0% found this document useful (0 votes)
17 views93 pages

The Algebra of Logic

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
17 views93 pages

The Algebra of Logic

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 93

Lecture Notes on The Algebra of Logic

Tommaso Moraschini
June 28, 2021
Contents

Contents i

1 Preliminaries 1
1.1 Algebras and equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Basic constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Completeness theorems 13
2.1 Propositional logics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Algebraic semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Universal algebra 23
3.1 Ultraproducts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Universal classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Quasi-varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4 Algebraizable logics 39
4.1 Algebraizability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Deductive filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Isomorphism theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5 Deduction theorems 61
5.1 Bridge theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.2 Quasi-varieties with EDPRC . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3 Sketches of structure theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

Bibliography 81

i
CHAPTER 1
Preliminaries

1.1 Algebras and equations


We begin by reviewing some fundamentals of general algebraic systems. For more
information, the reader may consult [12, 30].

Definition 1.1.

(i) A type is a map ρ : F → N, where F is a set of function symbols. In this case, ρ( f ) is


said to be the arity of the function symbol f , for every f ∈ F . Function symbols of
arity zero are called constants.

(ii) An algebra of type ρ is a pair A = h A; F i where A is a nonempty set and F = { f A :


f ∈ F } is a set of operations on A whose arity is determined by ρ, in the sense that
each f A has arity ρ( f ). The set A is called the universe of A.

When F = { f 1 , . . . , f n }, we shall write h A; f 1A , . . . , f nA i instead of h A; F i. In this case, we


often drop the superscripts, and write simply h A; f 1 , . . . , f n i.

Classical examples of algebras are groups and rings. For instance, the type of groups
ρG consists of a binary symbol +, a unary symbol − and a constant symbol 0. Then a
group is an algebra h G; +, −, 0i of type ρG in which + is associative, 0 is a neutral element
for + and − produces inverses.
Lattices, Heyting algebras and modal algebras are also algebras in the above sense.
For instance, the type of lattices ρ L consists of two binary symbols ∧ and ∨ and a lattice
is an algebra h A; ∧, ∨i of type ρ L that satisfies the idempotent, commutative, associative
and absorption laws. Similarly, the type of Heyting algebras ρ H consists of three binary
operations symbols ∧, ∨ and → and of two constant symbols 0 and 1. Then a Heyting
algebra is an algebra h A; ∧, ∨, →, 0, 1i such that h A; ∧, ∨, 0, 1i is a bounded lattice and, for
every a, b, c ∈ A,
a ∧ b 6 c ⇐⇒ a 6 b → c. (residuation law)

1
1. P RELIMINARIES

Boolean algebras can be viewed as the Heyting algebras that satisfy the following equa-
tional version of the excluded middle law:

x ∨ ( x → 0) ≈ 1.

In this case, the complement operation ¬ x can be defined as x → 0.


Perhaps less obviously, even algebraic structures whose operations are apparently
external can be viewed as algebras in the sense of the above definition. For instance,
modules over a ring R can be viewed as algebras whose type ρ R extends that of groups
with the unary symbols {λr : r ∈ R}. From this point of view, a module over R is an
algebra h G; +, −, 0, {λr : r ∈ R}i of type ρ R such that h G; +, −, 0i is an abelian group and,
for every r, s ∈ R and a, c ∈ G,

λr ( a + c ) = λr ( a ) + λr ( c )
λr +s ( a ) = λr ( a ) + λ s ( a )
λr (λs ( a)) = λr·s ( a)
λ1 ( a) = a.

Given a type ρ : F → N and a set of variables X disjoint from F , the set of terms of
type ρ over X is the least set Tρ ( X ) such that

(i) X ⊆ Tρ ( X );

(ii) if c ∈ F is a constant, then c ∈ Tρ ( X ); and

(iii) if ϕ1 , . . . , ϕρ( f ) ∈ Tρ ( X ) and f ∈ F , then f ϕ1 . . . ϕρ( f ) ∈ Tρ ( X ).

For the sake of readability, we shall often write f ( ϕ1 , . . . , ϕρ( f ) ) instead of f ϕ1 . . . ϕρ( f ) .
Similarly, if f is a binary operation +, we often write ϕ1 + ϕ2 instead of f ( ϕ1 , ϕ2 ).

Definition 1.2. Let ρ : F → N be a type and X a set of variables disjoint from F . The term
algebra Tρ ( X ) of type ρ over X is the unique algebra of type ρ whose universe is Tρ ( X ) and
with basic n-ary operations f defined, for every ϕ1 , . . . , ϕn ∈ Tρ ( X ), as

f Tρ (X ) ( ϕ1 , . . . , ϕn ) := f ( ϕ1 , . . . , ϕn ).

When no confusion might arise, we drop the subscript and write T ( X ) instead of
Tρ ( X ). Given a term ϕ ∈ Tρ ( X ), we write ϕ( x1 , . . . , xn ) to indicate that the variables
occurring in ϕ are among x1 , . . . , xn . Furthermore, given an algebra A of type ρ and
elements a1 , . . . , an ∈ A, we define an element

ϕ A ( a1 , . . . , a n )

of A, by recursion on the construction of ϕ, as follows:

(i) if ϕ is a variable xi , then ϕ A ( a1 , . . . , an ) := ai ;

(ii) if ϕ is a constant c, then c A is the interpretation of c in A;

2
1.2. Basic constructions

(iii) if ϕ = f (ψ1 , . . . , ψm ), then

ϕ A ( a1 , . . . , an ) := f A (ψ1A ( a1 , . . . , an ), . . . , ψm
A
( a1 , . . . , an )).

An equation of type ρ over X is an expression of the form ϕ ≈ ψ, where ϕ, ψ ∈ Tρ ( X ).


We denote by Eρ ( X ) the set of equations of type ρ over X. Such an equation ϕ ≈ ψ is valid
in an algebra A of type ρ, if

ϕ A ( a1 , . . . , an ) = ψ A ( a1 , . . . , an ), for every a1 , . . . , an ∈ A,

in which case we say that A satisfies ϕ ≈ ψ.


For instance, groups are precisely the algebras of type ρG that satisfy the equations

x + (y + z) ≈ ( x + y) + z x+0 ≈ x 0+x ≈ x x + −x ≈ 0 − x + x ≈ 0.

Similarly, lattices are the algebras of type ρ L that satisfy the equations

x∧x ≈ x x∨x ≈ x (idempotent laws)


x∧y ≈ y∧x x∨y ≈ y∨x (commutative laws)
x ∧ (y ∧ z) ≈ ( x ∧ y) ∧ z x ∨ (y ∨ z) ≈ ( x ∨ y) ∨ z (associative laws)
x ∧ (y ∨ x ) ≈ x x ∨ (y ∧ x ) ≈ x. (absorption laws)

From now on, we will work with a fixed denumerable set of variables

Var = { xn : n ∈ N}.

Accordingly, when we write x, y, z . . . for variables, it should be understood that these are
variables in Var.

1.2 Basic constructions


Algebras of the same type are called similar and can be compared by means of maps that
preserve their structure.

Definition 1.3. Given two similar algebras A and B, a homomorphism from A to B is a map
f : A → B such that, for every n-ary operation g of the common type and a1 , . . . , an ∈ A,

f ( g A ( a1 , . . . , an )) = g B ( f ( a1 ), . . . , f ( an )).

An injective homomorphism is called an embedding and, if there exists an embedding


from A to B, we say that A embeds into B. Lastly, a surjective embedding is called an
isomorphism. Accordingly, A and B are said to be isomorphic if there exists an isomorphism
between them, in which case we write A ∼ = B.
A simple induction on the construction of terms shows that, for every pair of algebras
A and B of type ρ and every term ϕ( x1 , . . . , xn ) of ρ, if f is a homomorphism from A to B,
then
f ( ϕ A ( a1 , . . . , an )) = ϕ B ( f ( a1 ), . . . , f ( an )),

3
1. P RELIMINARIES

for every a1 , . . . , an ∈ A. Therefore homomorphisms preserve not only basic operations,


but also arbitrary terms.
In the particular case where A and B are lattices, a homomorphism from A to B is a
map f : A → B such that, for every a, c ∈ A,

f ( a ∧ A c) = f ( a) ∧ B f (c) and f ( a ∨ A c) = f ( a) ∨ B f (c).

For instance, the inclusion map from the lattice hN; 6i into the lattice hZ; 6i is an injective
homomorphism, that is, an embedding. Similarly, given two sets Y ⊆ X, the inclusion
map from the powerset lattice hP (Y ); ⊆i to the powerset lattice hP ( X ); ⊆i is also an
embedding. On the other hand, if Y ( X, the map

(−) ∩ Y : P ( X ) → P (Y )

that sends every Z ⊆ X to Z ∩ Y is a noninjective homomorphism from hP ( X ); ⊆i to


hP (Y ); ⊆i.
Proposition 1.4. Let A be an algebra of type ρ and X a set of variables. Every function f : X → A
extends uniquely to a homomorphism f ∗ : Tρ ( X ) → A.

Proof. The unique extension f ∗ is defined, for every ϕ( xα1 , . . . xαn ) ∈ Tρ ( X ), as

f ∗ ( ϕ) = ϕ A ( f ( xα1 ), . . . , f ( xαn )). 

Exercise 1.5. Prove the above proposition. 


Corollary 1.6. If f : T (Var ) → A and g : B → A are homomorphisms and g is surjective, there
exists a homomorphism h : T (Var ) → B such that f = g ◦ h.

Proof. As g : B → A is surjective, for every n ∈ N there exists some bn ∈ B such that


f ( xn ) = g(bn ). By Proposition 1.4, there exists a homomorphism h : T (Var ) → B such that
h( xn ) = bn , for all n ∈ N. As a consequence,

g ◦ h( xn ) = g(bn ) = f ( xn ), for all n ∈ N.

Since Var is a set of generators for T (Var ) and both g ◦ h and f are homomorphisms, we
conclude that f = g ◦ h. 
Definition 1.7. Let A and B be algebras of the same type ρ : F → N. Then A is said to be
a subalgebra of B if A ⊆ B and f A is the restriction of f B to A, for every f ∈ F . In this case,
we write A 6 B.

Given a class of algebras K, let

I(K) := { A : A ∼
= B for some B ∈ K}
S(K) := { A : A 6 B for some B ∈ K}.

When K = { A}, we write I( A) and S( A) as a shorthand for I({ A}) and S({ A}), respec-
tively. The following observation is an immediate consequence of the definitions.

4
1.2. Basic constructions

Proposition 1.8. Let A and B be algebras of the same type. Then A ∈ IS( B) if and only if there
exists an embedding f : A → B. In this case, A is isomorphic to the unique subalgebra of B with
universe f [ A].

As we mentioned, homomorphisms can be used to compare similar algebras.

Definition 1.9. Given two similar algebras A and B, we say that A is a homomorphic image
of B if there exists a surjective homomorphism f : B → A.

Accordingly, given a class of algebras K, we set

H(K) := { A : A is a homomorphic image of some B ∈ K}.

As usual, when K = { A}, we write H( A) as a shorthand for H({ A}).


Observe that every (not necessarily surjective) homomorphism f : A → B induces a
homomorphic image of A.

Proposition 1.10. If f : A → B is a homomorphism, then f [ A] is the universe of a subalgebra of


B that, moreover, is a homomorphic image of A.

Proof. Observe that f [ A] is nonempty, because A is. Then consider an n-ary function
symbol g of the common type of A and B and b1 , . . . , bn ∈ f [ A]. Clearly, there are
a1 , . . . , an ∈ A such that f ( ai ) = bi , for every i 6 n. Since f is a homomorphism from A to
B, we obtain

g B (b1 , . . . , bn ) = g B ( f ( a1 ), . . . , g( an )) = f ( g A ( a1 , . . . , an )) ∈ f [ A].

Hence, we conclude that f [ A] is the universe of a subalgebra f [ A] of B.


Furthermore, f : A → f [ A] is a homomorphism, because for every basic n-ary function
symbol g of the common type and a1 , . . . , an ∈ A,

f ( g A ( a1 , . . . , an )) = g B ( f ( a1 ), . . . , f ( an )) = g f [ A] ( f ( a1 ), . . . , f ( an )),

where the first equality follows from the assumption that f : A → B is a homomorphism.
Since the map f : A → f [ A] is surjective, we conclude that f [ A] ∈ H( A). 

In view of the above result, when f : A → B is a homomorphism, we denote by f [ A] the


unique subalgebra of B with universe f [ A].
For instance, let f : Z → R be the absolute value map, that is, the function defined by
the rule
f (n) := the absolute value of n.
Observe that f is a nonsurjective homomorphism from the lattice of integers to that of
reals. Furthermore, the homomorphic image f [hZ; 6i] of hZ; 6i is the lattice of natural
numbers hN; 6i, which, in turn, is a subalgebra of lattice of reals.
Notably, the homomorphic images of an algebra A can be “internalized” as special
equivalence relations on A as follows.

5
1. P RELIMINARIES

Definition 1.11. A congruence of an algebra A is an equivalence relation θ on A such that,


for every basic n-ary opearation f of A and a1 , . . . , an , c1 , . . . , cn ∈ A,

if h a1 , c1 i, . . . , h an , cn i ∈ θ, then h f A ( a1 , . . . , an ), f A (c1 , . . . , cn )i ∈ θ. (1.1)

In this case, we often write a ≡θ c as a shorthand for h a, ci ∈ θ. The poset of congruences


of A ordered under the inclusion relation will be denoted by Con( A).

A simple induction on the construction of terms shows that, for every congruence θ of
A and every term ϕ( x1 , . . . , xn ),

if h a1 , c1 i, . . . , h an , cn i ∈ θ, then h ϕ A ( a1 , . . . , an ), ϕ A (c1 , . . . , cn )i ∈ θ,

for every a1 , . . . , an ∈ A. Therefore, congruences preserve not only basic operations, but
also arbitrary terms. Furthermore, a simple argument shows that Con( A) is a complete
(indeed algebraic) lattice whose maximum is the total relation A × A and whose minimum
is the identity relation id A := {h a, ai : a ∈ A}.

Example 1.12 (Heyting algebras). Recall that a filter of a Heyting algebra A is a nonempty
upset F ⊆ A closed under binary meets. We denote by Fi( A) the poset of filters of
A ordered under the inclusion relation. It is easy to see Fi( A) is a complete lattice.
Furthermore, the lattices Fi( A) and Con( A) are isomorphic via the inverse isomorphisms

Ω A (−) : Fi( A) → Con( A) and τ A (−) : Con( A) → Fi( A)

defined by the rules

Ω A ( F ) := {h a, ci ∈ A × A : a → c, c → a ∈ F }
τ A (θ ) := { a ∈ A : h a, 1i ∈ θ }.

Because of this, every congruence θ of a Heyting algebra A is induced by some filter F, in


the sense that θ = Ω A F. 
Example 1.13 (Modal algebras). A modal algebra is an algebra A = h A; ∧, ∨, ¬, , 0, 1i such
that h A; ∧, ∨, ¬, 0, 1i is a Boolean algebra and  is a unary operation such that

( a ∧ c) = a ∧ c and 1 = 1,

for every a, c ∈ A. An open filter of a modal algebra A is a filter of the Boolean reduct
of A that, moreover, is closed under the operation . The poset of open filters of A
ordered under the inclusion relation will be denoted by Op( A). It forms a complete lattice.
Furthermore, the lattices Op( A) and Con( A) are isomorphic via the inverse isomorphisms
described in Example 1.12. Because of this, every congruence of a modal algebra A has
the form θ = Ω A F, for some open filter F. 
Example 1.14 (Groups). Similarly, it is well known that the lattice of congruences of a
group is isomorphic to that of its normal subgroups. Because of this, every congruence of
a group is induced by some normal subgroup. 

6
1.2. Basic constructions

As we mentioned, there is a tight correspondence between the homomorphic images


and the congruences of an algebra A. On the one hand, every congruence θ of A gives
rise to a homomorphic image A/θ of A. Let F be the set of function symbols of A. Given
θ ∈ Con( A) and a basic n-ary function symbol f ∈ F , let f A/θ be the n-ary operation on
A/θ defined by the rule

f A/θ ( a1 /θ, . . . , an /θ ) := f A ( a1 , . . . , an )/θ.

Notice that f A/θ is well-defined, by condition (1.1). As a consequence, the structure

A/θ := h A/θ; { f A/θ : f ∈ F }i

is a well-defined algebra of the type as A. Furthermore, A/θ ∈ H( A), because the map
πθ : A → A/θ, defined, for every a ∈ A, as πθ ( a) := a/θ, is a surjective homomorphism
from A to A/θ. To prove this, consider a1 , . . . , an ∈ A. We have

πθ ( f A ( a1 , . . . , an )) = f A ( a1 , . . . , an )/θ
= f A/θ ( a1 /θ, . . . , an /θ )
= f A/θ (πθ ( a1 ), . . . , πθ ( an )),

where the second equality follows from the definition of the operation f A/θ .

Corollary 1.15. If θ is a congruence of an algebra A, then A/θ is a well-defined homomorphic


image of A.

In view of the above result, every congruence θ of an algebra A induces a homomorphic


image of A, namely A/θ. The converse is also true, as we proceed to explain.

Definition 1.16. The kernel of a homomorphism f : A → B is the binary relation

Ker( f ) := {h a, ci ∈ A × A : f ( a) = f (c)}.

Proposition 1.17. The kernel of a homomorphism f : A → B is a congruence of A.

Proof. It is obvious that Ker( f ) is an equivalence relation on A. Therefore, to prove that


Ker( f ) is a congruence of A, it suffices to show that it preserves the basic operations
of A. Consider a basic n-ary operation g of A and a1 , . . . , an , c1 , . . . , cn ∈ A such that
h a1 , c1 i, . . . , h an , cn i ∈ Ker( f ). By the definition of Ker( f ),

f ( ai ) = f (ci ), for every i 6 n.

It follows that g B ( f ( a1 ), . . . , f ( an )) = g B ( f (c1 ), . . . , f (cn )). Since f : A → B is a homomor-


phism, this yields

f ( g A ( a1 , . . . , an )) = g B ( f ( a1 ), . . . , f ( an )) = g B ( f (c1 ), . . . , f (cn )) = f ( g A (c1 , . . . , cn )).

Hence, we conclude that h g A ( a1 , . . . , an ), g A (c1 , . . . , cn )i ∈ Ker( f ), as desired. 


The behaviour of kernels is governed by the next principle.

7
1. P RELIMINARIES

Fundamental Homomorphism Theorem 1.18. If f : A → B is a homomorphism with kernel


θ, then there exists a unique embedding g : A/θ → B such that f = g ◦ πθ .

Proof. We begin by proving the existence of g. Let g : A/θ → B be the map defined as
g( a/θ ) := f ( a), for every a ∈ A. To show that g is well-defined, consider a, c ∈ A such
that a/θ = c/θ. Since θ = Ker( f ), this means that f ( a) = f (c), as desired. Furthermore,
the definition of g guarantees that f = g ◦ πθ .
Now, observe g is injective, because, for every a, c ∈ A such that g( a/θ ) = g(c/θ ), we
have f ( a) = f (c), that is, h a, ci ∈ Ker( f ) = θ and, therefore, a/θ = c/θ. Moreover, for
every basic n-ary operation p of A and a1 , . . . , an ∈ A, we have

g( p A/θ ( a1 /θ, . . . , an /θ )) = g( p A ( a1 , . . . , an )/θ )


= f ( p A ( a1 , . . . , an ))
= p B ( f ( a1 ), . . . , f ( an ))
= p B ( g( a1 /θ ), . . . , g( an /θ )).

The first equality above follows from the definition of A/θ, the second and the last from
the definition of g and the third from the assumption that f : A → B is a homomorphism.
Hence, we conclude that g : A/θ → B is a homomorphism and, therefore, an embedding,
as desired.
The uniqueness of g follows from the fact that, if a map g∗ satisfies the condition in the
statement of the theorem, then, for every a ∈ A,

f ( a) = g∗ ◦ πθ ( a) = g∗ ( a/θ ),

that is, g∗ coincides with g. 


Corollary 1.19. If f : A → B is a homomorphism, then f [ A] ∼
= A/Ker( f ). In particular, if f is
surjective, B ∼
= A/Ker( f ).
Proof. In the proof of the Fundamental Homomorphism Theorem we showed that the
map g : A/Ker( f ) → B, defined by the rule g( a/Ker( f )) := f ( a), is an embedding of
A/Ker( f ) into B. As g can be viewed as a surjective embedding of A/Ker( f ) into f [ A],
we conclude that f [ A] ∼
= A/Ker( f ). 
At this stage, it should be clear that if θ is a congruence on an algebra A, then πθ : A →
A/θ is a surjective homomorphism whose kernel is θ. Similarly, if f : A → B is a surjective
homomorphism, then A/Ker( f ) ∼ = B, by Corollary 1.19. As a consequence, for every class
of algebras K,
H(K) = I{ A/θ : A ∈ K and θ ∈ Con( A)}. (1.2)
Now, recall that the Cartesian product of a family of sets { Ai : i ∈ I } is the set

∏ Ai := { f : I →
[
Ai : f (i ) ∈ Ai , for all i ∈ I }.
i∈ I i∈ I

In particular, if I is empty, then ∏i∈ I Ai is the singleton containing only the empty map.

8
1.2. Basic constructions

Definition 1.20. The direct product of a family of similar algebras { Ai : i ∈ I } is the unique
algebra of the common type whose universe is the Cartesian product ∏i∈ I Ai and such
that, for every basic n-ary operation symbol f and every ~a1 , . . . ,~an ∈ ∏i∈ I Ai ,

f ∏i∈ I Ai (~a1 , . . . ,~an )(i ) = f Ai (~a1 (i ), . . . ,~an (i )), for every i ∈ I.

We denote this algebra by ∏i∈ I Ai .

