(W ). By Theorem 2.4, L(W ) is a modular lattice under fusion and meet, and f and g preserve these operations, so these restrictions are homomorphisms of modular lattices. By assumption, f and g agree on U, and since K′ ⊆ U they must also agree on K′. But K′ is an epic ′ ′ ′ sublattice of L , so f and g must also agree on L . Thus f|L′ = g|L′ , and so f = g since L is the generating set of V . Thus U is an R-epic subalgebra of V . However, U is a proper subalgebra. Let φ : A(K) → A(L) be the complete embedding of Theorem 3.2, and let Z = im φ. Since IK(K) ⊆A(K), we have ′ φ(IK(K)) = IL(K)= K , so K′ ⊆ Z, which implies U ⊆ Z since U is the smallest subalgebra that contains K′. Now for contradiction suppose that U = V . Then L′ ⊆ V = U ⊆ Z. Thus for any x ∈ L, we have (x]L ∈ Z, so there is some S ⊆ K such that φ(S)=(x]L. In particular then, there is an a ∈ S such that x ∈ φ({a}) ⊆ (a]L =⇒ x ≤ a. On the other hand, a ∈ φ({a}) ⊆ φ(S)=(x]L =⇒ a ≤ x. Thus x = a, so x ∈ K. But the element x ∈ L was arbitrary, so L = K, which is a contradiction. Theorem 4.2. ES fails for any class R of relation algebras with LRA ⊆ R ⊆ ARA. Proof. In [8, Theorem 3.3], Freese constructs modular lattices A ⊂ B such that A is a proper epic sublattice of B. B has no infinite chains, so is complete by Davey and Priestley [6, Theorem 2.41 (iii)]. Likewise, A is a {0,1}-sublattice of B, and as a sublattice is complete by Theorems 2.40 and 2.41 (i) of the same. We can thus apply Theorem 4.1 to A and B, and the result follows. Corollary 4.3. ES fails for the varieties of abelian, symmetric, and dense symmetric relation algebras. Corollary 4.4. The Beth Definability Property fails for KR.

5 Conclusion

Using modular lattices constructed by Freese, we have shown that epimorphisms need not be surjective in a wide class of relation algebras. This class includes the Boolean monoids, which shows that the Beth Definability Property fails for the super relevant logic KR. This should be contrasted with the result of [3, Theorem 8.5], which shows that the Beth Definability Property does hold for the super relevant logic RM. The super relevant logics thus exhibit more diversity than the relevant logics, where this property fails uniformly.

10 Acknowledgements The author would like to thank Katalin Bimb´ofor suggesting this problem for his under- graduate research project and supervising his work on it. Her advice and never-ending encouragement were instrumental in finding a solution.

References

[1] Alan Ross Anderson and Nuel D. Belnap. Entailment: The Logic of Relevance and Necessity, volume 1. Princeton University Press, 1975.

[2] Alan Ross Anderson, Nuel D. Belnap, and J. Michael Dunn. Entailment: The Logic of Relevance and Necessity, volume 2. Princeton University Press, 1992.

[3] Guram Bezhanishvili, Tommaso Moraschini, and James G. Raftery. Epimorphisms in varieties of residuated structures. Journal of Algebra, 492:185–211, 2017.

[4] W. J. Blok and Eva Hoogland. The Beth property in algebraic logic. Studia Logica, 83:49–90, 2006.

[5] Ross Brady, editor. Relevant Logics and Their Rivals, volume 2. Ashgate, 2003.

[6] B. A. Davey and H. A. Priestley. Introduction to Lattices and Order. Cambridge University Press, 2 edition, 2002.

[7] J. Michael Dunn and Gregory Restall. Relevance logic. In Handbook of Philosophical Logic, volume 6. Springer Netherlands, 2 edition, 2002.

[8] Ralph Freese. The variety of modular lattices is not generated by its finite members. Transactions of the American Mathematical Society, 255:277–300, 1979.

[9] Steven Givant. Duality Theories for Boolean Algebras with Operators. Springer Mono- graphs in Mathematics. Springer International Publishing, 2014.

[10] Steven Givant. Advanced Topics in Relation Algebras. Springer International Publishing, 2017.

[11] Steven Givant. Introduction to Relation Algebras. Springer International Publishing, 2017.

[12] George Gr¨atzer. General Lattice Theory. Birk¨auser Basel, 2 edition, 2003.

[13] Leon Henkin, J. Donald Monk, and Alfred Tarski. Cylindric Algebras, volume 2. North- Holland, 1985.

[14] Alison Duncan Kerr. A plea for KR. Synthese, 2019.

[15] Roger Maddux. Embedding modular lattices into relation algebras. Algebra Universalis, 12:242–246, 1981.

11 [16] Richard Routley, Robert K. Meyer, Val Plumwood, and Ross T. Brady. Relevant Logics and Their Rivals, volume 1. Ridgeview Publishing Company, 1982.

[17] Alasdair Urquhart. Beth’s definability theorem in relevant logics. In Logic at Work, volume 24 of Studies in Fuzziness and Soft Computing. Physica-Verlag Heidelberg, 1999.

[18] Alasdair Urquhart. The geometry of relevant implication. IFCoLog Journal of Logics and Their Applications, 4(3):591–604, 2017.

12