In this case, for every j ∈ I, the projection map on the j-th component p j : ∏i∈ I Ai →
A j , defined by the rule p j (~a) := ~a( j), is a surjective homomorphism from ∏i∈ I Ai to A j .
Given a class of similar algebras K, we set

P(K) := { A : A is a direct product of a family { Bi : i ∈ I } ⊆ K}.

As usual, when K = { A}, we write P( A) as a shorthand for P({ A}).


Notice that up to isomorphism, there exists a unique one-element algebra of a given
type. Because of this, one-element algebras are called trivial. Accordingly, when the set of
indexes I is empty, the direct product ∏i∈ I Ai is the trivial algebra of the given type. It
follows that P(K) contains always a trivial algebra, for every class of similar algebras K.

Example 1.21 (Powerset algebras). Boolean algebras of the form hP ( X ); ∩, ∪, −, ∅, X i are


called powerset Boolean algebras. Let B be the two-element Boolean algebra and observe
that IP( B) is the class of algebras isomorphic to some powerset Boolean algebra. To prove
this, observe that every powerset Boolean algebra P ( X ) is isomorphic to a direct product
of B via the characteristic function f X : P ( X ) → ∏ x∈X Bx , defined by the rule

1 if x ∈ Y
f (Y )( x ) :=
0 if x ∈
/ Y,

where Y ∈ P ( X ) and x ∈ X. By the same token, every direct product ∏i∈ I Bi of B is


isomorphic to the powerset Boolean algebra P ( I ) via the isomorphism f I . 

We close this section by reviewing the subdirect product construction.

Definition 1.22. A subalgebra B of a direct product ∏i∈ I Ai is said to be a subdirect product


of { Ai : i ∈ I } if the projection map pi is surjective, for every i ∈ I. Similarly, an
embedding f : B → ∏i∈ I Ai is said to be subdirect when f [ B] is a subdirect product of the
family { Ai : i ∈ I }.

Given a class of similar algebras K, we set

PSD (K) := { A : A is a subdirect direct product of a family { Bi : i ∈ I } ⊆ K}.

As usual, when K = { A}, we write PSD ( A) as a shorthand for PSD ({ A}). Clearly, PSD (K) ⊆
SP(K). Furthermore, PSD (K) contains always a trivial algebra.

9
1. P RELIMINARIES

Example 1.23 (Distributive lattices). Let DL be the class of distributive lattices and B
be the two-element distributive lattice. Birkhoff’s Representation Theorem states that
DL = IPSD ( B). The inclusion IPSD ( B) ⊆ DL follows from the fact that DL is closed under
I, S and P. For the other inclusion, consider a distributive lattice A and let I be the set of
its prime filters. By Birkhoff’s Representation Theorem, the map
γ: A → ∏ BF ,
F∈ I

defined, for every a ∈ A and F ∈ I, by the rule



1 if a ∈ F
γ( a)( F ) :=
0 if a ∈
/ F,
is a well-defined subdirect embedding. 
Example 1.24 (Boolean algebras). Similarly, Stone’s Representation Theorem states that
the class of Boolean algebras coincides with IPSD ( B), where B the two-element Boolean
algebra. 
The next result provides a general recipe to construct subdirect products.
Proposition 1.25. Let A be an algebra and {θi : i ∈ I } ⊆ Con( A). Then the map

∏ A/θi ,
\
f : A/ θi →
i∈ I i∈ I

defined, for every a ∈ A and j ∈ I, as


\
f ( a/ θi )( j) := a/θ j ,
i∈ I

is a subdirect embedding.
Proof. For the sake of readability, set B := A/ i∈ I θi . To prove that f is injective, consider
T

a, c ∈ A such that h a, ci ∈ / i∈ I θi . Then there exists j ∈ I such that h a, ci ∈


T
/ θ j and,
therefore, \ \
f ( a/ θi )( j) := a/θ j 6= c/θ j = f (c/ θi )( j).
i∈ I i∈ I
It follows that f ( a/ i∈ I θi ) 6= f (c/ i∈ I θi ). Thus, f is injective. Moreover, by the defini-
T T

tion of f , the composition pi ◦ f : B → A/θi is surjective, for every i ∈ I.


It only remains to prove that f is a homomorphism. Consider an n-ary basic operation
g and a1 , . . . , an ∈ A. For every j ∈ I, we have
f ( g B ( a1 / θi ))( j) = f ( g A ( a1 , . . . , an )/
\ \ \
θi , . . . , a n / θi )( j)
i∈ I i∈ I i∈ I
= g A ( a1 , . . . , an )/θ j
= g A/θ j ( a1 /θ j , . . . , an /θ j )
= g A/θ j ( f ( a1 /
\ \
θi )( j), . . . , f ( an / θi )( j))
i∈ I i∈ I
∏ A/θi
\ \
= g i ∈ I ( f ( a1 / θ i ), . . . , f ( an / θi ))( j).
i∈ I i∈ I

10
1.2. Basic constructions

It follows that

f ( g B ( a1 / θi )) = g∏i∈ I A/θi ( f ( a1 /
\ \ \ \
θi , . . . , a n / θ i ), . . . , f ( a n / θi )). 
i∈ I i∈ I i∈ I i∈ I

11
CHAPTER 2
Completeness theorems

2.1 Propositional logics


For general information on propositional logics we refer the reader to [20, 35, 42, 55, 56,
57, 115, 119]. Recall that a closure operator on a set A is a map C : P ( A) → P ( A) such that,
for every X ⊆ Y ⊆ A,

X ⊆ C ( X ) = C (C ( X )) and C ( X ) ⊆ C (Y ).

Given a closure operator C on A, a subset X ⊆ A is said to be closed if X = C ( X ). A closure


system on A is a family C ⊆ P ( A) that contains A and such that F , for every nonempty
T

F ⊆ C . Closure operators and systems on A are two faces of the same coin. More precisely,
the family of closed sets of a closure operator on A is a closure system on A. On the other
hand, if C is a closure system on A, then the map C : P ( A) → P ( A), defined by the rule
\
C ( X ) := {Y ∈ C : X ⊆ Y } ,

is a closure operator on A. These transformations between closure operators and systems


on A are one inverse to the other.
Exercise 2.1. Prove that these transformations are well-defined and one inverse to the
other. 
Another way of presenting closure operators or systems is by means of the following
concept.

Definition 2.2. A consequence relation on a set A is a binary relation ` ⊆ P ( A) × A such


that, for every X ∪ Y ∪ { a} ⊆ A,

(i) if a ∈ X, then X ` a; and

(ii) if X ` y for all y ∈ Y and Y ` a, then X ` a.

13
2. C OMPLETENESS THEOREMS

Furthermore, ` is said to be finitary when, for every X ∪ { a} ⊆ A,

if X ` a, there exists a finite Y ⊆ X such that Y ` a.

Remark 2.3. The relation X ` a should be read, intuitively, as “X proves a” or “a follows


from X”. In this reading, the demand expressed by condition (i) is rather natural, while
(ii) is an abstract of the Cut rule. 
Formally speaking, a consequence relation on a set A is a binary relation ` ⊆ P ( A) × A.
However, to simplify the notation, we will often write a1 , . . . , an ` c as a shorthand for
{ a1 , . . . , an } ` c. Similarly, we will use X, a ` c as a shorthand for X ∪ { a} ` c. Lastly, for
every set of formulas X ∪ Y ∪ { a, c}, we write

(i) X ` Y, when X ` y for every y ∈ Y;

(ii) a a ` c, when a ` c and c ` a; and

(iii) X a ` Y, when X ` Y and Y ` X.

Definition 2.4. Let ` be a consequence relation on a set A. A theory of ` is a subset X ⊆ A


such that, for every a ∈ A, if X ` a, then a ∈ X. The set of theories of A will be denoted
by T h(`).

It is easy to see that T h(`) is a closure system on A. Moreover, given a closure operator
C on A, the following is a consequence relation on A:

{h X, ai ∈ P ( A) × A : a ∈ C ( X )}.

Together with the correspondence between closure systems and operators, these trans-
formations induce a one-to-one correspondence between consequence relations, closure
operators and closure systems on A.
Exercise 2.5. Prove these facts. 
In the context of logic, the term algebra Tρ (Var ) is often called the algebra of formulas
(of type ρ) and its elements are referred to as formulas. An endomorphism of an algebra A is
a homomorphism whose domain and codomain is A. Endomorphisms of the algebra of
formulas play a fundamental role in logic.

Definition 2.6. A substitution of type ρ is an endomorphism σ of Tρ (Var ).

When the type ρ is clear from the context, we will simply say that σ is a substitution.
In view of Proposition 1.4 and of the fact that Var is a set of generators for Tρ (Var ),
every function σ : Var → Tρ (Var ) can be uniquely extended to a substitution σ+ of type ρ,
namely the function defined by the rule

ϕ( x1 , . . . , xn ) 7−→ ϕ(σ( x1 ), . . . , σ( xn )).

Because of this, substitutions can be presented by exhibiting functions σ : Var → Tρ (Var ).

14
2.1. Propositional logics

Definition 2.7. A logic of type ρ is a consequence relation ` on the set of formulas Tρ (Var )
that, moreover, is substitution invariant in the sense that for every substitution σ of type ρ
and every set of formulas Γ ∪ { ϕ} ⊆ Tρ (Var ),

if Γ ` ϕ, then σ[ Γ ] ` σ( ϕ).

Remark 2.8. As mentioned above, Γ ` ϕ should be read as “Γ proves ϕ” or “ϕ follows


from Γ”. The requirement that ` is substitution invariant, instead, is intended to capture
the idea that logical inferences are true only in virtue of their form (as opposed to their
content). 

Example 2.9 (Hilbert calculi). We work within a fixed, but arbitrary, type ρ. A rule is an
expression of the form Γ  ϕ, where Γ ∪ { ϕ} ⊆ Tρ (Var ). In this case, Γ is said to be the
set of premises of the rule and ϕ the conclusion. When Γ = ∅, the rule Γ  ϕ is sometimes
called an axiom. A rule Γ  ϕ is said to be valid in a logic `, when Γ ` ϕ. A Hilbert calculus
is a set of rules.
Every Hilbert calculus H induces a logic, as we proceed to explain. Consider a set
of formulas Γ ∪ { ϕ} ⊆ Tρ (Var ). A proof of ϕ from Γ in H is a well-ordered sequence
hψα : α 6 γi of formulas ψα ∈ Tρ (Var ) whose last element ψγ is ϕ and such that, for every
α 6 γ, either ψα ∈ Γ or there exist a substitution σ and a rule ∆  δ in H such that the
formulas in σ[∆] occur in the initial segment hψβ : β < αi and ψα = σ(δ).
The logic `H induced by H is defined, for every Γ ∪ { ϕ} ⊆ Tρ (Var ), as

Γ `H ϕ ⇐⇒ there exists a proof of ϕ from Γ in H.

As expected, `H is a logic in the sense of Definition 2.7. Furthermore, it is the least logic `
such that Γ ` ϕ, for every rule Γ  ϕ in H.
A logic ` is said to be axiomatized by a Hilbert calculus H when it coincides with `H .
Notice that every logic ` is vacuously axiomatized by the Hilbert calculus

{ Γ  ϕ : Γ ` ϕ }.

Because of this, axiomatizations in terms of Hilbert calculi H acquire special interest when
H is finite or, at least, recursive. 

When no confusion shall arise, given a sequence ~a and a set A, we write ~a ∈ A


to indicate that the elements of the sequence ~a belong to A. The following concept is
instrumental to exhibit further examples of logics.

Definition 2.10. Let K be a class of similar algebras We define a binary relation K ⊆


P ( Eρ (Var )) × Eρ (Var ) as follows:

Θ K ε ≈ δ ⇐⇒ for every A ∈ K and every ~a ∈ A,


if ϕ A (~a) = ψ A (~a) for all ϕ ≈ ψ ∈ Θ, then ε A (~a) = δ A (~a).

The relation K is known as the equational consequence relative to K.

15
2. C OMPLETENESS THEOREMS

Example 2.11 (Equationally defined logics). We work within a fixed, but arbitrary, type ρ.
Given a set of equations τ ( x ) in a single variable x and a set of formulas Γ ∪ { ϕ} ⊆ T (Var ),
we abbreviate
[
{ε( ϕ) ≈ δ( ϕ) : ε ≈ δ ∈ τ } as τ ( ϕ), and τ (γ) as τ [ Γ ].
γ∈ Γ

Given a class of algebras K and a set of equations τ ( x ), we define a logic `K,τ as


follows: for every Γ ∪ { ϕ} ⊆ T (Var ),

Γ `K,τ ϕ ⇐⇒ τ [ Γ ] K τ ( ϕ).

It is easy to prove that `K,τ is indeed a logic in the sense of Definition 2.7. Notice that, in
this case, ` is related to K by a completeness theorem witnessed by the set of equations τ ( x )
that allows to translate formulas into equations and, therefore, to interpret `K,τ into K .
For instance, the completeness theorem of classical propositional logic CPC with
respect to the class of Boolean algebras BA states precisely that CPC coincides with `BA,τ
where τ = { x ≈ 1}. Similarly, the completeness theorem of intuitionistic propositional
logic IPC with respect to the class of Heyting algebras HA states precisely that IPC
coincides with `HA,τ where τ = { x ≈ 1}. Because of this, CPC and IPC can be defined as
follows: for every set of formulas Γ ∪ { ϕ} of the appropriate type,

Γ `CPC ϕ ⇐⇒ τ [ Γ ] BA τ ( ϕ)
Γ `IPC ϕ ⇐⇒ τ [ Γ ] HA τ ( ϕ),

where τ = { x ≈ 1}. 

2.2 Algebraic semantics


The relation between logic and algebra is often explained in terms of the existence of
equational completeness theorems. The following definition makes this concept precise.
As we will see, however, equational completeness theorems alone are not sufficient to
account for the relation between logic and algebra.

Definition 2.12. A logic ` is said to admit an equational completeness theorem if there are a
set of equations τ ( x ) and a class K of algebras such that for all Γ ∪ { ϕ} ⊆ T (Var ),

Γ ` ϕ ⇐⇒ τ [ Γ ] K τ ( ϕ).

In this case, ` coincides with `K,τ and K is said to be a τ-algebraic semantics (or simply an
algebraic semantics) for `.

This notion was introduced in [21] and studied in depth in [28, 104, 113]. For instance, the
classes of Boolean and Heyting algebras are, respectively, τ-algebraic semantics for CPC
and IPC where τ = { x ≈ 1}.
Another familiar example of equational completeness theorem arises from the field of
modal logic. Let Fr be the class of all Kripke frames. We can associate two distinct logics

16
2.2. Algebraic semantics

with Fr, see for instance [89, 90]. The global consequence Kg of the modal system K is the
logic defined, for every set of modal formulas Γ ∪ { ϕ}, as follows:

Γ `Kg ϕ ⇐⇒ for every hW, Ri ∈ Fr and evaluation v in hW, Ri,


if w, v Γ for all w ∈ W, then w, v ϕ for all w ∈ W.

On the other hand, the local consequence K` of the modal system K is defined, for every set
of modal formulas Γ ∪ { ϕ}, as follows:

Γ `K` ϕ ⇐⇒ for every hW, Ri ∈ Fr, w ∈ W, and evaluation v in hW, Ri,


if w, v Γ, then w, v ϕ.

It is easy to see that Kg and K` are logics. Moreover, they are distinct, because

x `Kg x and x 0K` x. (2.1)

Exercise 2.13. Prove that Kg and K` are logics. Notice also that the modal system K is not
a logic itself, because it is not a consequence relation. Indeed, there are two ways to turn
K into a logic, namely, Kg and K` . 
Exercise 2.14. Prove that Kg and K` have the same theorems, i.e., formulas provable from
the empty set. Prove also that the set of their theorems is the modal system K. This
indicates that, even in the modal setting, logics should not be identified with their sets of
theorems. 
The global consequence Kg is related to the class MA of modal algebras by the following
equational completeness theorem.

Theorem 2.15. For every set Γ ∪ { ϕ} of modal formulas,

Γ `Kg ϕ ⇐⇒ τ [ Γ ] MA τ ( ϕ),

where τ = { x ≈ 1}. Consequently, the class of modal algebras is a τ-algebraic semantics for Kg .

In order to prove it, recall that a filter on a Boolean algebra A is said to be proper when
it differs from A. Moreover, a proper filter U of A is said to be a ultrafilter of A if it is
maximal among the proper filters of A or, equivalently, if

a ∈ U or ¬ a ∈ U, for every a ∈ A.

While the following result holds in ZFC, it cannot be proved in ZF (although it is strictly
weaker then the axiom of choice).

Ultrafilter Lemma 2.16. Every proper filter on a Boolean algebra can be extended to a ultrafilter.

We are now ready to prove Theorem 2.15.

17
2. C OMPLETENESS THEOREMS

Proof sketch. It suffices to prove that

Γ 0Kg ϕ ⇐⇒ τ [ Γ ] 2MA τ ( ϕ).

Suppose first that Γ 0Kg ϕ. Then there are a Kripke frame hW, Ri, an evaluation v in it
and a world u such that

w, v Γ for all w ∈ W and u, v 1 ϕ. (2.2)

Then consider the complex algebra of hW, Ri, that is, the structure

A := hP (W ); ∩, ∪, −, , ∅, W i,

where − is set theoretic complement and, for every V ⊆ W,

V := {w ∈ W : if hw, ti ∈ R, then t ∈ V }.
It is easy to prove that A is a modal algebra. Then consider the unique homomorphism
f : T (Var ) → A such that
f ( x ) = {w ∈ W : w, v x },
for every x ∈ Var. A simple induction of the construction of terms shows that, for every
formula ψ,
f (ψ) = {w ∈ W : w, v ψ}.
Together with (2.2), this yields

f [ Γ ] ⊆ {W } and f ( ϕ) 6= W.

Hence, we conclude that τ [ Γ ] 2MA τ ( ϕ).


To prove the converse, suppose that τ [ Γ ] 2MA τ ( ϕ). Then there are a modal algebra A
and a homomorphism f : T (Var ) → A such that

f [ Γ ] ⊆ {1} and f ( ϕ) 6= 1.

Then consider the Kripke frame dual to A, that is, the structure hW, Ri, where W is the set
of ultrafilters of A and R the binary relation on W defined as follows:

R := {hU, V i ∈ W × W : { a ∈ A : a ∈ U } ⊆ V }.

Let then v : Var → P (W ) be the evaluation in hW, Ri defined by the rule

v( x ) := {U ∈ W : f ( x ) ∈ U }.

An easy induction on the construction of terms shows that, for every formula ψ,

{U ∈ W : f (ψ) ∈ U } = {U ∈ W : U, v ψ }. (2.3)

Now, since f ( ϕ) 6= 1, the Ultrafilter Lemma guarantees the existence of an ultrafilter F


such that f ( ϕ) ∈/ F. Furthermore, as every ultrafilter contain 1, from f [ Γ ] ⊆ {1} it follows
that f [ Γ ] ⊆ U, for all U ∈ W. In short,

f [ Γ ] ⊆ U for all U ∈ W and f ( ϕ) ∈


/ F.

18
2.2. Algebraic semantics

Together with (2.3), this yields

U, v Γ for all U ∈ W and F, v 1 ϕ.

Hence, we conclude that Γ 0Kg ϕ. 

At this stage, it is tempting to conjecture that the relation between logic and algebra
can be explained in terms of equational completeness theorems only. As we anticipated,
however, this is not the case. For instance, the relation between CPC and BA cannot be
explained in terms of completeness theorems only, because the class of Heyting algebras
HA is also an algebraic semantics for CPC. To explain why, it is convenient to recall the
following classical result relating CPC and IPC [67].

Givenko’s Theorem 2.17. For every set of formulas Γ ∪ { ϕ},

Γ `CPC ϕ ⇐⇒ {¬¬γ : γ ∈ Γ } `IPC ¬¬ ϕ.

As a consequence, we obtain the desired result.

Corollary 2.18. The class of Heyting algebras is an algebraic semantics for CPC.

Proof. For every set of formulas Γ ∪ { ϕ}, we have

Γ `CPC ϕ ⇐⇒ {¬¬γ : γ ∈ Γ } `IPC ¬¬ ϕ


⇐⇒ {¬¬γ ≈ 1 : γ ∈ Γ } HA ¬¬ ϕ ≈ 1.

The first equivalent above is Glivenko’s Theorem, while the second is a consequence of the
completeness theorem of IPC with respect to HA. As a consequence, the class of Heyting
algebras is a τ-algebraic semantics for IPC, where τ = {¬¬ x ≈ 1}. 

This means that the univocal relation between CPC and the class of Boolean algebras
cannot be explained in terms of the existence of completeness theorems only. As we shall
see, this relation arises from a deeper phenomenon, known as algebraizability [21].
Exercise 2.19. One may wonder whether the fact that CPC has many distinct algebraic
semantics cannot be amended by restricting our attention to τ-algebraic semantics where
τ = { x ≈ 1}. This is not the case, as this exercise asks you to check. Let A be the three-
element algebra h{0, 1, a}; ∧, ∨, ¬, 0, 1i where h A; ∧, ∨i is the lattice with order 0 < a < 1
and ¬ : A → A is the map described by the rule

¬0 = ¬ a = 1 and ¬1 = 0.

Clearly, A is not a Boolean algebra (as there is no three-element Boolean algebra). Prove
that { A} is τ-algebraic semantics for CPC where τ = { x ≈ 1}. Hint: use the fact that the
two-element Boolean algebra is a homomorphic image of A. 

19
2. C OMPLETENESS THEOREMS

Indeed the existence of equational completeness theorems between a logic and a class
of algebras turns out to be a very weak relation, as shown in [28, 104]. For instance, while
many interesting logics lack a natural equational completeness theorem, they still admit a
nonstandard one. This is the case of K` , as we proceed to explain.
A logic ` is said to be protoalgebraic if there exists a set ∆( x, y) of formulas such that
∅ ` ∆( x, x ) and x, ∆( x, y) ` y. Notice that all logics ` with a binary connective → such
that ∅ ` x → x and x, x → y ` y are protoalgebraic, as witnessed by the set ∆ := { x → y}.
Furthermore, a logic ` is said to be nontrivial if x 0 y.

Theorem 2.20 (M. [104, Thm. 9.3]). A nontrivial protoalgebraic logic. ` has an algebraic
semantics if and only it there are two distinct formulas ϕ and ψ that are logically equivalent in the
sense that
δ( ϕ, ~z) a ` δ(ψ, ~z), for all δ( x, ~z) ∈ T (Var ).

As a consequence, we obtain the following.

Corollary 2.21. The logic K` has an algebraic semantics.

Proof. Clearly, K` is nontrivial and protoalgebraic. Furthermore, the formulas x and x ∧ x


are distinct, but logical equivalent in K` . Therefore, K` has an algebraic semantics in view
of Theorem 2.20. 

On the other hand, K` lacks any natural equational completeness theorem.

Theorem 2.22 (M. [104, Cor. 9.7]). No class of modal algebras is an algebraic semantics for K` .

Proof. We begin by proving that, for all ϕ, ψ ∈ T (Var ),

MA  ϕ ≈ ψ ⇐⇒ ϕ a `K` ψ. (2.4)

To this end, observe that

MA  ϕ ≈ ψ ⇐⇒ MA  ϕ ↔ ψ ≈ 1
⇐⇒ ∅ `Kg ϕ ↔ ψ
⇐⇒ ∅ `K` ϕ ↔ ψ
⇐⇒ ϕ a `K` ψ.

The above equivalence are justified as follows. The first is an easy property of Boolean
algebras, the second is a consequence of Theorem 2.15, the third holds because Kg and
K` have the same theorems (see Exercise 2.14) and the last one because K` has a standard
deduction theorem.
Now, suppose, with a view to contradiction, that K` has a τ-algebraic semantics
K ⊆ MA. This implies that there exists an equation ε ≈ δ ∈ τ such that MA 2 ε ≈ δ. Thus,
in view of the above display, we can assume, by symmetry, that ε 0K` δ. This means that
there are a Kripke frame X = h X, Ri, an element w ∈ X and a valuation v in X such that
w, v ε and w, v 1 δ.

20
2.2. Algebraic semantics

Let X+ = h X + ; R+ i be the Kripke frame obtained by adding a new point w+ to X and


defining the relation R+ as follows:

h p, qi ∈ R+ ⇐⇒ p = w+ or h p, qi ∈ R.

Let also v+ be the unique evaluation in X+ such that for every y ∈ Var and q ∈ X+ :

q, v+ y ⇐⇒ either (q ∈ X and q, v y) or q = w+ .

From the definition of X+ and v+ it follows that

q, v+ ϕ ⇐⇒ q, v ϕ

for all ϕ ∈ T (Var ) and q ∈ X. Consequently, as w, v ε and w, v 1 δ,

w+ , v+ x and w+ , v+ 1 (ε → δ).

This implies
x 0K` (ε → δ).
On the other hand, clearly ∅ `K` (δ → δ). Consequently,

x, (δ → δ) 0K` (ε → δ). (2.5)

Now, observe that, for every ϕ, ψ ∈ T (Var ),

ε( x ) ≈ δ( x ), ϕ((δ → δ)) ≈ ψ((δ → δ)) K ϕ((ε → δ)) ≈ ψ((ε → δ)).

Since ε ≈ δ ∈ τ ( x ), this implies

τ ( x ), τ ((δ → δ)) K τ ((ε → δ)).

Since K is a τ-algebraic semantics for K` , this yields x, (δ → δ) `K` (ε → δ), a
contradiction with (2.5). 
While, in view of Theorem 2.22, most logics have an algebraic semantics, examples of
logics lacking any algebraic semantics are known since [17].
Exercise 2.23. Prove that K` has a standard deduction theorem, i.e., that for every set of
formulas Γ ∪ {ψ, ϕ},
Γ, ψ `K` ϕ ⇐⇒ Γ `K` ψ → ϕ.
Prove that this is not the case for Kg . 
Exercise 2.24. Prove that no class of distributive lattices is an algebraic semantics for the
h∧, ∨i-fragment CPC∧∨ of CPC. Hint: use the fact that every equation in a single variable
holds in the class of distributive lattices.
Furthermore, prove that CPC∧∨ has a nonstandard algebraic semantics. To this end,
consider the three-element algebra A = h{0+ , 0− , 1}; ∧, ∨i whose binary commutative
operations are defined by the following tables

21
2. C OMPLETENESS THEOREMS

∧ 0− 0+ 1 ∨ 0− 0+ 1
0− 0+ 0+ 0+ 0− 0+ 0+ 1
0+ 0− 0+ 0+ 0− 1
1 1 1 1

and prove that { A} is a τ-algebraic semantics for τ = { x ≈ x ∧ x }. Conclude that CPC∧∨


is another example of logic that admits a nonstandard algebraic semantics, but lacks a
standard one. 

22
CHAPTER 3
Universal algebra

3.1 Ultraproducts
In order to understand the relation between logic and algebra, we need to take a short
detour in universal algebra and the theory of quasi-varieties. We begin by reviewing a
product-like construction known as ultraproduct [10, 34]. First, recall that ultrafilters on
powerset Boolean algebras P ( X ) are also called ultrafilters on X. Then let { Ai : i ∈ I } be a
family of similar algebras. The equalizer J~a = ~c K of a pair of elements ~a, ~c ∈ ∏i∈ I Ai is the
set of indexes on which the sequences ~a and ~c agree, that is,

J~a = ~c K := {i ∈ I : ~a(i ) = ~c(i )}.

Moreover, given an ultrafilter U on the index set I, let θU be the binary relation on the
Cartesian product ∏i∈ I Ai defined as

θU := {h~a, ~c i : J~a = ~c K ∈ U }.

Proposition 3.1. If { Ai : i ∈ I } is a family of similar algebras and U an ultrafilter on I, then θU


is a congruence of ∏i∈ I Ai .

Proof. We begin by proving that θU is an equivalence relation on ∏i∈ I Ai . To this end,


consider ~a,~b, ~c ∈ ∏i∈ I Ai . We have

J~a = ~aK = {i ∈ I : ~a(i ) = ~a(i )} = I.

Observe that I ∈ U, since U is a nonempty upset of P ( I ). Together with the above display,
this yields J~a = ~aK ∈ U and, therefore, h~a,~ai ∈ θU . It follows that θU is reflexive. To prove
that it is symmetric, suppose that h~a, ~c i ∈ θU . Then J~a = ~c K ∈ U. Since J~a = ~c K = J~c = ~aK,
this implies J~c = ~aK ∈ U and, therefore, h~c,~ai ∈ θU . Lastly, to prove that θU is transitive,
suppose that h~a,~bi, h~b, ~c i ∈ θU , that is, J~a = ~bK, J~b = ~c K ∈ U. Since U is closed under
binary meets,
J~a = ~bK ∩ J~b = ~c K ∈ U

23
3. U NIVERSAL ALGEBRA

Clearly, J~a = ~bK ∩ J~b = ~c K ⊆ J~a = ~c K. Since U is an upset of P ( I ), we obtain that


J~a = ~c K ∈ U, whence h~a, ~c i ∈ θU . We conclude that θU is an equivalence relation.
To prove that θU is a congruence, it only remains to show that it preserves the basic
operations. Accordingly, let f be a basic n-ary operation and ~a1 , . . . ,~an , ~c1 , . . . , ~cn ∈ ∏i∈ I Ai
such that
h~a1 , ~c1 i, . . . , h~an , ~cn i ∈ θU .
By definition of θU , this amounts to J~a1 = ~c1 K, . . . , J~an = ~cn K ∈ U. Since U is a filter, it is
closed under finite meets, whence

J~a1 = ~c1 K ∩ · · · ∩ J~an = ~cn K ∈ U. (3.1)

We will show that

J~a1 = ~c1 K ∩ · · · ∩ J~an = ~cn K ⊆ J f ∏i∈ I Ai (~a1 , . . . ,~an ) = f ∏i∈ I Ai (~c1 , . . . , ~cn )K. (3.2)

To this end, consider j ∈ J~a1 = ~c1 K ∩ · · · ∩ J~an = ~cn K. We have

~a1 ( j) = ~c1 ( j), . . . ,~an ( j) = ~cn ( j).

Consequently,

f ∏i∈ I Ai (~a1 , . . . ,~a)( j) = f A j (~a1 ( j), . . . ,~an ( j))


= f A j (~c1 ( j), . . . , ~cn ( j))
= f ∏i∈ I Ai (~c1 , . . . , ~c)( j),

that is, j ∈ J f ∏i∈ I Ai (~a1 , . . . ,~an ) = f ∏i∈ I Ai (~c1 , . . . , ~cn )K. This establishes (3.2). Since U is an
upset of P ( I ), from (3.1) and (3.2) it follows

J f ∏i∈ I Ai (~a1 , . . . ,~an ) = f ∏i∈ I Ai (~c1 , . . . , ~cn )K ∈ U.

Hence, we conclude that h f ∏i∈ I Ai (~a1 , . . . ,~an ), f ∏i∈ I Ai (~c1 , . . . , ~cn )i ∈ θU , as desired. 
In view of the above result, we can make the following definition.

Definition 3.2. An ultraproduct of a family of similar algebras { Ai : i ∈ I } is an algebra of


the form ∏i∈ I Ai /θU , for some ultrafilter U on I.

Given a class of similar algebras K, we set

PU (K) := { A : A is an ultraproduct of a family { Bi : i ∈ I } ⊆ K}.

Notice that PU (K) ⊆ HP(K). Furthermore, as usual, when K = { A}, we write PU ( A) as a


shorthand for PU ({ A}).
Exercise 3.3. Prove that if U is not free (that is, it is principal), then ∏i∈ I Ai /θU is isomorphic
to some Ai . Conclude that if I is finite, then ∏i∈ I Ai /θU belongs to I{ Ai : i ∈ I }. Because
of this, interesting ultraproducts arise from free ultrafilters only. 
Exercise 3.4. Prove that K is a finite set of finite algebras, PU (K) ⊆ I(K). 

24
3.1. Ultraproducts

The importance of ultraproducts is largely due to the following result [30, Thm. V.2.9].
Łoś’ Theorem 3.5. Let { Ai : i ∈ I } be a family of similar algebras, U an ultrafilter on I and
φ( x1 , . . . , xn ) a first order formula. For every ~a1 , . . . ,~an ∈ ∏i∈ I Ai ,

∏ Ai /θU  φ(~a1 /θU , . . . ,~an /θU ) ⇐⇒ {i ∈ I : Ai  φ(~a1 (i), . . . ,~an (i))} ∈ U.


i∈ I

Corollary 3.6. Let { Ai : i ∈ I } be a family of similar algebras, U an ultrafilter on I and φ a


sentence. If φ is valid in all the Ai , then it is valid in ∏i∈ I Ai /θU .
In view Łos’ Theorem, ultraproducts are instrumental to construct nonstandard models
of first order theories. For instance, let N = hN; s, +, ·, 0i be the standard model of Peano
Arithmetic. If U is an ultrafilter on N, the ultraproduct ∏n∈N Nn /U is elementarily
equivalent to N, that is, it satisfies the same sentences as N. On the other hand, it is not
hard to see that if U is free, ∏n∈N Nn /U is uncountable and, therefore, contains many
“infinite” (or nonstandard) natural numbers.
For the present purpose, however, we will not need the full strength of Łoś’ Theorem
and, therefore, we shall omit its proof. Instead, we shall focus on a particular embedding
theorem for ultraproducts that depends on the following notion.
Definition 3.7. A local subgraph X of an algebra A is a finite subset X ⊆ A endowed with
the restriction of finitely many basic operations of A to X.
In this case, X is a finite partial algebra of finite type (even when the type of A is infinite).
Let A and B be similar algebras and X a local subgraph of A. A map f : X → B is said
to be an embedding of X into B if it is injective and, for every basic n-ary operation g of the
type of X and a1 , . . . , an ∈ X such that g A ( a1 , . . . , an ) ∈ X,
f ( g A ( a1 , . . . , an )) = g B ( f ( a1 ), . . . , f ( an )).
Theorem 3.8. Let K ∪ { A} be a class of similar algebras. If every local subgraph of A can be
embedded into some member of K, then A ∈ ISPU (K).
Proof. Let I be the set of local subgraphs of A. By assumption, for every X ∈ I there are
an algebra BX ∈ K and an embedding hX : X → BX . We define a partial order v on I as
follows:
X v Y ⇐⇒ X ⊆ Y and the type of Y extends that of X.
Then, for every X ∈ I, define
JX := {Y ∈ I : X v Y}.
Moreover, let F be the filter of P ( I ) generated by { JX : X ∈ I }. Recall that
F = {Y ⊆ I : JX1 ∩ · · · ∩ JXn ⊆ Y, for some X1 , . . . , Xn ∈ I }.
We will prove that F is proper. To this end, consider X1 , . . . , Xn ∈ I. Then let Y be the
local subgraph of A with universe Y := X1 ∪ · · · ∪ Xn and whose type in the union of the
types of the various Xi . Then
Xi v Y, for every i 6 n,

25
3. U NIVERSAL ALGEBRA

that is, Y ∈ JX1 ∩ · · · ∩ JXn . It follows that ∅ ∈


/ F and, therefore, that F is proper. As F is
a proper filter, by the Ultrafilter Lemma, it can be extended to an ultrafilter U on I.
Now, consider a map
f : A → ∏ BX
X∈ I
such that f ( a)(X) = hX ( a), for every a ∈ A and X ∈ I such that a ∈ X. Moreover, let
f∗: A → ∏ BX /θU
X∈ I

be the map defined by the rule


f ∗ ( a) := f ( a)/θU .
We will show f ∗ is an embedding of A into ∏X∈ I BX /θU .
In order to prove that f ∗ is injective, consider a pair of distinct elements a, c ∈ A.
Consider a local subgraph Y of A containing a and c. We will show that
JY ⊆ {X ∈ I : f ( a)(X) 6= f (c)(X)} (3.3)
Consider X ∈ JY . Then Y v X and, therefore, a, c ∈ Y ⊆ X. Since a, c ∈ X, we have
f ( a)(X) = hX ( a) and f (c)(X) = hX (c).
Furthermore, hX ( a) 6= hX (c), because hX is injective and a 6= c. This yields f ( a)(X) 6=
f (c)(X), establishing (3.3).
Recall that the definition of U guarantees that JY ∈ F ⊆ U. Therefore, since U is an
upset of P ( I ), we can apply (3.3) obtaining
I r J f ( a) = f (c)K = {X ∈ I : f ( a)(X) 6= f (c)(X)} ∈ U.
Since U is a proper filter, this implies
J f ( a ) = f ( c )K ∈
/U
and, therefore,
f ∗ ( a) = f ( a)/θU 6= f (c)/θU = f ∗ (c).
Hence, we conclude that f ∗ is injective.
To prove that it is a homomorphism, consider a basic n-ary operation g and a1 , . . . , an ∈
A. Then consider a local subgraph Y of A whose universe contains a1 , . . . , an , g A ( a1 , . . . , an )
and whose type contains g. We will prove that
JY ⊆ J f ( g A ( a1 , . . . , an )) = g∏X∈ I BX ( f ( a1 ), . . . , f ( an ))K. (3.4)
Consider V ∈ JY . Since Y v V, the type of V contains g and a1 , . . . , an , g A ( a1 , . . . , an ) ∈ V.
Since a1 , . . . , an , g A ( a1 , . . . , an ) ∈ V, we have
f ( a1 )(V) = hV ( a1 )
..
.
f ( an )(V) = hV ( an )
f ( g ( a1 , . . . , an ))(V) = hV ( g A ( a1 , . . . , an )).
A

26
3.1. Ultraproducts

Furthermore, as the type of V contains g,

hV ( g A ( a1 , . . . , an )) = g BV (hV ( a1 ), . . . , hV ( an )).

From the above displays it follows

f ( g A ( a1 , . . . , an ))(V) = g BV ( f ( a1 )(V), . . . , f ( an )(V)) = g∏X∈ I BX ( f ( a1 ), . . . , f ( an ))(V),

that is, V ∈ J f ( g A ( a1 , . . . , an )) = g∏X∈ I BX ( f ( a1 ), . . . , f ( an ))K. This establishes (3.4). Lastly,


as JY ∈ U and U is an upset of P ( I ), condition (3.4) implies

J f ( g A ( a1 , . . . , an )) = g∏X∈ I BX ( f ( a1 ), . . . , f ( an ))K ∈ U,

and, therefore,

f ∗ ( g A ( a1 , . . . , an )) = f ( g A ( a1 , . . . , an ))/θU
= g∏X∈ I BX ( f ( a1 ), . . . , f ( an ))/θU
= g∏X∈ I BX /θU ( f ( a1 )/θU , . . . , f ( an )/θU )
= g∏X∈ I BX /θU ( f ∗ ( a1 ), . . . , f ∗ ( an )).

Hence, we conclude that f ∗ is a homomorphism and, therefore, an embedding of A into


∏Y∈ I BY /θU . As a consequence,

A ∈ ISPU ({ BX : X ∈ I }) ⊆ ISPU (K). 

Corollary 3.9. Every algebra embeds into an ultraproduct of its finitely generated subalgebras.

Exercise 3.10. Consider the consequence relation ` on the set of sentences of a given
algebraic language defined as follows:

Γ ` ϕ ⇐⇒ for every algebra A of the appropriate type,


if A  Γ, then A  ϕ.

Notice that, in view of the Completeness Theorem, ` is classical first order logic for
algebraic languages. This exercise asks you to prove the Compactness Theorem for `
using the ultraproduct construction (the same proof works for relational languages too).
To this end, consider a set of sentences Γ ∪ { ϕ} such that ∆ 0 ϕ, for every finite ∆ ⊆ Γ.
Then we can associate with every finite ∆ ⊆ Γ an algebra A∆ such that

A∆  ∆ and A 2 ϕ.

Prove that there exists an ultraproduct B of the family { A∆ : ∆ is a finite subset of Γ } such
that B  Γ and B 2 ϕ. Then conclude that Γ 0 ϕ, as desired. Hint: use Łoś’ Theorem. 

27
3. U NIVERSAL ALGEBRA

3.2 Universal classes


Definition 3.11. A sentence is said to be universal if it is of the form ∀ x1 , . . . , xn ϕ for some
quantifier free formula ϕ. Accordingly, a class of similar algebras is said to be universal if
it can be axiomatized by a set of universal sentences.

The following concept is instrumental for describing universal classes.

Definition 3.12. Let X be a local subgraph of an algebra A and assume that the universe
and the type of X are, respectively, { a1 , . . . , an } and f 1 , . . . , f t .

(i) The positive atomic diagram of X is the set of equations

D + (X) := { f i ( xm1 , . . . , xmk ) ≈ x j : m1 , . . . mk , j 6 n and i 6 t and f iA ( am1 , . . . , amk ) = a j }.

(ii) The negative atomic diagram of X is the set of negated equations

D − (X) := { xm 6≈ xk : m, k 6 n and am 6= ak }.

Theorem 3.13 (Loś). A class of similar algebras is universal if and only if it is closed under I, S
and PU .

Proof. The implication from left to right follows from the easy observation that the validity
of universal sentences is preserved by I, S and PU . To prove the converse, consider a class
K of similar algebras closed under I, S and PU . Moreover, let Th∀ (K) be the set of universal
sentences valid in K and let K∀ be the class of algebras satisfying Th∀ (K). In order to
conclude the proof, it suffices to show that K = K∀ .
The inclusion K ⊆ K∀ follows immediately from the fact that K  Th∀ (K). To prove
the other inclusion, consider A ∈ K∀ . We will show that every local subgraph of A can
be embedded into an element of K. To this end, consider a local subgraph X of A with
universe { a1 , . . . , an } and take the sentence
 
Φ := ∃ x1 , . . . , xn &D + (X) & D − (X) .
&
Now, suppose, with a view to contradiction, that K  ¬Φ. Since ¬Φ is equivalent to
a universal sentence and A  Th∀ (K), we obtain A  ¬Φ. But this is false, as witnessed
by the assignment x1 7−→ a1 , . . . , xn 7−→ an in A. Thus, we conclude that K 2 ¬Φ.
Consequently, there exists an element B ∈ K such that B  Φ. Let then b1 , . . . , bn be the
elements that witness the validity of the existential part of Φ in B. The map f : X → B
defined by the rule ai 7−→ bi is an embedding of X into B, as desired.
Since every local subgraph of A can be embedded into a member of K, we can apply
Theorem 3.8 obtaining that A ∈ ISPU (K). As, by assumption, K is closed under I, S and
PU , this yields A ∈ K. Hence, we conclude that K∀ ⊆ K. 
Given a class of similar algebras K, the least universal class extending K exists and will
be denoted by U(K) and called the universal class generated by K.

Corollary 3.14. If K is a class of similar algebras, U(K) = ISPU (K).

28
3.2. Universal classes

Proof. From Theorem 3.13 and the fact that U(K) is a universal class it follows that it is
closed under I, S and PU , whence ISPU (K) ⊆ U(K). To prove the reverse inclusion, observe
that U(K) is the class of all algebras satisfying all the universal sentences valid in K. In the
proof of Theorem 3.13, we showed that this guarantees the inclusion U(K) ⊆ ISPU (K). 

Corollary 3.15. Let K ∪ { A} be a class of similar algebras. If A ∈ U(K), then every local
subgraph of A embeds into some member of K.

Proof. This is established in the second part of the proof of Theorem 3.13. 

The following provides an algebraic path to the strong finite model property in logic.

Definition 3.16. A class of similar algebras K is said to have the finite embeddability property
(FEP, for short) if every local subgraph of a member of K can be embedded into a finite
member of K.

Given a class of algebras K, we denote by K<ω the class of its finite members.

Proposition 3.17. A universal class K has the FEP if and only if K = U(K<ω ).

Proof. Suppose first that K has the FEP. Then

K ⊆ ISPU (K<ω ) = U(K<ω ).

The first inclusion in the above display follows from Theorem 3.8 and the second from
Corollary 3.14. Furthermore, since K is a universal class, U(K<ω ) ⊆ K. Therefore, we
conclude that K = U(K<ω ). To prove the converse, suppose that K = U(K<ω ). By Corollary
3.15, every local subgraph of a member of K can be embedded into some element of K<ω ,
that is, K has the FEP. 

The universal theory Th∀ (K) of a class of algebras K is the set of universal sentences
valid in K. Th∀ (K) is said to be decidable when so is the problem of determining whether
a universal sentence is valid in K. Furthermore, we say that a class of algebras is finitely
axiomatizable if it can be axiomatized by finitely many sentences. The following result can
be traced back at least to [97].

Proposition 3.18. Let K be a finitely axiomatizable class of algebras. If K has the FEP, then
Th∀ (K) is decidable.

Proof. In order to prove that Th∀ (K) is decidable it suffices to show that

(i) the problem of determining whether a universal sentence belongs to Th∀ (K) is
semidecidable; and

(ii) the problem of determining whether a universal sentence does not belong to Th∀ (K)
is semidecidable.

29
3. U NIVERSAL ALGEBRA

Condition (i) holds, because K is finitely axiomatizable. Therefore, it only remains to


prove (ii). To this end, let Σ be a finite set of axioms for K and { f 1 , . . . , f n } the function
symbols that appear in Σ. Given a universal sentence Φ, we enumerate the finite models
A1 , A2 . . . of Σ in the type { f 1 , . . . , f n , g1 , . . . , gm }, where g1 , . . . , gm are the function sym-
bols that occur in Φ. This can be done mechanically, because Σ is finite. Our algorithm
tests if Φ fails in some An . If this is the case, it stops and answers that Φ does not belong
to Th∀ (K), otherwise it runs forever.
In order to establish (ii), it suffices to prove that the algorithm stops if and only if
Φ∈ / Th∀ (K). First, it if stops, then Φ fails in some An . Let then B be an algebra of the type
of K obtained by expanding An with an arbitrary interpretation of the missing function
symbols. From An 2 Φ it follows B 2 Φ. Moreover, since An  Σ, we obtain B  Σ and,
therefore, B ∈ K (as Σ axiomatizes K). Hence, we conclude that Φ ∈ / Th∀ (K).
To prove the converse, consider a universal sentence ∀~x ϕ that fails in K. Then there
exist B ∈ K and b1 , . . . , bn ∈ B such that B 2 ϕ(b1 , . . . , bn ). Let X be the local subgraph of B
whose universe is {b1 , . . . , bn } and whose type consists of the function symbols occurring
in ϕ. Since K has the FEP, there is an embedding f : X → C, for some C ∈ K<ω . It follows
that C 2 ϕ( f ( a1 ), . . . , f ( an )), whence ∀~x ϕ. Let C − be the reduct of C is the language L
consisting of the function symbols occurring in Σ ∪ { ϕ}. Clearly, C − 2 ϕ( f ( a1 ), . . . , f ( an ))
and, therefore, ∀~x ϕ fails in C − . As C − is a finite model of Σ in the language L , there must
be some n ∈ N such that C − ∼ = An . It follows that ∀~x ϕ fails in An , whence the algorithm
stops, as desired. 
Example 3.19 (Lattices). We will prove that the class Latt of all lattices has the FEP. For
consider a lattice A and let X be one of its local subgraphs. Let also

B := { a1 ∧ A · · · ∧ A an : a1 , . . . , an ∈ X and n ∈ N}.

Since the operation ∧ A is idempotent, commutative and associative, the set B is finite.
furthermore, B can be viewed as a subposet of A. Let B+ be the poset obtained extending
h B; 6i with a new top element. Since B+ is a finite meet-semilattice with maximum, it is
also a lattice. Furthermore, X embeds into B+ . Hence, Latt has the FEP. By Propositions
3.18 and 3.17, Th∀ (Latt) is decidable [124] and Latt = U(Latt<ω ). On the other hand, the
first order theory of distributive lattices (and, therefore, of any nontrivial equational class
of lattices) is undecidable [70]. 
Example 3.20 (Heyting algebras). We will prove that the class HA of Heyting algebras has
the FEP. For consider a Heyting algebra A and let X be one of its local subgraphs. Then
let B be the bounded sublattice of A generated by X. Notice that B is finite, because it
is a finitely generated bounded distributive lattice. Since B is a finite distributive lattice,
it can be viewed as a finite Heyting algebra B+ . Furthermore, it is easy to see that the
identity map is an embedding of X into B+ . Hence, we conclude that HA has the FEP. By
Propositions 3.18 and 3.17, Th∀ (HA) is decidable and HA = U(HA<ω ) [98]. On the other
hand, the first order theory of nontrivial equational class of Heyting algebras other than
that of Boolean algebras is known to be undecidable [29].
Notably, the fact that Th∀ (HA) is decidable implies that IPC is also decidable, in
the sense that, given a finite set of formulas {γ1 , . . . , γn , ϕ}, we can decide whether

30
3.3. Quasi-varieties

γ1 , . . . , γn `IPC ϕ. This is because, in view of the fact that HA is a { x ≈ 1}-algebraic


semantics for IPC, we have

γ1 , . . . , γn `IPC ϕ ⇐⇒ γ1 ≈ 1, . . . , γn ≈ 1 HA ϕ ≈ 1
⇐⇒ HA  ∀~x ((γ1 ≈ 1 & . . . & γn ≈ 1) =⇒ ϕ ≈ 1).

As ∀~x ((γ1 ≈ 1 & . . . & γn ≈ 1) =⇒ ϕ ≈ 1) is a universal sentence, we can decide whether


it holds in HA or not, because the universal theory of HA is decidable. 

Exercise 3.21. The following proof the class MA of modal algebras has the FEP originates
in [97]. Let X be a local subgraph of a model algebra A. Moreover, let B be the Boolean
subalgebra of A generated by X ∪ {0}. We consider the algebra algebra B+ obtained by
endowing B with a unary operation  defined as follows:
B
+
B b := { A a ∈ B :  A a ∈ B and a 6 b}.
_

Prove that B+ is a modal algebra and that X embeds into B+ . Then conclude that Th∀ (MA)
is decidable and MA = U(MA<ω ). Use these facts to infer that the logic Kg is decidable. 

3.3 Quasi-varieties
For a detailed presentation of the theory of quasi-varieties, we refer the reader to [68, 93].

Definition 3.22. A class of similar algebras closed under I, S, P and PU is said to be a


quasi-variety.

Examples of quasi-varieties include the classes of Boolean, Heyting and modal algebras,
as well as the class of (bounded) distributive lattices and groups. Our aim will be to prove
that quasi-varieties are precisely the classes of algebras axiomatized by the following kind
of first order formulas.

Definition 3.23. A quasi-equation of type ρ is an expression Φ of the form

( ϕ1 ≈ ψ1 & . . . & ϕn ≈ ψn ) =⇒ ε ≈ δ,

where { ϕ1 ≈ ψ1 , . . . , ϕn ≈ ψn , ε ≈ δ} is a set of equations of type ρ. Then Φ is valid in an


algebra A of type ρ when so is its universal closure ∀~x Φ, that is, for every ~a ∈ A,

if ϕ1A (~a) = ψ1A (~a), . . . , ϕnA (~a) = ψnA (~a), then ε A (~a) = δ A (~a).

In this case, we often say that A satisfies Φ and write A  Φ. A quasi-equation is said to be
an equation when its antecedent is empty.

Notably, for every class K of algebras and equations ϕ1 ≈ ψ1 , . . . , ϕn ≈ ψn , ε ≈ δ,

K( &
i 6n
ϕi ≈ ψi ) =⇒ ε ≈ δ iff { ϕ1 ≈ ψ1 , . . . , ϕn ≈ ψn } K ε ≈ δ.

31
3. U NIVERSAL ALGEBRA

Remark 3.24. The reader might have noticed that expressions of the form ε ≈ δ and
∅ =⇒ ε ≈ δ are both called equations. This is not a problem, because they are synonyms,
in the sense that an algebra satisfies ε ≈ δ if and only if it satisfies ∅ =⇒ ε ≈ δ. Because of
this, we will continue to denote equations by ε ≈ δ, while keeping in mind that they can
be viewed as quasi-equations whose antecedent is empty. 
The aim of this section is to prove the following classical result.

Maltsev’s Theorem 3.25. A class of similar algebras is a quasi-variety if and only if it can be
axiomatized by a set of quasi-equations.

Proof. The “only if” part follows from the fact that the validity of quasi-equations is
preserved by the class operators I, S, P and PU . To prove the converse, consider a quasi-
variety K and let Σ the set of quasi-equations valid in K. Let also K+ be the class of algebras
axiomatized by Σ. Our aim is to prove that K = K+ .
The inclusion K ⊆ K+ is straightforward. To prove the other one, consider an algebra
A ∈ K+ . In view of Theorem 3.8, in order to show that A ∈ K, it suffices to prove that
every local subgraph of A embeds in some members of K. This is because, in this case,
A ∈ ISPU (K) ⊆ K. Since K is closed under I, S and PU , this implies A ∈ K, as desired.
Then consider a local subgraph X of A with universe { a1 , . . . , an }. Observe that both
D + (X) and D − (X) are finite sets. Then take an enumeration

D − (X) = {ε 1 6≈ δ1 , . . . , ε t 6≈ δt }.

Moreover, for each i 6 t, consider the quasi-equation


 
Φi := & D + (X) =⇒ ε i ≈ δi .

As witnessed by the natural assignment

x1 7−→ a1 , . . . , xn 7−→ an ,

the quasi-equations Φ1 , . . . , Φt fail in A. Since A satisfies all the quasi-equations valid in


K, this implies that each Φi fails in some Bi ∈ K under an assignment

x1 7−→ b1i , . . . , xn 7−→ bni . (3.5)

Now, consider the map h : X → ( B1 × · · · × Bt ), defined by the rule

a1 7−→ hb11 , . . . , b1t i, . . . , an 7−→ hbn1 , . . . , bnt i.

We will prove that h is an embedding of X into B1 × · · · × Bt . To prove that h is injective,


consider two distinct elements a p , aq ∈ X. Then the formula x p 6≈ xq belongs to D − (X).
Consequently, there exists i 6 t such that
 
Φi = & D + (X) =⇒ x p ≈ xq .

32
3.3. Quasi-varieties

Since Φi fails in Bi under the assignment in (3.5), we obtain bip 6= bqi . As a consequence,

h( a p )(i ) = bip 6= bqi = h( aq )(i )

and, therefore, h( a p ) 6= h( aq ). Hence, h is injective. To prove that it preserves the partial


operations, consider a basic m-ary operation f in the type of X and ak1 , . . . , akm ∈ X such
that f A ( ak1 , . . . , akm ) ∈ X. Then there exists some p 6 n such that a p = f A ( ak1 , . . . , akm ).
Moreover, the equation
f ( xk1 , . . . , xk m ) ≈ x p
belongs to D + (X). As each quasi-equation Φi fails under the assignment in (3.5), the same
assignment satisfies the antecedent of Φi , namely D + (X). It follows that

f Bi (bki 1 , . . . , bki m ) = bip , for each i 6 t.

As a consequence, for every i 6 t,

h( f A ( ak1 , . . . , akm ))(i ) = h( a p )(i )


= bip
= f Bi (bki 1 , . . . , bki m )
= f Bi (h( ak1 )(i ), . . . , h( akm )(i ))
= f B1 ×···× Bt (h( ak1 ), . . . , h( akm ))(i ).

Thus, h( f A ( ak1 , . . . , akm )) = f B1 ×···× Bt (h( ak1 ), . . . , h( akm )). We conclude that h : X → ( B1 ×
· · · × Bt ) is an embedding. Since B1 , . . . , Bt ∈ K and K is closed under P, the direct product
B1 × · · · × Bt belongs to K. Hence, X embeds into some member of K, as desired. 

Exercise 3.26. In view of Łoś’ Theorem quasi-equations persist in ultraproducts. If you are
not familiar with the proof of Łoś’ Theorem, offer a direct proof of this fact. 
Given a class of similar algebras K, the least quasi-variety extending K exists. It will be
denoted by Q(K) and called the quasi-variety generated by K.

Corollary 3.27. Let K be a class of similar algebras. Then Q(K) = ISPPU (K). If in addition K is
a finite set of finite algebras, Q(K) = ISP(K).

Proof. The inclusion ISPPU (K) ⊆ Q(K) is straightforward. To prove the other one, con-
sider A ∈ Q(K). By Maltsev’s Theorem, Q(K) is the class of all algebras satisfying the
quasi-equations valid in K. The proof of the hard part of Maltsev’s Theorem show that
A ∈ ISPU P(K). Therefore, it only remains to show that PU P(K) ⊆ ISPPU (K), which is left
as an exercise. This shows that Q(K) = ISPPU (K).
To prove the second part of the statement, suppose that K is a finite set of finite
algebras. This guarantees that PU (K) ⊆ I(K) (see Exercise 3.4). As a consequence, Q(K) =
ISPPU (K) = ISP(K), as desired. 

Exercise 3.28. Prove that if K is a class of similar algebras, then PU P(K) ⊆ ISPPU (K). 

33
3. U NIVERSAL ALGEBRA

Example 3.29 (Quasi-varieties). In view of Examples 3.19 and 3.20, we know that

Latt = ISPU (Latt<ω ) and HA = ISPU (HA<ω ).

This implies that Latt ⊆ Q(Latt<ω ) and HA ⊆ Q(HA<ω ). Since both Latt and HA are
closed under I, S, P and PU , this yields

Latt = Q(Latt<ω ) and HA = Q(HA<ω ).

Let DL be the class of distributive lattices and B the two-element distributive lattice.
In view of Example 1.23, we have DL = IPSD ( B). Clearly, IPSD ( B) ⊆ ISP( B) ⊆ Q( B). On
the other hand, since DL is closed under I, S, P and PU , we obtain Q( B) ⊆ DL. Hence,
DL = Q( B). On the other hand, U( B) = ISPU ( B) = I( B). Similarly, the class of Boolean
algebras is the quasi-variety generated by the two-element Boolean algebra (see 1.24, if
necessary). 

Exercise 3.30. Prove that there is no finite Heyting algebra A such that HA = Q( A), cf.
with HA = Q(HA<ω ). Prove that, however, there exists an infinite Heyting algebra A such
that HA = Q( A) [83, 131], see also [50, 106]. 
The next observation will be needed later on.

Theorem 3.31. If K is a quasi-variety, the consequence relation K is finitary.

Proof. In view of Maltsev’s Theorem, K can be axiomatized by a set Σ of quasi-equations.


Formally speaking, this means that K is the class of models of the set of sentences

Σ∀ := {∀~x ϕ : ϕ(~x ) ∈ Σ}.

Let then Θ ∪ { ϕ ≈ ψ} ⊆ E(Var ) be such that Θ K ϕ ≈ ψ. Moreover, let { an : n ∈ N}


be a set of new constants. Given a formula ϕ( x1 , . . . , xn ), we write

ϕ(~a) as a shorthand for ϕ( a1 , . . . , an ).

Since Σ∀ axiomatizes K, the fact that Θ K ϕ ≈ ψ is equivalent to the demand that

Σ∀ ∪ {ε(~a) ≈ δ(~a) : ε ≈ δ ∈ Θ} ` ϕ(~a) ≈ ψ(~a),

where ` is the derivability symbol of classical first order logic. Consequently, by the
Compactness Theorem, there exists a finite ∆ ⊆ Θ such that

Σ∀ ∪ {ε(~a) ≈ δ(~a) : ε ≈ δ ∈ ∆} ` ϕ(~a) ≈ ψ(~a).

Since Σ∀ axiomatizes K, this means that ∆ K ϕ ≈ ψ, as desired. 

Observe that, for every class K of similar algebras, I(K) ∪ PU (K) ⊆ HP(K). Because of
this, the following classes of algebras [12, 30] are also quasi-varieties.

Definition 3.32. A class of similar algebras closed under H, S and P is said to be a variety.

34
3.3. Quasi-varieties

Notably, varieties coincide with equational classes of algebras [12, Thm. 4.41].

Birkhoff’s Theorem 3.33. A class of similar algebras is a variety if and only if it can be axioma-
tized by a set of equations.

Given a class of similar algebras K, the least variety extending K exists. It will be
denoted by V(K) and called the variety generated by K.

Corollary 3.34. Let K be a class of similar algebras. Then V(K) = HSP(K).

As a consequence, examples of varieties include the classes of (distributive) lattices,


Heyting, Boolean and modal algebras. While every variety is a quasi-variety, the converse
is not true in general, as we proceed to explain.

Example 3.35 (Lattices). Consider the lattices A and B depicted below. We will show that
the quasi-variety Q( B) is not closed under H and, therefore, is not a variety.

a8

a5 a6 a7

a2 a3 a4

a1
A B

To this end, let D + ( A) be the positive atomic diagram of A written with the variables
x1 , . . . , x8 corresponding to the elements a1 , . . . , a8 and consider the quasi-equation

Φ= & D+ ( A) =⇒ x1 ≈ x8.
Notice that B validates Φ. To prove this, consider an assignment f : { x1 , . . . , x8 } → B
that validates D + ( A) in B. Using the definition of D + ( A), it is easy to see that the
map h : A → B that sends ai to f ( ai ) is a homomorphism from A to B. Since Con( A) =
{id A , A × A}, the kernel Ker(h) is either id A or A × A. Notice that there is no embedding of
A into B. Therefore, Ker(h) cannot be the identity relation. It follows that Ker(h) = A × A.
In particular, h a1 , a8 i ∈ Ker(h) and, therefore, f ( x1 ) = h( a1 ) = f ( a8 ) = f ( x8 ). Hence, we
conclude that B  Φ, as desired. Moreover, Φ fails in A, as witnessed by the assignment

x1 7−→ a1 , . . . , x8 7−→ a8 .

In brief, Φ holds in B but fails in A. By Maltsev’s Theorem, we conclude that A


does not belong to the quasi-variety Q( B) generated by B. On the other hand, A is a
homomorphic image of B (obtained by glueing two pairs of elements of B). Thus, the
quasi-variety Q( B) is not closed under H. 

35
3. U NIVERSAL ALGEBRA

Example 3.36 (Heyting algebras). We assume the reader is familiar with Esakia duality for
Heyting algebras [53, 54]. Let A be the finite Heyting algebra whose dual Esakia space is
the following poset A∗ endowed with the discrete topology.

• • •
• •

Or aim is to show that Q( A) is not closed under H and, therefore, fails to be a variety.
To this end, consider the finite Heyting algebra B whose dual Esakia space is the the
rooted poset B∗ (endowed with the discrete topology) depicted below.

• • •

Notice that, as B∗ is rooted and finite, the algebra B is subdirectly irreducible. Moreover,
observe that, as B∗ is an upset of A∗ , by Esakia duality we obtain B ∈ H( A). Therefore, it
suffices to show that B ∈/ Q( A). Suppose the contrary, with a view to contradiction. By
Corollary 3.27,
B ∈ ISP( A) ⊆ IPSD S( A).
As B is subdirectly irreducible, we conclude that B ∈ IS( A). By Esakia duality, this means
that B∗ is a p-morphic image of A∗ . But a quick inspection of the posets A∗ and B∗ shows
that this is impossible, a contradiction. Hence, we conclude that Q( A) does not contain B
and, therefore, is not closed under H. 

Exercise 3.37. Let A1 , . . . , A5 be the Heyting algebras whose Esakia duals are the five posets
depicted below.


• • • • • •
• • • • • • • • • • • • •
• • • • • • • • •
A variety K is said to be primitive whenever every quasi-variety M ⊆ K is a variety. Prove
that if a variety K of Heyting algebras is primitive, then it omits A1 , . . . , A5 .
Notably, the converse is also true: a variety K of Heyting algebras is primitive if and
only if it omits A1 , . . . , A5 [36, 37], see also [13]. A similar description of primitive varieties
of K4-algebras has been obtain in [122]. These results admit a logical interpretation in
terms of structural completeness [11, 118, 123]. 
If K is a quasi-variety and θ a congruence of some A ∈ K, the algebra A/θ need not
belong to K. This makes the following concept attractive.

Definition 3.38. Let K ∪ { A} be a class of similar algebras. A congruence θ ∈ Con( A) is


said to be a K-congruence of A if A/θ ∈ K. We denote the poset of K-congruences of A,
ordered under the inclusion relation, by ConK ( A).

36
3.3. Quasi-varieties

Proposition 3.39. If K is a quasi-variety, then ConK ( A) is a complete lattice in which meets are
intersections, for every algebra A of the type of K.

Proof. Then let A be an algebra of the type of K. Since K is a quasi-variety, it is closed


under PSD and it contains a trivial algebra (the subdirect product of the empty family).
Therefore, as K is closed under I by assumption, it contains all trivial algebras and, in
particular, A/( A × A). Thus, A × A ∈ ConK ( A). Then consider a nonempty family
{θi : i ∈ I } ⊆ ConK ( A). By Proposition 1.25,
\
A/ θi ∈ IPSD ({ A/θi : i ∈ I }).
i∈ I

Observe that { A/θi : i ∈ I } ⊆ K, since the various θi are K-congruences of A. Together


with the above display and the assumption that K is closed under I and PSD , this yields
A/ i∈ I θi ∈ K, whence i∈ I θi ∈ ConK ( A). It follows that ConK ( A) has a minimum,
T T

namely A × A and infima of nonempty families. Therefore, arbitrary infima exist in


ConK ( A) and, therefore, ConK ( A) is a complete lattice. 
Corollary 3.40. If K is a quasi-variety and A an algebra of the same type, the map

CgKA : P ( A × A) → P ( A × A)

that associates the least K-congruence of A containing a subset X with a subset X ⊆ A × A is a


closure operator on A × A.

Remark 3.41. The proof of Proposition 3.39 depends only on the fact that K is closed
under I, S and P. Classes of similar algebras closed under I, S and P are called prevarieties.
Notably, they coincide with the classes of algebras that can be axiomatized by proper classes
of infinitary quasi-equations. The demand that proper classes can be replaced by sets in
the axiomatization of prevarieties is equivalent to an independent set theoretical principle
known as Vopěnka Principle [68, Prop. 2.3.18], see also [1]. 

37
CHAPTER 4
Algebraizable logics

4.1 Algebraizability
We are now ready to introduce a robust theory of algebraization that will account for the
relation between logic and algebra [21], see also [23, 42, 55, 56, 57, 73, 74]. To this end, for
every set of formulas ∆( x, y) and set of equations Θ ∪ {ε ≈ δ}, we shall abbreviate
[
{ ϕ(ε, δ) : ϕ( x, y) ∈ ∆} as ∆(ε, δ), and ∆( ϕ, ψ) as ∆[Θ].
ϕ≈ψ∈Θ

Definition 4.1 (Blok & Pigozzi). A finitary logic ` is said to be algebraizable if there are a
finite set of equations τ ( x ), a finite set of formulas ∆( x, y), and a quasi-variety K such that

Γ ` ϕ ⇐⇒ τ [ Γ ] K τ ( ϕ) (Alg1)
Θ K ε ≈ δ ⇐⇒ ∆[Θ] ` ∆(ε, δ) (Alg2)
ϕ a ` ∆[τ ( ϕ)] (Alg3)
ε ≈ δ =||=K τ [∆(ε, δ)] (Alg4)

for every set of formulas Γ ∪ { ϕ} and every set of equations Θ ∪ {ε ≈ δ}. In this case, K is
said to be an equivalent algebraic semantics for `. In addition, we say that τ, ∆ and K witness
the algebraizability of the logic `.
Condition (Alg1) expresses the demand that K is a τ-algebraic semantics for `, namely,
that ` can be interpreted into K by means of the set of equations τ ( x ) that allows to
translate a set of formulas Γ into a set of equations τ [ Γ ]. Condition (Alg2) states that
this interpretation can be reversed, in the sense that K can also be interpreted into `
by means of the set of formulas ∆( x, y) that allows to translate sets of equations Θ into
sets of formulas ∆[Θ]. Lastly, conditions (Alg3) and (Alg4) guarantee that these two
interpretations are inverses of each other up to provability equivalence.
Remark 4.2. While the problem of determining whether logics presented by Hilbert calculi
are algebraizable is undecidable [100], the same problem becomes decidable, although

39
4. A LGEBRAIZABLE LOGICS

complete for EXPTIME [103], for logics presented by finite sets of finite logical matrices in
the sense of [129]. 
The definition of an algebraizable logic can be made more concise as follows.

Proposition 4.3. The following conditions are equivalent for a finitary logic `:

(i) ` is algebraizable;

(ii) There are a set of equations τ ( x ), a set of formulas ∆( x, y) and a quasi-variety K that satisfy
(Alg1) and
x ≈ y =||=K τ [∆( x, y)]; (Alg4*)

(iii) There are a set of equations τ ( x ), a set of formulas ∆( x, y) and a quasi-variety K that satisfy
(Alg2) and
x a ` ∆[τ ( x )]. (Alg3*)

In this case, τ, ∆, and K witness the algebraizability of `.

Proof. The implication (i)⇒(ii) is straightforward. To prove (ii)⇒(iii), observe that (Alg4*)
implies that, for every ϕ ≈ ψ ∈ τ,

ϕ ≈ ψ =||=K τ [∆( ϕ, ψ)].

{τ [∆( ϕ, ψ)] : ϕ ≈ ψ ∈ τ }, this yields


S
Since τ [∆[τ ( x )]] =

τ ( x ) =||=K τ [∆[τ ( x )]].

By (Alg1), we conclude that x a ` ∆[τ ( x )]. Then consider a set of equations Θ ∪ {ε ≈ δ}.
We have
Θ K ε ≈ δ ⇐⇒ τ [∆[Θ]] K τ [∆(ε, δ)] ⇐⇒ ∆[Θ] ` ∆(ε, δ),
where the first equivalence follows from (Alg4*) and the second from (Alg1).
(iii)⇒(i): Since ` is finitary and ∆[τ ( x )] ` x, there exists a finite τ 0 ⊆ τ such that
∆[τ ( x )] ` x. Together with the assumption that x ` ∆[τ ( x )], this yields x a ` ∆[τ 0 ( x )].
Therefore, the pair τ 0 and ∆ satisfy conditions (Alg2) and (Alg3*). Furthermore, a proof
similar to the one of the implication (ii)⇒(iii) establishes (Alg1) and (Alg4*) for τ 0 and ∆.
Now, by (Alg4*), we have x ≈ y =||=K τ 0 [∆( x, y)]. Since, by Theorem 3.31, the relation
K is finitary, there exists a finite ∆0 ⊆ ∆ such that x ≈ y =||=K τ 0 [∆0 ( x, y)]. Therefore,
the pair τ 0 and ∆0 satisfy conditions (Alg1) and (Alg4*). Furthermore, a proof similar
to the one of the implication (ii)⇒(iii) establishes (Alg2) and (Alg3*) for τ 0 and ∆0 . As a
consequence, the pair of finite sets τ 0 and ∆0 satisfy (Alg1), (Alg2), (Alg3*) and (Alg4*).
Since conditions (Alg3*) and (Alg4*) imply (Alg3) and (Alg4), respectively, we conclude
that ` is algebraizable and that its algebraizability is witnessed by τ, ∆, and K. 

Example 4.4 (Algebraizable logics). Consider the following sets of equations and formulas,
respectively,
τ ( x ) := { x ≈ 1} and ∆( x, y) := { x → y, y → x }.

40
4.1. Algebraizability

We shall prove that IPC is algebraizable and that its algebraizability is witnessed by τ, ∆
and the variety HA of Heyting algebras. First, recall that HA is a τ-algebraic semantics for
IPC. Moreover, observe that

x ≈ y =||=HA { x → y ≈ 1, y → x ≈ 1}. (4.1)

To prove this, consider a Heyting algebra A and a, c ∈ A. Using the residuation law and
the fact that 1 is the maximum of A, we obtain

a = c ⇐⇒ ( a 6 c and c 6 a)
⇐⇒ (1 ∧ a 6 and 1 ∧ c 6 a)
⇐⇒ (1 6 a → c and 1 6 c → a)
⇐⇒ a → c = 1 and c → a = 1.

This establishes (4.1), which is precisely (Alg4*). Hence, IPC satisfies (Alg1) and (Alg4*)
with respect to τ, ∆ and HA. By Proposition 4.3, we conclude that IPC is algebraizable and
that its algebraizability is witnessed by τ, ∆ and HA.
A similar argument shows that Kg is algebraizable and that its algebraizability is
witnessed by τ, ∆ and the class of modal algebras MA. Similarly, CPC is algebraizable and
its algebraizability is witnessed by τ, ∆ and the class of Boolean algebras BA. 
Example 4.5 (A nonalgebraizable logic). On the other hand, not every logic is algebraizable.
As an exemplification, we will prove that K` is not algebraizable.
Suppose the contrary, with a view to contradition. Then there are a set of equations
τ ( x ), a set of formulas ∆( x, y) and a quasi-variety K witnessing the algebraizbaility of K` .
We shall see that K is a class of modal algebras. As the class MA of modal algebras is a
variety, it suffices to show that every equation valid in MA is also valid in K. To this end,
let ε ≈ δ be an equation valid in MA. From (Alg1) and (Alg4) it follows

∅ `K` ∆( x, x ) ⇐⇒ ∅ K τ [∆( x, x )] ⇐⇒ ∅ K x ≈ x.

Since ∅ K x ≈ x, we conclude that ∅ `K` ∆( x, x ). By substitution invariance, this yields

∅ `K` ∆(ε, ε). (4.2)

Since ε ≈ δ is valid in MA, for every ϕ( x, y) ∈ ∆, we have

MA  ϕ(ε, ε) ≈ ϕ(ε, δ).

Together with (2.4), this yields ∆(ε, ε) a `K` ∆(ε, δ). Since, by (4.2), ∅ `K` ∆(ε, ε), this
implies ∅ `K` ∆(ε, δ). By (Alg1), we obtain ∅ K τ [∆(ε, δ)]. Using (Alg4), we conclude
that K satisfies ε ≈ δ and, therefore, that K is a class of modal algebras.
Since K` is related to K by (Alg4), this implies that K` has an algebraic semantics,
namely K, that is a class of modal algebras. But this contradicts Theorem 2.22. 
As we mentioned, the theory of algebraizable logics allows to associate with a logic a
unique distinguished algebraic semantics.

41
4. A LGEBRAIZABLE LOGICS

Theorem 4.6. If τ1 , ∆ 1 , K1 and τ2 , ∆ 2 , K2 witness the algebraizability of a logic `, then


K1 = K2 τ1 ( x ) =||=K1 τ2 ( x ) ∆ 1 ( x, y) a ` ∆ 2 ( x, y).
Proof. We begin by proving ∆ 1 ( x, y) a ` ∆ 2 ( x, y). By symmetry, it suffices to prove
∆ 1 ( x, y) ` ∆ 2 ( x, y). To this end, consider ϕ ∈ ∆ 2 . Since τ1 , ∆ 1 , K1 witness the alge-
braizability of `, from (Alg4) and (Alg1) it follows
τ1 ( x ), x ≈ y K1 τ1 (y) ⇐⇒ τ1 ( x ), τ1 [∆ 1 ( x, y)] K1 τ1 (y) ⇐⇒ x, ∆ 1 ( x, y) ` y
and
x ≈ y K1 ϕ( x, x ) ≈ ϕ( x, y) ⇐⇒ ∆ 1 ( x, y) ` ∆ 1 ( ϕ( x, x ), ϕ( x, y)).
Since τ1 ( x ), x ≈ y K1 τ1 (y) and x ≈ y K1 ϕ( x, x ) ≈ ϕ( x, y), we obtain that
x, ∆ 1 ( x, y) ` y and ∆ 1 ( x, y) ` ∆ 1 ( ϕ( x, x ), ϕ( x, y)).
Moreover, by substitution invariance, we get ϕ( x, x ), ∆ 1 ( ϕ( x, x ), ϕ( x, y)) ` ϕ( x, y). To-
gether with the above display, we conclude that
ϕ( x, x ), ∆ 1 ( x, y) ` ϕ( x, y).
Now, as explained in Example 4.5, the fact that τ2 , ∆ 2 , K2 witness the algebraizability
of ` guarantees that ∅ ` ∆ 2 ( x, x ). As a consequence, ∅ ` ϕ( x, x ). Together with the
above display, this yields ∆ 1 ( x, y) ` ϕ( x, y) and, therefore, ∆ 1 ( x, y) ` ∆ 2 ( x, y). Hence, we
conclude that ∆ 1 ( x, y) a ` ∆ 2 ( x, y), as desired.
Then we turn to prove that K1 = K2 . Since K1 and K2 are quasi-varieties, it suffices
to show that they satisfy the same quasi-equations with variables in Var. To prove this,
consider a finite set of equation Θ ∪ {ε ≈ δ} ⊆ E(Var ). We have
K1  & Θ =⇒ ε ≈ δ ⇐⇒ Θ K 1
ε≈δ
⇐⇒ ∆ 1 [Θ] ` ∆ 1 (ε, δ)
⇐⇒ ∆ 2 [Θ] ` ∆ 2 (ε, δ)
⇐⇒ Θ K2 ε ≈ δ
⇐⇒ K2  & Θ =⇒ ε ≈ δ.
The above equivalence can be justified as follows. The first and the last are straightforward,
the second from (Alg2) and the assumption that τ1 , ∆ 1 , K1 witness the algebraizability of
`, the third from ∆ 1 ( x, y) a ` ∆ 2 ( x, y) and the fourth from the assumption that τ2 , ∆ 2 , K2
witness the algebraizability of `. Hence, we conclude that K1 = K2 .
It only remains to prove that τ1 ( x ) =||=K1 τ2 ( x ). To prove this, observe that
τ1 ( x ) =||=K1 τ2 ( x ) ⇐⇒ ∆ 1 [τ1 ( x )] a ` ∆ 1 [τ2 ( x )]
⇐⇒ ∆ 1 [τ1 ( x )] a ` ∆ 2 [τ2 ( x )]
⇐⇒ x a ` x.
The above equivalence can be justified as follows. The first follows from (Alg2) and the
assumption that τ1 , ∆ 1 , K1 witness the algebraizability of `, the third from ∆ 1 ( x, y) a `
∆ 2 ( x, y) and the last from (Alg3) and the fact that both τ1 , ∆ 1 , K1 and τ2 , ∆ 2 , K2 witness
the algebraizability of `. Since x a ` x, we conclude that τ1 ( x ) =||=K1 τ2 ( x ). 

42
4.1. Algebraizability

Corollary 4.7. Every algebraizable logic has a unique equivalent algebraic semantics.

Proof. Immediate from Theorem 4.6. 


Example 4.8 (Equivalent algebraic semantics). In view of Example 4.5,

(i) The unique equivalent algebraic semantics of CPC is BA;

(ii) The unique equivalent algebraic semantics of IPC is HA;

(iii) The unique equivalent algebraic semantics of Kg is MA.

In particular, while CPC has various algebraic semantics (for instance, BA and HA), the
class of Boolean algebras is its unique equivalent algebraic semantics. 
At this stage, it is natural to wonder whether every quasi-variety is the equivalent
algebraic semantics of some algebraizable logic. While this is not the case in general
[58] (as explained in the next exercise), still every quasi-variety is categorically equivalent
to the equivalent algebraic semantics of an algebraizable logic [105], see also [27]. As
a consequence, the categorical properties of every every quasi-variety can be studied
though the eyes of an algebraizable logic. Notably, category equivalences and adjunctions
between quasi-varieties have been described in detail in [49, 95, 101], see also [59].
Exercise 4.9. A class of algebras K that satisfies f ( x, . . . , x ) ≈ x for each of its basic opera-
tions f is said to be idempotent. For instance, all classes of lattices are idempotent. Prove
that a nontrivial idempotent quasi-variety cannot be the equivalent algebraic semantics of
any algebraizable logic.
To this end, you might wish to use the following strategy: suppose, with a view to
contradiction, that there exists a nontrivial idempotent quasi-variety K that, moreover, is
the equivalent algebraic semantics of some algebraizable logic `. First show that ∅ ` x.
By substitution invariance, derive ∅ ` ∆( x, y). Use this fact to infer that K is trivial and,
therefore, to obtain a contradiction. 
Definition 4.10. Given two logics ` and `∗ of the same type. Then `∗ is said to be

(i) an extension of ` if, Γ `∗ ϕ, for every Γ ∪ { ϕ} ⊆ T (Var ) such that Γ ` ϕ; and

(ii) an axiomatic extension of ` if there exists a set of formulas Σ such that σ[Σ] ⊆ Σ for
every substitution σ and, for every Γ ∪ { ϕ} ⊆ T (Var ),

Γ `∗ ϕ ⇐⇒ Γ, Σ ` ϕ.

In this case, we say that Σ axiomatizes `∗ relative to `.

Axiomatic extensions of IPC have been called superintuitionistic logics in the literature.
Remark 4.11. Notice that when a logic `∗ is an axiomatic extension of a logic `, axiomatized
relative to ` by Σ, it can be axiomatized by extending any Hilbert axiomatization for `
with the set of axioms {∅  ϕ : ϕ ∈ Σ}. 
Definition 4.12. Let K and M be quasi-varieties of the same type. Then M is said to be

43
4. A LGEBRAIZABLE LOGICS

(i) a subquasi-variety of K if M ⊆ K; and

(ii) a relative subvariety of K if it is axiomatized relative to K by a set of equations.

Notice that the relative subvarieties of a variety are precisely its subvarieties.

Given a finitary logic `, the posets fEx(`) and aEx(`) of finitary and axiomatic exten-
sions of `, respectively, ordered under the inclusion relation form two complete lattices.
Similarly, given a quasi-variety K, the posets sQ(K) and sV(K) of subquasi-varieties and
relative subvarieties of K, respectively, ordered under the inclusion relation form two com-
plete lattices. In view of the next result, axiomatic extensions of an algebraizable logic can
be studied through the lenses of the relative subvarieties of its equivalent algebraic seman-
tics, which, in turn, are amenable to the methods of model theory, universal algebra and
duality theory. This approach has proved very fruitful in the study of superintuitionistic
and normal modal logics [33, 89].

Theorem 4.13. Let ` be an algebraizable logic and K its equivalent algebraic semantics. The
lattices fEx(`) and sQ(K) are dually isomorphic under the map that sends a finitary extension of
` to its equivalent algebraic semantics. This dual isomorphism restricts to one between aEx(`)
and sV(K).

Remark 4.14. Notice that the above result implicitly states that algebraizability persists
under the formation of finitary extensions. Moreover, it generalizes the well-known
fact that the lattices of superintuitionistic logics and of normal modal logics are dually
isomorphic to those of varieties of Heyting and modal algebras, respectively. 
Remark 4.15. The notion of an algebraizable logic can be extended to other formalisms, in-
cluding Gentzen systems and hypersequent calculi [28, 66, 112, 117, 120] and has inspired
a notion of equivalence between arbitrary deductive systems [15, 16, 61, 64]. 

4.2 Deductive filters


Deductive closed sets of formulas have been called theories. This notion can be extended
as follows.

Definition 4.16. A subset F of the universe of an algebra A is said to be a deductive filter of


a logic ` on A when, for every Γ ∪ { ϕ} ⊆ T (Var ),

if Γ ` ϕ, then for every homomorphism f : T (Var ) → A,


if f [ Γ ] ⊆ F, then f ( ϕ) ∈ F.

We denote by Fi` ( A) the poset of deductive filters of ` on A ordered under inclusion.

The following observations is a direct consequence of the definition of a deductive


filter.

Proposition 4.17. If ` is a logic and A an algebra, then Fi` ( A) is a complete lattice in which
meets are intersections.

44
4.2. Deductive filters

Recall that the set of theories of a logic ` is denoted by T h(`). When convenient, we
regard T h(`) as a lattice ordered under inclusion.

Proposition 4.18. If ` is a logic, the lattice T h(`) of theories of ` coincides with the lattice
Fi` ( T (Var )) of filters of ` on the formula algebra T (Var ).

Proof. Consider a set of formulas Γ. Suppose first that Γ is a theory of `. Then consider a
set of formulas Σ ∪ { ϕ} such that Σ ` ϕ and a homomorphism σ : T (Var ) → T (Var ) such
that σ[Σ] ⊆ Γ. Notice that σ is a substitution. Consequently, as ` is substitution invariant,
σ[Σ] ` σ( ϕ). Now, since σ[Σ] ⊆ Γ, we have Γ ` σ[Σ]. Together with σ[Σ] ` σ( ϕ) and the
Cut principle, this implies Γ ` σ( ϕ). Since Γ is a theory of `, this yields σ( ϕ) ∈ Γ. Hence,
we conclude that Γ is a filter of ` on T (Var ).
Conversely, suppose that Γ is a filter of ` on T (Var ). Then consider a formula ϕ such
that Γ ` ϕ and let id : T (Var ) → T (Var ) be the identity homomorphism. Since id[ Γ ] = Γ
and Γ ` ϕ, the assumption that Γ is a filter of ` on T (Var ) guarantees that ϕ = id( ϕ) ∈ Γ.
Hence, we conclude that Γ is a theory of `. 
The next observation is instrumental to describe deductive filters in concrete cases.

Proposition 4.19. Let ` be the logic axiomatized by an Hilbert calculus H. A subset F of the
universe of an algebra A is a deductive filter of ` on A if and only if F is closed under the
interpretation of the rules in H, that is, for every rule Γ  ϕ in H and every homomorphism
f : T (Var ) → A,
if f [ Γ ] ⊆ F, then f ( ϕ) ∈ F.

Proof. The “only if” part is straightforward. To prove the “if” one, suppose that F is closed
under the interpretation of the rules in H. Then consider a set of formulas Γ ∪ { ϕ} such
that Γ ` ϕ and a homomorphism f : T (Var ) → A such that f [ Γ ] ⊆ F. Since H axiomatizes
`, there exists a formal proof hψα : α 6 γi of ϕ from Γ in H. We will prove, by complete
induction on α, that f (ψα ) ∈ F. Accordingly, assume that { f (ψβ ) : β < α} ⊆ F. Then we
have two cases: either ψα ∈ Γ or there exists a rule Σ  δ in H and a substitution σ such
that σ[∆] ⊆ {ψβ : β < α} and σ(δ) = ψα . If ψα ∈ Γ, then it is clear that f (ψα ) ∈ f [ Γ ] ⊆ F.
Then we consider the second case. By the inductive hypothesis, we have

f (σ[∆]) ⊆ { f (ψβ ) : β < α} ⊆ F.

Furthermore, consider the homomorphism f ◦ σ : T (Var ) → A. From the above display it


follows f ◦ σ[∆] ⊆ F. Since F is closed under the interpretation of the rules in H (and, in
particular, under ∆  δ), this yields

f (ψα ) = f (σ(δ)) = f ◦ σ(δ) ∈ F.

This concludes the inductive proof. Since ϕ = ψγ , we obtain that f ( ϕ) ∈ F and, therefore,
that F is a deductive filter of ` on A. 
Example 4.20 (Deductive filters). We will prove that

(i) The deductive filters of IPC on a Heyting algebra A are the lattice filters of A;

45
4. A LGEBRAIZABLE LOGICS

(ii) The deductive filters of CPC on a Boolean algebra A are the lattice filters of A;

(iii) The deductive filters of Kg on a modal algebra A are the open filters of A.

(i): Consider a subset F of A. Suppose first that F is a deductive filter of IPC. As ∅ `IPC 1
and F is a deductive filter of IPC on A, we obtain 1 ∈ F. Then consider a, c ∈ A. To prove
that F is an upset, suppose that a ∈ F and a 6 c. From the residuation law it follows

a 6 c =⇒ 1 ∧ a 6 c =⇒ 1 6 a → c =⇒ a → c = 1.

Since 1 ∈ F, this yields a, a → c ∈ F. Together with x, x → y `IPC y and the assumption


that F is a deductive filter of IPC on A, this yields c ∈ F, as desired. Lastly, to prove that F
is closed under binary meets, suppose that a, c ∈ F. Since F is a deductive filter of IPC on
A, the fact that x, y `IPC x ∧ y and a, c ∈ F implies a ∧ c ∈ F. Hence, we conclude that F is
a lattice filter of A.
Conversely, suppose that F is a lattice filter of A. In view of Proposition 4.19, it
suffices to show that F is closed under the interpretation of the rules of an Hilbert calculus
axiomatizing IPC. Accordingly, let H be the Hilbert calculus whose set of axioms is

{∅  ϕ : ϕ ∈ T (Var ) and ∅ `IPC ϕ}

and whose sole rule is modus ponens x, x → y  y. As H axiomatizes IPC, it only remains
to prove that F is closed under the interpretation of its rules. Let then ∅  ϕ( x1 , . . . , xn )
be an axiom of H. Clearly, ∅ `IPC ϕ( x1 , . . . , xn ). Since the class of Heyting algebras is an
{ x ≈ 1}-algebraic semantics for IPC, it follows that, for all a1 , . . . , an ∈ A,

ϕ A ( a1 , . . . , an ) = 1 ∈ F.

The only rule in H is modus ponens x, x → y  y. To prove that F is closed under its
interpretation, consider a, c ∈ A such that a, a → c ∈ F. Since F is closed under binary
meets, a ∧ ( a → c) ∈ F. Moreover, by the residuation law,

a → c 6 a → c ⇐⇒ a ∧ ( a → c) 6 c.

Since a → c 6 a → c always holds, we get a ∧ ( a → c) 6 c. As F is an upset that contains


a ∧ ( a → c), we obtain that c ∈ F. This shows that F is a deductive filter of IPC on A.
(ii): Analogous to the proof of (i).
(iii): Consider a subset F of A. Suppose first that F is a deductive filter of Kg on A. The
proof that F is a lattice filter of A is analogous to the one detailed in the case of (i). In order
to prove that F is also closed under , consider a ∈ F. Since x `Kg x, the fact that F is a
deductive filter of Kg on A and a ∈ F implies that a ∈ F. Hence, we conclude that F is
an open filter of A, as desired.
To prove the converse, suppose that F is an open filter of A. In view of Proposition
4.19, it suffices to show that F is closed under the interpretation of the rules in the Hilbert
calculus axiomatizing Kg . Accordingly, let H be the Hilbert calculus whose set of axioms is

{∅  ϕ : ϕ ∈ T (Var ) and ∅ `Kg ϕ}

46
4.2. Deductive filters

and whose rules are modus ponens x, x → y  y and necessitation x  x. As H axioma-
tizes Kg , it only remains to show that F is closed under the interpretation of its rules. The
proof that F is closed under the interpretation of the axioms in H and of modus ponens
is analogous to the one detailed for the case of (i). Therefore, it only remains to prove
that F is closed under the interpretation of the necessitation rule x  x. But this is an
immediate consequence of the fact that the filter F is open. 
Exercise 4.21. Prove that the deductive filters of K` on a modal algebra A are precisely the
lattice filters of A. Use the fact that K` can be axiomatized by the Hilbert calculus whose
set of axioms is
{∅  ϕ : ϕ ∈ T (Var ) and ∅ `K` ϕ}
and whose sole rule is modus ponens x, x → y  y. Hint: you may use the fact that K`
and Kg have the same theorems. 
The notion of a deductive filter can be extended to relative equational consequences as
follows.
Definition 4.22. Let K ∪ { A} be a class of similar algebras. A set θ ⊆ A × A is said to be a
deductive filter of K on A when, for every Θ ∪ {ε ≈ δ} ⊆ E(Var ),

if Θ K ϕ ≈ ψ, then for every homomorphism f : T (Var ) → A,


if h f ( ϕ), f (ψ)i ∈ θ for all ϕ ≈ ψ ∈ Θ, then h f (ε), f (δ)i ∈ θ.

Proposition 4.23. Let K be a quasi-variety and A an algebra of the same type. The deductive
filters of K on A are precisely the K-congruences of A.
Proof. Consider a subset θ of A × A. First suppose that θ is a deductive filters of K on A.
Notice that

∅ K x ≈ x x ≈ y K y ≈ x x ≈ y, y ≈ z K x ≈ z.

Furthermore, for every basic n-ary operation f , we have

x1 ≈ y1 , . . . , xn ≈ yn K f ( x1 , . . . , xn ) ≈ f (y1 , . . . , yn ).

Since θ is a deductive filter of K on A, the above displays guarantee that θ is a congruence


of A. To prove that it is also a K-congruence, it remains to show that A/θ ∈ K. Since K
is a quasi-variety, it suffices to prove that A satisfies all the quasi-equations valid in K.
Accordingly, consider a quasi-equation & Θ =⇒ ε ≈ δ valid in K and let f : T (Var ) → A/θ
be a homomorphism such that f ( ϕ) = f (ψ), for all ϕ ≈ ψ ∈ Θ. Since the canonical
homomorphism π : A → A/θ is surjective, we can apply Corollary 1.6, obtaining a
homomorphism g : T (Var ) → A such that f = π ◦ g. For every ϕ ≈ ψ ∈ Θ, we have

g( ϕ)/θ = π ◦ g( ϕ) = f ( ϕ) = f (ψ) = π ◦ g(ψ) = g(ψ)/θ

and, therefore, h g( ϕ), g(ψ)i ∈ θ. Since θ is a deductive filter of K and Θ K ε ≈ δ, this


yields h g(ε), g(δ)i ∈ θ. In turn, this implies

f ( ε ) = π ◦ g ( ε ) = π ◦ g ( δ ) = f ( δ ).

47
4. A LGEBRAIZABLE LOGICS

Hence, we conclude that A/θ ∈ K, as desired.


To prove the converse, suppose that θ is a K-congruence of A. Consider a set of
equations Θ ∪ {ε ≈ δ} ⊆ E(Var ) such that Θ K ε ≈ δ and a homomorphism f : T (Var ) →
A such that h f ( ϕ), f (ψ)i ∈ θ, for all ϕ ≈ ψ ∈ Θ. Now, let π : A → A/θ be the canonical
projection. We have π ◦ f ( ϕ) = π ◦ f (ψ), for all ϕ ≈ ψ ∈ Θ. Since A/θ ∈ K, this yields
π ◦ f (ε) = π ◦ f (δ), which is h f (ε), f (δ)i ∈ θ. 

Definition 4.24. Let K be a quasi-variety. A set of equations Θ ⊆ E(Var ) is said to be a


theory of K when, for every ε ≈ δ ∈ E(Var ),

if Θ K ε ≈ δ, then ε ≈ δ ∈ Θ.

When ordered under the inclusion relation, the theories of K form a lattice that we denote
by T h(K ).

Remark 4.25. Notice that K is a consequence relation on the set of equations E(Var ).
Therefore, the above definition is a special instance of Definition 2.4.
Recall that, formally speaking, equations are ordered pairs of formulas, e.g., the
expression ε ≈ δ is a suggestive notation for the ordered pair hε, δi. The following result
builds on this observation.

Proposition 4.26. If K is a quasi-variety, the lattice T h(K ) of theories of K coincides with the
lattice ConK ( T (Var )) of K-congruences of the formula algebra T (Var ).

Proof. An argument analogous to the one detailed in the proof of Proposition 4.18 shows
that T h(K ) coincides with the lattice of deductive filters of K on T (Var ). But, in view of
Proposition 4.23, the latter coincides with ConK ( T (Var )). 

Deductive filters are closed under inverse endomorphisms, as we proceed to explain.


First, the set of endomorphism of an algebra A will be denoted by End( A). Then, given an
endomorphism σ and a congruence θ of A, we set

σ−1 [θ ] := {h a, ci ∈ A × A : hσ( a), σ(c)i ∈ θ }.

Lemma 4.27. Let ` be a logic, K a quasi-variety, A an algebra and σ ∈ End( A).

(i) If F ∈ Fi` ( A), then σ−1 [ F ] ∈ Fi` ( A).

(ii) If θ ∈ ConK ( A), then σ−1 [θ ] ∈ ConK ( A).

Proof. (i): Suppose that F ∈ Fi` ( A). Then consider Γ ∪ { ϕ} ⊆ T (Var ) such that Γ ` ϕ and
a homomorphism f : T (Var ) → A such that f [ Γ ] ⊆ σ−1 [ F ]. Clearly, σ ◦ f [ Γ ] ⊆ F. Since
σ ◦ f : T (Var ) → A is a homomorphism and F ∈ Fi` ( A), this implies σ ◦ f ( ϕ) ⊆ F. Hence,
we conclude f ( ϕ) ⊆ σ−1 [ F ], as desired.
(ii): Recall from Proposition 4.23 that ConK ( A) is the set of deductive filters of K
on A. Because of this, we can mimic the proof detailed for condition (i) and obtain that
σ−1 [θ ] ∈ ConK ( A), for every θ ∈ ConK ( A). 

48
4.3. Isomorphism theorem

Remark 4.28. Conditions (i) and (ii) in the above lemma can be generalized as follows. Let
` be a logic, K a quasi-variety and f : A → B a homomorphism.
(i) If F is a deductive filter of ` on B, then f −1 [ F ] is a deductive filter of ` on A; and

(ii) If θ is a K-congruence of B, then f −1 [θ ] is a K-congruence of A. 


In view of Lemma 4.27, given a logic ` and an algebra A, the lattice Fi` ( A) of deductive
filters of ` on A can be expanded with the unary operations {σ−1 : σ ∈ End( A)}. Similarly,
given a quasi-variety K, the lattice ConK ( A) of K-congruences of A can also be expanded
with the unary operations {σ−1 : σ ∈ End( A)}. Accordingly, we set

Fi` ( A)+ := hFi` ( A); ∧, ∨, {σ−1 : σ ∈ End( A)}i


ConK ( A)+ := hConK ( A); ∧, ∨, {σ−1 : σ ∈ End( A)}i.

The above structures can be viewed as algebras whose type comprises two binary oper-
ations ∧ and ∨ an a family of unary operations {σ−1 : σ ∈ End( A)}. From this perspec-
tive, an isomorphism from Fi` ( A)+ to ConK ( A)+ is a lattice isomorphism Φ : Fi` ( A) →
ConK ( A) that commutes with inverse endomorphisms, in the sense that

Φ(σ−1 [ F ]) = σ−1 [Φ( F )], for every σ ∈ End( A).

Recall from Propositions 4.18 and 4.26 that

T h(`) = Fi` ( T (Var )) and T h(K ) = ConK ( T (Var )). (4.3)

Because of this, when A = T (Var ), we will denote Fi` ( A)+ and ConK ( A)+ by

T h(`)+ and T h(K )+ .

The importance of these structures will become apparent in the next section.

4.3 Isomorphism theorem


In many familiar examples the congruences of an algebra correspond to certain distin-
guished subsets of its universe. This happens, for instance, in the algebra of logic, where
the congruence lattice Con( A) of a Heyting algebra A is isomorphic to the lattice Fi( A)
of its filters. Similarly, the congruence lattice of a modal algebra A is isomorphic to the
the lattice Op( A) of its open filters (see Examples 1.12 and 1.13, if necessary). Analogous
correspondences can be found in classical algebra, where the congruences of groups and
ideals are related to normal subgroups and two-sided ideals in a similar manner. As we
shall see, all these correspondences can be viewed as consequences of the algebraization
of some logic.

Isomorphism Theorem 4.29 (Blok & Pigozzi). The following conditions are equivalent for a
finitary logic ` and a quasi-variety K:

(i) ` is algebraizable with equivalent algebraic semantics K;

49
4. A LGEBRAIZABLE LOGICS

(ii) Fi` ( A)+ ∼


= ConK ( A)+ , for every algebra A of the suitable type;
(iii) T h(`)+ ∼
= T h(K )+ .
In view of the implication (i)⇒(ii) in the Isomorphism Theorem, every algebraizable
logic induces an isomorphism between lattices of deductive filters and of K-congruences.
Most of the known correspondences between congruences and distinguished sets are
consequences of this fact, as we proceed to explain.

Corollary 4.30. The following conditions hold:

(i) If A is a Heyting algebra, Con( A) is isomorphic to the lattice of filters of A;

(ii) If A is a modal algebra, Con( A) is isomorphic to the lattice of open filters of A;

(iii) If A is a group, Con( A) is isomorphic to the lattice of normal subgroups of A;

(iv) If A is a ring, Con( A) is isomorphic to the lattice of two-sided ideals of A.

Proof. (i): Let A be a Heyting algebra. As detailed in Example 4.20, the deductive filters of
IPC on A are precisely the lattice filters of A, in symbols,

FiIPC ( A) = Fi( A).

Since IPC is algebraizable and its equivalent algebraic semantics is the class HA of Heyting
algebras, the implication (i)⇒(ii) in the Isomorphism Theorem guarantees that the lattices
FiIPC ( A) and ConHA ( A) are isomorphic. Furthermore, since A is a Heyting algebra and
HA is a variety, we have ConHA ( A) = Con( A), whence

Fi( A) = FiIPC ( A) ∼
= ConHA ( A) = Con( A).

Thus, we conclude that Fi( A) ∼ = Con( A), as desired.


The proof of condition (ii) is analogous to that of (i), because Kg is algebraizable with
equivalent algebraic semantics the variety of modal algebras and the deductive filters of
Kg on a modal algebra A are precisely the open filters of A (see Example 4.20).
It only remains to prove conditions (iii) and (iv). We will outline only the proof of (iii),
as the one of (iv) is analogous. Let Gr be the variety of groups in the type h·, (−)−1 , 1i. We
will show that Gr is the equivalent algebraic semantics of an algebraizable logic. To this
end, consider the sets

τ ( x ) := { x ≈ 1} and ∆( x, y) := { x · y−1 }

and observe that x ≈ y =||=Gr x · y−1 ≈ 1, that is,

x ≈ y =||=Gr τ [∆( x, y)]. (4.4)

The logic of groups G is defined, for every set of formulas Γ ∪ { ϕ}, as follows:

Γ `G ϕ ⇐⇒ τ [ Γ ] Gr τ ( ϕ).

50
4.3. Isomorphism theorem

This definition and condition (4.4) imply that G, τ, ∆ and G satisfy the conditions (Alg1)
and (Alg4*). By Proposition 4.3, we conclude that G is algebraizable with equivalent
algebraic semantics the variety of groups Gr.
Furthermore, the deductive filters of G on a group A are precisely the normal sub-
groups of A. On the one hand, every deductive filter of G on A is a normal subgroup,
because the following rules are valid in G:

x, y  x · y x  x −1 ∅1 x  y · ( x · y −1 ).

On the other hand, every deductive filter is a normal subgroup. For if F is a normal sub-
group of A, we have F = π −1 [{1}], where π : A → A/θ F the canonical homomorphism
and θ F the congruence of A induced by F. Now, the definition of G guarantees that {1} is
a deductive filter of G on A/θ F . In view of Remark 4.28, its inverse image π −1 [{1}] is a
deductive filter of G on A. Since F = π −1 [{1}], we are done.
Therefore, we can apply the implication (i)⇒(ii) in the Isomorphism Theorem, obtain-
ing that the lattice of normal subgroups of A is isomorphic to ConGr ( A). But, as Gr is a
variety, Con( A) = ConGr ( A). 

The implication (i)⇒(ii) in the Isomorphism Theorem is also instrumental to disprove


that certain logics are algebraizable.

Example 4.31. Recall that the logic K` is not algebraizable. We will present an alternative
proof of this fact, based on the Isomorphism Theorem. Suppose, with a view to contradic-
tion, that K` is algebraizable and let K be its equivalent algebraic semantics. Then consider
the modal algebra A = h{0, a, c, 1}; ∧, ∨, ¬, , 0, 1i such that

h{0, a, c, 1}; ∧, ∨, ¬, 0, 1i

is the four-element Boolean algebra with minimum 0 and maximum 1 and

0 = a = c = 0 and 1 = 1.

As the deductive filters of K` on a modal algebra are precisely the lattice filters (Exer-
cise 4.21), we know that FiK` ( A) is a four-element set. On the other hand, Con( A) =
{id A , A × A}. Consequently, ConK ( A) has cardinality at most two. On cardinality grounds,
it follows that the lattices FiK` ( A) and ConK ( A) cannot be isomorphic. As this contradicts
the Isomorphism Theorem, we conclude that K` is not algebraizable. 

The role of the inverse endomorphisms in the structures Fi` ( A)+ and ConK ( A)+ be-
comes apparent in the following example.

Example 4.32 (Implication free fragments). We will prove that also the h∧, ∨, ¬i-fragment
of IPC, in symbols IPC∧∨¬ , fails to be algebraizable. Suppose the contrary, with a view to
contradiction. Then consider the algebra A = h A; ∧, ∨, ¬i, where h A; ∧, ∨i is the lattice
with order 0 < c < a < 1 and ¬ is the unary operation defined as follows:

¬1 = ¬ a = ¬c = 0 and ¬0 = 1.

51
4. A LGEBRAIZABLE LOGICS

By inspection, we see that A has precisely five congruences, namely, id A , A × A and


θ1 := the congruence whose equivalence classes are {1, a}, {c}, {0};
θ2 := the congruence whose equivalence classes are {1}, { a, c}, {0};
θ3 := the congruence whose equivalence classes are {1, a, c}, {0}.
Therefore, Con( A) is the lattice depicted below.

A×A

θ3

θ1 θ2

id A

Then consider the endomorphism σ1 : A → A and σ2 : A → A such that


σ1 (1) = σ1 ( a) = 1 σ1 (c) = c σ1 (0) = 0
and
σ2 (1) = 1 σ2 ( a) = σ2 (c) = c σ2 (0) = 0
and observe that
θ1 = σ1−1 [θ2 ] and θ2 = σ2−1 [θ1 ]. (4.5)
We claim that if K is a quasi-variety and the lattice underlying ConK ( A)+
contains a
four-element chain, then ConK ( A) has cardinality five. To prove this, observe that, since
ConK ( A) is a subposet of Con( A) and Con( A) is the lattice depicted above, if ConK ( A)+
has a four-element chain, then
Con( A) r {θ1 } ⊆ ConK ( A) or Con( A) r {θ2 } ⊆ ConK ( A).
Since ConK ( A)+ is closed under the unary operations σ1−1 and σ2−1 , from (4.5) it follows
that θ1 , θ2 ∈ ConK ( A)+ . Together with the above display, this yields Con( A) = ConK ( A)
and, therefore, that ConK ( A) has precisely five elements.
Lastly, as A is the h∧, ∨, ¬i-reduct of a Heyting algebra, it is not hard to see that the
deductive filters of IPC∧∨¬ on A are precisely the lattice filters of A. Therefore, the lattice
of deductive filters of IPC∧∨¬ on A is a four-element chain. By the Isomorphism Theorem,
the lattice of K-congruences of A is also a four-element chain, where K is equivalent
algebraic semantics of IPC∧∨¬ . But, by the claim, this implies that ConK ( A) is a five
element set, a contradiction. 
Exercise 4.33. An ordered algebra is a pair h A; 6i such that A is an algebra and 6 a partial
order on A. The logic preserving degrees of truth `6 K associated with a class of ordered
algebras K is defined as follows: for every Γ ∪ { ϕ} ⊆ T (Var ),
Γ `6
K ϕ ⇐⇒ for every h A; 6i ∈ K, homomorphism f : T (Var ) → A and a ∈ A,
if a 6 f (γ) for all γ ∈ Γ, then a 6 f ( ϕ).

52
4.3. Isomorphism theorem

A bi-Heyting algebra [121, 130] is a structure A = h A; ∧, ∨, →, ←, 0, 1i such that both

h A; ∧, ∨, →, 0, 1i and h A; ∨, ∧, ←, 1, 0i

are Heyting algebras. Use the Isomorphism Theorem to show that the logic preserving
degrees of truth associated with the class of bi-Heyting algebras (endowed with the lattice
order) is not algebraizable. Hint: take inspiration from Example 4.31.
You might also wish to prove that IPC and K` are the logics preserving degrees of
truth associated, resepctively, with the classes of Heyting and modal algebras (endowed
with the lattice order). Hint: try to adapt the proof of Theorem 2.15. 
We shall now present a proof of the Isomorphism Theorem.

Proof. (i)⇒(ii): Let τ and ∆ be the sets of equations and formulas that, together with K,
witness the algebraizability of `. Moreover, let A be an algebra of the appropriate type.
For every F ∈ Fi` ( A) and θ ∈ ConK ( A), we define

Ω A F := {h a, ci ∈ A × A : ∆A ( a, c) ⊆ F }
S A (θ ) := { a ∈ A : h ϕ( a), ψ( a)i ∈ θ, for all ϕ ≈ ψ ∈ τ }.

We will show that the maps

Ω A : Fi` ( A)+ → ConK ( A)+ and S A : ConK ( A)+ → Fi` ( A)+

are well-defined isomorphisms, one inverse to the other.


We begin by proving that the map Ω A : Fi` ( A)+ → ConK ( A)+ is well defined. To this
end, consider F ∈ Fi` ( A). Observe that

∅ K x ≈ x
x ≈ y K y ≈ x
x ≈ y, y ≈ z K x ≈ z
x 1 ≈ y 1 , . . . , x n ≈ y n K f ( x 1 , . . . , x n ) ≈ f ( y 1 , . . . , y n ) ,

for every basic n-ary symbol. From condition (Alg2) in the definition of an algebraizable
logic, it follows that

∅ ` ∆( x, x )
∆( x, y) ` ∆(y, x )
∆( x, y) ∪ ∆(y, z) ` ∆( x, z)
∆( x1 , y1 , ) ∪ · · · ∪ ∆( xn , yn ) ` ∆( f ( x1 , . . . , xn ), f (y1 , . . . , yn )).

Since F is a deductive filter of `, these conditions guarantee that Ω A F is a congruence of


A. For instance, in order to prove that Ω A F is transitive, suppose that h a, bi, hb, ci ∈ Ω A F.
By the definition of Ω A F, we have ∆A ( a, b) ∪ ∆A (b, c) ⊆ F. As F is a deductive filter of
` on A, by the third condition in the above display we obtain ∆A ( a, c) ⊆ F. This in turn
yields h a, ci ∈ Ω A F, establishing transitivity.

53
4. A LGEBRAIZABLE LOGICS

As Ω A F is a congruence of A, it only remains to prove that A/Ω A F ∈ K. Since K


is a quasi-variety, Maltsev’s Theorem guarantees that it can be axiomatized by a set of
quasi-equations. Therefore, it suffices to show that every quasi-equation

( ϕ1 ≈ ψ1 , . . . , ϕn ≈ ψn ) =⇒ ε ≈ δ (4.6)

valid in K is also valid in A/Ω A F. To this end, consider a homomorphism f : T (Var ) →


A/Ω A F such that f ( ϕi ) = f (ψi ) for every i 6 n. By Corollary 1.6, there exists a homo-
morphism g : T (Var ) → A such that f = π ◦ g, where π : A → A/Ω A F is the canonical
projection. For every i 6 n, we have

π ◦ g( ϕi ) = f ( ϕi ) = f (ψi ) = π ◦ g(ψi )

and, therefore, h g( ϕi ), g(ψi )i ∈ Ω A F. By the definition of Ω A F, this amounts to

∆A ( g( ϕ1 ), g(ψ1 )) ∪ · · · ∪ ∆A ( g( ϕn ), g(ψn )) ⊆ F.

Since g is a homomorphism, this can be rewritten as

g[∆( ϕ1 , ψ1 ) ∪ · · · ∪ ∆( ϕn , ψn )] ⊆ F. (4.7)

Now, since the quasi-equation in (4.6) is valid in K, we have

ϕ1 ≈ ψ1 , . . . , ϕn ≈ ψn K ε ≈ δ.

By condition (Alg2) in the definition of an algebraizable logic, this implies

∆( ϕ1 , ψ1 ) ∪ · · · ∪ ∆( ϕn , ψn ) ` ∆(ε, δ).

Together with the assumption that F is a deductive filter of ` on A and (4.7), this yields

∆A ( g(ε), g(δ)) = g[∆(ε, δ)] ⊆ F.

By definition of Ω A F, this yields h g(ε), g(δ)i ∈ Ω A F. Hence, we conclude that

f (ε) = π ◦ g(ε) = g(ε)/Ω A F = g(δ)/Ω A F = π ◦ g(δ) = f (δ).

This shows that A/Ω A F satisfies the quasi-equation in (4.6) and, therefore, that A/Ω A F ∈
K. We conclude that the map Ω A : Fi` ( A)+ → ConK ( A)+ is well defined.
A similar argument shows that S A : ConK ( A)+ → Fi` ( A)+ is also well defined. To
conclude the proof, it only remains to show that

Ω A : Fi` ( A)+ → ConK ( A)+ and S A : ConK ( A)+ → Fi` ( A)+ (4.8)

are isomorphisms, one inverse to the other. To this end, consider F ∈ Fi` ( A). We have

S A (Ω A F ) = { a ∈ A : h ϕ( a), ψ( a)i ∈ Ω A F, for all ϕ ≈ ψ ∈ τ }


= { a ∈ A : ∆A ( ϕ( a), ψ( a)) ⊆ F, for all ϕ ≈ ψ ∈ τ }
= { a ∈ A : ∆A [τ A ( a)] ⊆ F }
= F.

54
4.3. Isomorphism theorem

The first and the third equalities above are straightforward, the second follows from the
definition of Ω A F, and the fourth follows from condition (Alg3) in the definition of an
algebraizable logic and the fact that F is a deductive filter of `. A similar argument shows
that θ = Ω A (S A (θ )), for every θ ∈ ConK ( A). Hence, we conclude that that the maps in
(4.8) are bijections, one inverse to the other.
It is straightforward to see that they are order preserving. Then consider F, G ∈ Fi` ( A)
such that Ω A F ⊆ Ω A G. Since S A is order preserving, we obtain S A (Ω A F ) ⊆ S A (Ω A G ).
As Ω A and S A are one inverse to the other, we conclude that

F = S A (Ω A F ) ⊆ S A (Ω A G ) = G.

It follows that Ω A : Fi` ( A)+ → ConK ( A)+ is also order reflecting and, therefore, an order
embedding. As it is surjective, we conclude that it is a lattice isomorphism. To prove
that it commutes with inverse endomorphism, consider F ∈ Fi` ( A), a pair a, c ∈ A and
σ ∈ End( A). We have

h a, ci ∈ σ−1 [Ω A F ] ⇐⇒ hσ( a), σ(c)i ∈ Ω A F


⇐⇒ ∆A (σ( a), σ(c)) ⊆ F
⇐⇒ σ[∆A ( a, c)] ⊆ F
⇐⇒ ∆A ( a, c) ⊆ σ−1 [ F ]
⇐⇒ h a, ci ∈ Ω A σ−1 [ F ].

The first and the fourth equivalences above are straightforward, the second holds by
definition of Ω A F, and the third because σ is an endomorphism. The last equivalence
follows from the definition of Ω A too, because σ−1 [ F ] is a deductive filter, by Lemma 4.27.
Hence, we conclude that Ω A : Fi` ( A)+ → ConK ( A)+ is an isomorphism and, therefore,
that Fi` ( A)+ ∼= ConK ( A)+ .
(ii)⇒(iii): In view of condition (4.3), when A = T (Var ), we obtain Fi` ( A)+ = T h(`)+
and ConK ( A)+ = T h(K )+ . Therefore, condition (iii) is a special instance of (ii).
(iii)⇒(i): Consider an isomorphism Φ : T h(`)+ → T h(K )+ . Let also

Cn` : P ( T (Var )) → P ( T (Var )) and CnK : P ( E(Var )) → P ( E(Var ))

be the closure operators associated with the consequence relations ` and K (equiv. with
the closure systems T h(`) and T h(K )), respectively. The proof proceeds through a series
of technical claims.
Claim 4.34. For every set of formula Γ, set of equations Θ and substitution σ,

Φ(Cn` (σ[ Γ ])) = CnK (σ [Φ(Cn` Γ )])


Φ−1 [CnK (σ[Θ])] = Cn` (σ[Φ−1 (CnK (Θ))]).

Proof of the Claim. We detail the proof of the first equality only, as that of the second is
analogous. First, since CnK is a closure operator, we have

σ[Φ(Cn` ( Γ ))] ⊆ CnK (σ[Φ(Cn` ( Γ ))])

55
4. A LGEBRAIZABLE LOGICS

and, therefore,
Φ(Cn` ( Γ )) ⊆ σ−1 [CnK (σ[Φ(Cn` ( Γ ))])]. (4.9)
We will prove that the two sets in the above display are theories of K . For Φ(Cn` ( Γ ))
this is a consequence of the fact that Cn` ( Γ ) ∈ T h(`) and Φ sends theories of ` to
theories of K . To prove that σ−1 [CnK (σ[Φ(Cn` ( Γ ))])] is also a theory of K , observe that
CnK (σ[Φ(Cn` ( Γ ))]) ∈ T h(K ). Since T h(K ) is closed under inverse substitutions (as
the structure T h(K )+ is well defined), this yields the desired result.
Since the two sets in(4.9) are theories of K and Φ−1 : T h(K )+ → T h(`)+ is also an
isomorphism, from (4.9) it follows

Cn` ( Γ ) = Φ−1 Φ(Cn` ( Γ ))


⊆ Φ−1 (σ−1 [CnK (σ[Φ(Cn` ( Γ ))])])
= σ−1 [Φ−1 (CnK (σ[Φ(Cn` ( Γ ))]))].
Consequently,
σ[Cn` ( Γ )] ⊆ Φ−1 (CnK (σ[Φ(Cn` ( Γ ))])).
Notice that the right hand side of the above display belongs to T h(`), because Φ−1 sends
theories of K to theories of `. Therefore, we obtain

Cn` (σ[Cn` ( Γ )]) ⊆ Φ−1 (CnK (σ[Φ(Cn` ( Γ ))])). (4.10)

Lastly,

Φ(Cn` (σ[ Γ ])) ⊆ Φ(Cn` (σ[Cn` ( Γ )]))


⊆ ΦΦ−1 (CnK (σ[Φ(Cn` ( Γ ))]))
= CnK (σ[Φ(Cn` ( Γ ))]).
The inclusions above are justified as follows. To prove the first, notice that σ[ Γ ] ⊆
σ[Cn` ( Γ )]. Since both Cn` and Φ are order preserving, this yields Φ(Cn` (σ[ Γ ])) ⊆
Φ(Cn` (σ [Cn` ( Γ )])), as desired. The second inclusion follows from (4.10) and the fact that
Φ is order preserving, while the last equality follows from the fact that Φ is a bijection.
This establishes the left to right inclusion of the first equality in the statement of
the Claim. To prove the other inclusion, observe that σ[ Γ ] ⊆ Cn` (σ[ Γ ]) and, therefore,
Γ ⊆ σ−1 [Cn` (σ[ Γ ])]. This, in turn, yields Cn` ( Γ ) ⊆ Cn` (σ−1 [Cn` (σ[ Γ ])]). As T h(`)
is closed under inverse substitutions (because T h(`)+ is well defined), we obtain that
σ−1 [Cn` (σ[ Γ ])] is also a theory of `. Consequently,

Cn` ( Γ ) ⊆ Cn` (σ−1 [Cn` (σ[ Γ ])]) = σ−1 [Cn` (σ[ Γ ])].

As at the left and right hand sides of the above displays we have two theories of ` and Φ
is order preserving and commutes with inverse substitutions, we obtain

Φ(Cn` ( Γ )) ⊆ Φ(σ−1 [Cn` (σ[ Γ ])]) = σ−1 [Φ(Cn` (σ[ Γ ]))].

Consequently, σ[Φ(Cn` ( Γ ))] ⊆ Φ(Cn` (σ[ Γ ])) and, therefore,

CnK (σ[Φ(Cn` ( Γ ))]) ⊆ CnK (Φ(Cn` (σ[ Γ ]))) = Φ(Cn` (σ[ Γ ])),

56
4.3. Isomorphism theorem

where the last equality follows from the assumption that Φ sends theories of ` to theories
of K . This establishes the right to left inclusion of the first equality in the statement of the
Claim. 
We will rely on the following formulation of substitution invariance.
Claim 4.35. For every set of formulas Γ, set of equations Θ and substitution σ,

Cn` (σ[Cn` ( Γ )]) = Cn` (σ[ Γ ]) and CnK (σ[CnK (Θ)]) = CnK (σ[Θ]).

Proof of the Claim. As before, we detail the proof of the first equality only. As Γ ⊆ Cn` ( Γ ),
we have σ[ Γ ] ⊆ σ[Cn` ( Γ )] and, therefore, Cn` (σ[ Γ ]) ⊆ Cn` (σ[Cn` ( Γ )]). To prove the
reverse inclusion, it suffices to show that σ[Cn` ( Γ )] ⊆ Cn` (σ[ Γ ]). To this end, consider
ϕ ∈ σ[Cn` ( Γ )]. Then there exists ψ ∈ Cn` ( Γ ) such that ϕ = σ(ψ). Moreover, Γ ` ψ. As `
is substitution invariant, this yields σ[ Γ ] ` σ(ψ), that is, σ[ Γ ] ` ϕ. Hence, we conclude
that ϕ ∈ Cn` (σ[ Γ ]). 
Now, let σx (resp. σx,y ) be the substitution that sends all variables to x (resp. all variables
other than y to x, and leaves y untouched). We define a set of equations τ ( x ) and a set of
formulas ∆( x, y) as follows:

τ ( x ) := σx [Φ(Cn` ({ x }))] and ∆( x, y) := σx,y [Φ−1 (CnK ({ x ≈ y}))].

Our aim is to prove that τ, ∆ and K witness the algebraizability of `.


Claim 4.36. For every formula ϕ,

Φ(Cn` ({ ϕ})) = CnK (τ ( ϕ)).

Proof of the Claim. From Claim 4.34 it follows

Φ(Cn` ({ x })) = Φ(Cn` ({σx ( x )})) = CnK (σx [Φ(Cn` ({ x }))]) = CnK (τ ( x )). (4.11)

Now, consider a formula ϕ and let σ be any substitution such that σ( x ) = ϕ. We have

Φ(Cn` ({ ϕ})) = Φ(Cn` ({σ( x )}))


= CnK (σ[Φ(Cn` ({ x }))])
= CnK (σ[CnK (τ ( x ))])
= CnK (σ[τ ( x )])
= CnK (τ ( ϕ)).

The above equalities are justified as follows. The first holds by the definition of σ, the
second follows from Claim 4.34, the third from (4.11), the fourth from Claim 4.35, and the
last one from the definition of σ. 
Claim 4.36 can be extended to sets of formulas as follows.
Claim 4.37. For every set of formulas Γ,

Φ(Cn` ( Γ )) = CnK (τ [ Γ ]).

57
4. A LGEBRAIZABLE LOGICS

Proof of the Claim. We have

T h(`) T h(K ) T h(K )


Φ(Cn` ( Γ )) = Φ( Φ(Cn` ({γ})) =
_ _ _
Cn` ({γ})) = CnK (τ (γ)).
γ∈ Γ γ∈ Γ γ∈ Γ

The first and the equality above follows from the basic properties of closure operators,
the second from the assumption that Φ : T h(`)+ → T h(K )+ is an isomorphism (and,
therefore, preserves arbitrary joins), and the third from Claim 4.36.
Furthermore, we have

T h(K )
_ [
CnK (τ (γ)) = CnK ( τ (γ)) = CnK (τ [ Γ ]).
γ∈ Γ γ∈ Γ

The first equality above follows from the description of joins in closure systems and the
second from the definition of τ [ Γ ]. 

In view of Proposition 4.3, in order to prove that τ, ∆ and K witness the algebraizability
of `, it suffices to check that conditions (Alg1) and (Alg4*) hold. To prove (Alg1), consider
a set of formulas Γ ∪ { ϕ}. Applying the fact that Φ is an order isomorphism and Claim
4.37, we obtain

Γ ` ϕ ⇐⇒ Cn` ({ ϕ}) ⊆ Cn` ( Γ ) ⇐⇒ Φ(Cn` ({ ϕ})) ⊆ Φ(Cn` ( Γ ))


⇐⇒ CnK (τ ( ϕ)) ⊆ CnK (τ [ Γ ]) ⇐⇒ τ [ Γ ] K τ ( ϕ).

This establishes (Alg1). To prove condition (Alg4*), observe that

CnK ({ x ≈ y}) = CnK (σx,y [{ x ≈ y}])


= CnK (σx,y [CnK ({ x ≈ y})])
= ΦΦ−1 (CnK (σx,y [CnK ({ x ≈ y})]))
= Φ(Cn` (σx,y [Φ−1 (CnK ({ x ≈ y}))]))
= Φ(Cn` (∆( x, y))
= CnK (τ [∆( x, y)]).

The equalities above are justified as follows. The second follows from Claim 4.35, the
fourth from Claim 4.34, and the sixth from Claim 4.37. From the above display it follows
(Alg4*), thus concluding the proof. 

Corollary 4.38. Let ` be an algebraizable logic whose algebraizability is witnessed by τ, ∆ and


K. For every algebra A of the suitable type, the map Ω A : Fi` ( A)+ → ConK ( A)+ , defined by the
rule
Ω A F := {h a, ci ∈ A × A : ∆A ( a, c) ⊆ F },

is an isomorphism.

58
4.3. Isomorphism theorem

In algebraic logic, the map Ω A : Fi` ( A)+ → ConK ( A)+ is known as the Leibniz operator
[42, 55, 57, 115]. Its behavior influences the definability of logical equivalence [20, 38, 43, 74]
and of truth predicates [72, 102, 113] in matrix semantics and, more in general, in equality
free model theory [23, 32, 48, 51, 52].
Remark 4.39. The implication (ii)⇒(i) in the Isomorphism Theorem suggests the non-
mathematical thesis that every correspondence between congruences and distinguished
sets is induced by the algebraization of some logic. Large classes of varieties for which
such a correspondence exists have been identified in [126, 3, 4, 5, 127] and [88], see also
[2, 71]. However, in some of these cases, the isomorphism between congruences and distin-
guished sets does not commute with inverse substitutions. The notion of an algebraizable
logic has been weakened in [46] to accommodate for these situations too. 
Remark 4.40. At this stage, it is worth mentioning that algebraizable logics admit the
following purely syntactic description [55, Thm. 3.21].

Theorem 4.41. A finitary logic ` is algebraizable if and only if there are a finite set of equations τ
and a finite set of formulas ∆ such that, for every n-ary connective f ,

∅ `∆( x, x )
x, ∆( x, y) `y
∆( x1 , y1 ) ∪ · · · ∪ ∆( xn , yn ) `∆( f ( x1 , . . . , xn ), f (y1 , . . . , yn ))
x a `∆[τ ( x )].

In view of the Isomorphism Theorem, the above result identifies syntactic conditions
equivalent to the existence of an isomorphism between lattices of K-congruences and de-
ductive filters, for a suitable quasi-variety K. Similar results play a central role in universal
algebra, where Maltsev conditions provide a syntactic description of structural properties
of congruence lattices [9, 65, 85, 107, 125]. The connection with Maltsev conditions has
been explored in [80, 81, 82]. 

59
CHAPTER 5
Deduction theorems

5.1 Bridge theorem


For general information on deduction theorems in algebraic logic, the reader might consult
[22, 26, 40, 41, 42, 45, 114].

Definition 5.1. A logic ` has a deduction-detachment theorem (DDT) if there exists a finite
set of formulas I ( x, y) such that for every set of formulas Γ ∪ { ϕ, ψ},

Γ, ϕ ` ψ ⇐⇒ Γ ` I ( ϕ, ψ).

In this case, we say that I ( x, y) witnesses the DDT for `.

It is well known that, for every set of formulas Γ ∪ { ϕ, ψ},

Γ, ϕ `IPC ψ ⇐⇒ Γ `IPC ϕ → ψ.

Therefore, IPC has a DDT witnessed by I ( x, y) := { x → y}. The same holds for K` (see
Exercise 2.23). On the other hand, we will prove that Kg lacks any DDT (Example 5.52).
Exercise 5.2. Prove that the DDT persists in axiomatic extensions. Conclude that all super-
intuitionistic logics and all axiomatic extensions of K` have a DDT. On the other hand, the
DDT does not persist in arbitrary (i.e., not necessarily axiomatic) extensions. 
Let K be a quasi-variety and A an algebra. Recall from Corollary 3.40 that the closure
operator of K-congruence generation on A is denoted by

CgKA : P ( A × A) → P ( A × A).

Furthermore, given a, c ∈ A, we abbreviate CgKA ({h a, ci}) as CgKA ( a, c). Accordingly, the
K-congruences of A of the form CgKA ( a, c) will be called principal.
When K is a variety and A ∈ K, we drop the subscript K in CgKA , because CgKA ( X ) is
the least congruence of A extending X.

61
5. D EDUCTION THEOREMS

Definition 5.3. A quasi-variety K is said to have equationally definable principal relative


congruences (EDPRC) when there exists a finite set of equations Φ( x, y, z, v) such that, for
every A ∈ K and a, b, c, d ∈ A,

h a, bi ∈ CgKA (c, d) ⇐⇒ A  Φ(c, d, a, b).

In this case, we say that Φ witnesses EDPRC for K. When K is a variety, it is common to use
the expression equationally definable principal congruences (EDPC), as opposed to (EDPRC).

This notion originates in [8] and was studied in the series of papers [19, 18, 24, 25, 87].
Remark 5.4. EDPRC persists in relative subvarieties, but not necessarily in subquasi-
varieties (cf. Exercise 5.2). 
Exercise 5.5. Prove that, for every Heyting algebra A and c, d ∈ A,

Cg A (c, d) = {h a, bi ∈ A × A : c ↔ d 6 a ↔ b}.

Use this fact to infer that the set of equations

Φ( x, y, z, v) := { x ↔ y 6 z ↔ v}

witnesses EDPC for the variety of Heyting algebras. As EDPC persists in subvarieties, this
shows that every variety of Heyting algebras has EDPC. 
Exercise 5.6. The case of modal algebras is more complex. For every n ∈ N, we define

n x := x ∧ x ∧ x ∧ · · · ∧ n x.

Prove that, for every modal algebra A and c, d ∈ A,

Cg A (c, d) = {h a, bi ∈ A × A : there exists n ∈ N such that n (c ↔ d) 6 a ↔ b}.

A variety K of modal algebras is said to be weakly transitive if

K  n x 6 n+1 x, for some n ∈ N.

Prove that every weakly transitive variety of modal algebras has EDPC. Hint: use sets of
equations of the form

Φ( x, y, z, v) := {n ( x ↔ y) 6 z ↔ v}.

As shown in Theorem 5.55, the converse is also true, whence the varieties of modal
algebras with EDPC are precisely the weakly transitive ones. 
Our aim is to prove the following bridge theorem, connecting the DDT on the logic
side with EDPRC on the algebra side [26].

Theorem 5.7 (Blok & Pigozzi). An algebraizable logic has a DDT if and only if its equivalent
algebraic semantics has EDPRC.

62
5.1. Bridge theorem

To this end, it is convenient to extend the definition of a DDT to relative equational


consequences.
Definition 5.8. Given a quasi-variety K, we say that K has a deduction-detachment theorem
(DDT) if there exists a finite set of equations Φ( x, y, z, v) such that, for every Θ ∪ { ϕ ≈
ψ, ε ≈ δ} ⊆ E(Var ),
Θ, ϕ ≈ ψ K ε ≈ δ ⇐⇒ Θ K Φ( ϕ, ψ, ε, δ).
In this case, we say that Ψ witnesses the DDT for K .
Proposition 5.9. An algebraizable logic ` has a DDT if and only if the equational consequence
K relative to its equivalent algebraic semantics K has one.
Proof. We shall detail only the proof of the “only if” part, as the other one is analogous. To
this end, assume that ` has a DDT witnessed by a finite set of formulas I ( x, y). For every
n ∈ N, we define a set In ( x1 , . . . , xn , y) recursively by the following rule:
I0 (y) := {y}
[
Ik+1 ( x1 , . . . , xk+1 , y) := { I ( x1 , ϕ) : ϕ ∈ Ik ( x2 , . . . , xk+1 , y)}.
A straightforward inductive argument shows that for every n ∈ N and every set of
formulas Γ ∪ {ψ} ∪ { ϕi : i < n},
Γ ∪ { ϕi : i < n} ` ψ ⇐⇒ Γ ` In ( ϕ0 , . . . , ϕn−1 , ψ). (5.1)
As ` is algebraizable, there are a finite set of equations τ ( x ) and a finite set of formulas
∆( x, y) such that the tripleτ, ∆ and K witnesses the algebraizability of `. As ∆ is finite, it
has the form {δi : i < n} for some n ∈ N. Then, consider the following set of equations
Φ( x, y, z, v) :=
[
τ [ In (δ0 ( x, y), . . . , δn−1 ( x, y), δi (z, v))].
i <n

Observe that Φ is finite, as so are ∆ and τ. We shall see that Φ witnesses a DDT for K . To
this end, consider a set of equations Θ ∪ { ϕ ≈ ψ, α ≈ β} ⊆ E(Var ). We have
Θ, ϕ ≈ ψ K α ≈ β ⇐⇒ ∆[Θ], ∆( ϕ, ψ) ` ∆(α, β)
⇐⇒ ∆[Θ] ∪ {δi ( ϕ, ψ) : i < n} ` δj (α, β), for all j < n
[
⇐⇒ ∆[Θ] ` In (δ0 ( ϕ, ψ), . . . , δn−1 ( ϕ, ψ), δi (α, β))
i <n
[
⇐⇒ τ [∆[Θ]] K τ [ In (δ0 ( ϕ, ψ), . . . , δn−1 ( ϕ, ψ), δi (α, β))]
i <n
[
⇐⇒ Θ K τ [ In (δ0 ( ϕ, ψ), . . . , δn−1 ( ϕ, ψ), δi (α, β))]
i <n
⇐⇒ Θ K Φ( ϕ, ψ, α, β).
The above equivalences justified as follows: the first follows from (Alg2), the second is
obvious, the third holds because I witnesses a DDT for `, the fourth follows from (Alg1),
the fifth from (Alg4) and the last one from the definition of Φ. Hence, we conclude that Φ
witnesses a DDT for K , as desired. 

63
5. D EDUCTION THEOREMS

Accordingly, in order to establish Theorem 5.7, it suffices to prove the following.

Proposition 5.10. A quasi-variety K has EDPRC if and only if K has a DDT.

To this end, we rely on two technical lemmas.

Lemma 5.11. Let K be a quasi-variety, A an algebra and a, c ∈ A. Then

CgKA ( a, c) = {CgKB ( a, c) : B ∈ S( A) is finitely generated and a, c ∈ B}.


[

Proof. The inclusion from right to left follows from the fact that if B is a subalgebra of A
and θ a K-congruence of A, then θ ∩ ( B × B) is a K-congruence of B. This observation will
be used without further notice in the proof.
In order to prove the inclusion from left to right, consider the relation

{CgKB ( a, c) : B ∈ S( A) is finitely generated and a, c ∈ B}.


[
θ :=

It is easy to see that θ is a congruence of A containing the pair h a, ci. In order to check that
θ is also a K-congruence of A, consider a quasi-equation

&
i 6n
ϕi ≈ ψi =⇒ ε ≈ δ (5.2)

valid in K and a homomorphism f : T (Var ) → A/θ such that f ( ϕi ) = f (ψi ), for all i 6 n.
Moreover, let π : A → A/θ be the canonical surjection. By Corollary 1.6, there exists a
homomorphism g : T (Var ) → A such that f = π ◦ g.
For every i 6 n, we have

g( ϕi )/θ = π ( g( ϕi )) = f ( ϕi ) = f (ψi ) = π ( g(ψi )) = g(ψi )/θ

and, therefore, h g( ϕi ), g(ψi )i ∈ θ. By definition of θ, for every i 6 n there exists a finite


Bi
subset Xi of A such that h g( ϕi ), g(ψi )i ∈ CgK ( a, c), where Bi is the subalgebra of A
generated by Xi . As the definition of θ requires Bi to contain a and c, we may assume that
a, c ∈ Xi .
Let then { x1 , . . . , xm } be the set of variables occurring in the quasi-equation in (5.2)
and C the subalgebra of A generated by

X1 ∪ · · · ∪ Xn ∪ { g( x1 ), . . . , g( xm )}.

Clearly, C is finitely generated, it contains a, c, and

B1 , . . . , Bn ∈ S(C ).
Bi C
Since CgK ( a, c) ⊆ CgK ( a, c) for every i 6 n, we have
C
h g( ϕi ), g(ψi )i ∈ CgK ( a, c), for all i 6 n.

Consider a homomorphism h : T (Var ) → C such that

h ( x1 ) = g ( x1 ), . . . , h ( x m ) = g ( x m ).

64
5.1. Bridge theorem

This is possible because g( x1 ), . . . , g( xm ) ∈ C, by definition of C. Let also p : C →


C
C/CgK ( a, c) be the canonical surjection. For every i 6 n, we have
C C
p(h( ϕi )) = p( g( ϕi )) = g( ϕi )/CgK ( a, c) = g(ψi )/CgK ( a, c) = p( g(ψi )) = p(h(ψi )).
C
Since CgK ( a, c) is a K-congruence of C, the above display implies p(h(ε)) = p(h(δ)). In
turn, this guarantees that
C C
g(ε)/CgK ( a, c) = p( g(ε)) = p(h(ε)) = p(h(δ)) = p(h(δ)) = g(δ)/CgK ( a, c).
C
As a consequence, h g(ε), g(δ)i ∈ CgK ( a, c) ⊆ θ. Hence, we conclude that

f (ε) = π ( g(ε)) = g(ε)/θ = g(δ)/θ = π ( g(δ)) = f (δ).

It follows that θ is a K-congruence of A containing h a, ci, whence CgKA ( a, c) ⊆ θ. 

Lemma 5.12. Let K be a quasi-variety and A and algebra. If θ ∈ ConK ( A) and a, b ∈ A, then

CgKA (θ ∪ {h a, bi}) = {hc, di ∈ A × A : hc/θ, d/θ i ∈ CgKA/θ ( a/θ, b/θ )}.

Proof. We define

φ := {hc, di ∈ A × A : hc/θ, d/θ i ∈ CgKA/θ ( a/θ, b/θ )}.

The proof that φ is a K-congruence of A is left to you as an exercise.


Now, the definition of φ guarantees that θ ∪ {h a, bi} ⊆ φ. Since φ is a K-congruence
of A, this implies CgKA (θ ∪ {h a, bi}) ⊆ φ. To prove the reverse inclusion, we reason by
/ CgKA (θ ∪ {h a, bi}).
contraposition. Accordingly, consider a pair c, d ∈ A such that hc, di ∈
A
Then let f : A/θ → A/CgK (θ ∪ {h a, bi}) be the map defined by the rule

f (e/θ ) := e/CgKA (θ ∪ {h a, bi}).

Notice that f is well defined homomorphism, because θ ⊆ CgKA (θ ∪ {h a, bi}). Together


with A/CgKA (θ ∪ {h a, bi}) ∈ K, this implies that Ker( f ) is a K-congruence of A/θ. Further-
more,
h a/θ, b/θ i ∈ Ker( f ) and hc/θ, d/θ i ∈ / Ker( f ),
because h a, bi ∈ CgKA (θ ∪ {h a, bi}) and hc, di ∈
/ CgKA (θ ∪ {h a, bi}). In view of the above
display, we obtain
/ CgKA/θ ( a/θ, b/θ )
hc/θ, d/θ i ∈

/ φ. Hence, we conclude that φ ⊆ CgKA/θ ( a/θ, b/θ ), as desired.


and, therefore, hc, di ∈ 

Exercise 5.13. Complete the proof of the above lemma by showing that the relation φ is
indeed a K-congruence of A. 
We are now ready to prove Proposition 5.10.

65
5. D EDUCTION THEOREMS

Proof. In order to prove the “if” part, suppose that K has a DDT witnessed by a finite set
of equations Φ( x, y, z, v). We shall see that Φ witnesses EDPRC for K. In view of Lemma
5.11, it suffices to show that for every finitely generated A ∈ K and a, b, c, d ∈ A,

h a, bi ∈ CgKA (c, d) ⇐⇒ A  Φ(c, d, a, b). (5.3)

Suppose first that h a, bi ∈ CgKA (c, d). Since A is finitely generated, we can choose some
generators e1 , . . . , en for it. Then there are ϕ a , ϕb , ϕc and ϕd in variables x1 , . . . , xn such that

ϕ aA (e1 , . . . , en ) = a
ϕbA (e1 , . . . , en ) = b
ϕcA (e1 , . . . , en ) = c
ϕdA (e1 , . . . , en ) = d.

Furthermore, consider the set of equations

Θ := {ε( x1 , . . . , xn ) ≈ δ( x1 , . . . , xn ) ∈ E(Var ) : ε A (e1 , . . . , en ) = δ A (e1 , . . . , en )}.

We claim that
Θ, ϕc ≈ ϕd K ϕ a ≈ ϕb .
To prove this, consider an algebra B ∈ K and elements p1 , . . . , pn ∈ B such that

ε B ( p1 , . . . , pn ) = δ B ( p1 , . . . , pn ) and ϕcB ( p1 , . . . , pn ) = ϕdB ( p1 , . . . , pn ), (5.4)

for every ε ≈ δ ∈ Θ. Moreover, let f : A → B be the map defined, for every formula
γ( x1 , . . . , xn ), by the rule

γ A (e1 , . . . , en ) 7−→ γ B ( p1 , . . . , pn ).

From the left hand side of (5.4) and the fact that A is generated by e1 , . . . , en it follows that
f is a well-defined homomorphism from A to B. Therefore, Ker( f ) is a congruence of A.
Furthermore, as A/Ker( f ) ∈ IS( B) and B ∈ K, we obtain that Ker( f ) is a K-congruence
of A. We have that

f ( a) = f ( ϕ aA (e1 , . . . , en )) = ϕ aB ( p1 , . . . , pn )
f (b) = f ( ϕbA (e1 , . . . , en )) = ϕbB ( p1 , . . . , pn )
f (c) = f ( ϕcA (e1 , . . . , en )) = ϕcB ( p1 , . . . , pn )
f (d) = f ( ϕdA (e1 , . . . , en )) = ϕdB ( p1 , . . . , pn ).

Therefore, the right hand side of (5.4) guarantees that hc, di ∈ Ker( f ). As a consequence,
Ker( f ) is a K-congruence of A containing the pair hc, di. It follows that CgKA (c, d) ⊆ Ker( f ).
Together with the assumption that h a, bi ∈ CgKA (c, d), this implies

ϕ aB ( p1 , . . . , pn ) = f ( a) = f (b) = ϕbB ( p1 , . . . , pn ),

establishing the claim.

66
5.1. Bridge theorem

As Φ witnesses a DDT for K , the claim implies

Θ K Φ ( ϕ c , ϕ d , ϕ a , ϕ b ) .

Now, the definition of Θ guarantees that ε A (e1 , . . . , en ) = δ A (e1 , . . . , en ), for every ε ≈ δ ∈


Θ. Together with the above display, this yields

A  Φ( ϕc (e1 , . . . , en ), ϕd (e1 , . . . , en ), ϕ a (e1 , . . . , en ), ϕb (e1 , . . . , en )),

i.e., A  Φ(c, d, a, b). This establishes the implication from left to right in (5.3).
To prove the other implication in (5.3), notice that from Φ( x, y, z, v) K Φ( x, y, z, v)
and the assumption that Φ witnesses a DDT for K , it follows

x ≈ y, Φ( x, y, z, v) K z ≈ v. (5.5)

Suppose that A  Φ(c, d, a, b). Then the set of premises of the above display is valid in
A/CgKA (c, d) under the assignment

x 7−→ c/CgKA (c, d) y 7−→ d/CgKA (c, d)


z 7−→ a/CgKA (c, d) v 7−→ b/CgKA (c, d).

Therefore, by (5.5), we obtain that a/CgKA (c, d) = b/CgKA (c, d), i.e., h a, bi ∈ CgKA (c, d). As
this establishes (5.3), we conclude that K has EDPRC.
To prove the “only if” part of the statement, suppose that K has EDPRC witnessed by
set Φ( x, y, z, v). We shall see that Φ witnesses a DDT for K .
Recall from Proposition 4.26 that the closure systems ConK ( T (Var )) and T h(K ) coin-
cide. Accordingly, the closure operators
T (Var )
CgK : P ( Eq) → P ( Eq) and CnK : P ( Eq) → P ( Eq)

associated with them coincide too. Bearing this in mind, consider a set of equations
Θ ∪ { ϕ ≈ ψ, ε ≈ δ} and define
T (Var )
θ := CgK (Θ).
Bearing in mind that equations are ordered pairs of formulas, we obtain

Θ, ϕ ≈ ψ K ε ≈ δ ⇐⇒ ε ≈ δ ∈ CnK (Θ ∪ { ϕ ≈ ψ})
T (Var )
⇐⇒ hε, δi ∈ CgK (Θ ∪ {h ϕ, ψi})
T (Var )
⇐⇒ hε, δi ∈ CgK (θ ∪ {h ϕ, ψi})
T (Var )/θ
⇐⇒ hε/θ, δ/θ i ∈ CgK ( ϕ/θ, ψ/θ )
⇐⇒ T (Var )/θ  Φ( ϕ/θ, ψ/θ, ε/θ, δ/θ )
⇐⇒ Φ( ϕ, ψ, ε, δ) ⊆ θ
T (Var )
⇐⇒ Φ( ϕ, ψ, ε, δ) ⊆ CgK (Θ)
⇐⇒ Φ( ϕ, ψ, ε, δ) ⊆ CnK (Θ)
⇐⇒ Θ K Φ( ϕ, ψ, ε, δ).

67
5. D EDUCTION THEOREMS

The above equivalences are justified as follows. The first and the latter are obvious, the
T (Var )
second and second to last follow from the the fact that the closure operators CgK and
CnK coincide, and the third and the third to last from the definition of θ and the fact that
T (Var )
CgK is a closure operator. Lastly, the fourth equality is a consequence of Lemma 5.12,
while the fifth and the sixth are immediate applications of the definitions. 

5.2 Quasi-varieties with EDPRC


The aim of this section is to prove that the demand that a quasi-variety K has EDPRC is
equivalent to a purely order theoretic property of lattices of K-congruences.

Definition 5.14. Let A be a complete lattice.

(i) An element a ∈ A is said to be compact if for every X ⊆ A,


_ _
if a 6 X, then there is a finite Y ⊆ X such that a 6 Y.

(ii) A is said to be algebraic if every element is a join of compact elements.

We denote by C( A) the set of compact elements of A.

In order to present canonical examples of algebraic lattices, recall that a semilattice is


an algebra A = h A; ∗i such that ∗ is a binary idempotent, commutative and associative
operation. Every semilattice A can be associated with two partial orders on A, namely the
meet order 6m and the join order 6 j , defined respectively by the following rules:

a 6m c ⇐⇒ a ∗ c = a and a 6 j c ⇐⇒ a ∗ c = c.

Accordingly, we say that A is a meet semilattice (resp. join semilattice) when we give priority
to the meet order (resp. join order), which, in this case, will be denoted simply by 6.

Example 5.15 (Algebraic lattices). Let A be a join semilattice with minimum element. A
set I ⊆ A is said to be an ideal of A if it is a nonempty downset such that if a, c ∈ I, then
a ∗ c ∈ I. The poset I( A) of ideals of A ordered under the inclusion relation is an algebraic
lattice, whose compact elements are the principal downsets of A. 

We will prove that every algebraic lattice arises in this way (Theorem 5.17). To this end,
we rely on the following immediate consequence of the definition of a compact element.

Proposition 5.16. If A is a complete lattice and 0 its minimum element, then

if a, c ∈ C( A), then a ∨ c ∈ C( A).

Consequently, C( A) can be viewed as a join semilattice with minimum hC( A); ∨, 0i, whose order
coincides with the restriction of the order of A to C( A).

As a consequence, we obtain a representation theorem for algebraic lattices.

68
5.2. Quasi-varieties with EDPRC

Theorem 5.17. A poset is an algebraic lattice if and only if it is isomorphic to the lattice of ideals
of a join semilattice with minimum.

Proof. The “if” part is supplied in Example 5.15. To prove the “only if” part, consider an
algebraic lattice A. In view of Proposition 5.16, C( A) is a join semilattice with minimum.
It is not hard to prove that the map f : A → I(C( A)), defined by the rule

a 7−→ C( A) ∩ ↓ a,

is an isomorphism. 

Exercise 5.18. Complete the above proof by showing that f is indeed an isomorphism. 
Algebraic lattices play a central role in algebra, partly because of the next observation.

Proposition 5.19. Let K be a quasi-variety. If A is an algebra of the same type, ConK ( A) is an


algebraic lattice, whose compact elements are precisely the finitely generated K-congruences of A,
i.e., the K-congruence of the form CgKA ( X ) for some finite X ⊆ A × A.

Remark 5.20. Notably, nothing more can be said about lattices of K-congruences in general,
as every algebraic lattice is isomorphic to the congruence lattice of some algebra [69]. 
Exercise 5.21. Prove Proposition 5.19. 
The following semilattices are instrumental to characterize quasi-varieties with ED-
PRC.

Definition 5.22. A dually Brouwerian semilattice is an algebra A = h A; ∨, ←i such that


h A; ∨i is a join semilattice and ← a binary operation such that, for every a, b, c ∈ A,

c 6 a ∨ b ⇐⇒ c ← b 6 a.

Typical examples of dually Brouwerian semilattices arise from Heyting algebras. More
precisely, if A is a Heyting algebra with order 6, the join order of the semilattice h A; ∧i is
the dual relation >. Because of this, from the residuation law it follows that h A; ∧, →i is a
dually Brouwerian semilattice.
Our aim is to establish the following result [87].

Theorem 5.23 (Köhler & Pigozzi). A quasi-variety K has EDPRC if and only if, for every algebra
A ∈ K, the join semilattice C(ConK ( A)) can be endowed with a binary operation ← that turns it
into a dually Brouwerian semilattice.

In order to prove Theorem 5.23, we rely on the following observation.

Lemma 5.24. Let A be a join semilattice generated by a set X ⊆ A. Suppose that for every
b, c ∈ X there exists an element c L99 b ∈ A such that for all a ∈ A,

c 6 a ∗ b ⇐⇒ c L99 b 6 a.

Then there is a binary operation ← on A such that h A; ∗, ←i is a dually Brouwerian semilattice.

69
5. D EDUCTION THEOREMS

Proof. As X generates A, it suffices to show that, for every b1 , . . . , bn ∈ X and c ∈ A, there


exists an element c ← (b1 ∗ · · · ∗ bn ) ∈ A such that, for all a ∈ A,

c 6 a ∗ (b1 ∗ · · · ∗ bn ) ⇐⇒ c ← (b1 ∗ · · · ∗ bn ) 6 a.

We will prove this by induction on n.


For the base case, consider two elements b ∈ X and c ∈ A. As X generates h A; ∗i, there
are c1 , . . . , cn ∈ X such that c = c1 ∗ · · · ∗ cn . We set

c ← b := (c1 L99 b) ∗ · · · ∗ (cn L99 b).

Observe that for all a ∈ A,

c 6 a ∗ b ⇐⇒ c1 ∗ · · · ∗ cn 6 a ∗ b
⇐⇒ ci 6 a ∗ b, for all i 6 n
⇐⇒ ci L99 b 6 a, for all i 6 n
⇐⇒ (c1 L99 b) ∗ · · · ∗ (cn L99 b) 6 a
⇐⇒ c ← b 6 a.

For the induction step, consider b1 , . . . , bk+1 ∈ X and c ∈ A. By the inductive hypothe-
sis, we can define

c ← (b1 ∗ · · · ∗ bk+1 ) := (c ← (b1 ∗ · · · ∗ bk )) ← bk+1 .

For every a ∈ A, we have

c 6 a ∗ (b1 ∗ · · · ∗ bk+1 ) ⇐⇒ c 6 ( a ∗ bk+1 ) ∗ (b1 ∗ · · · ∗ bk )


⇐⇒ c ← (b1 ∗ · · · ∗ bk ) 6 a ∗ bk+1
⇐⇒ (c ← (b1 ∗ · · · ∗ bk )) ← bk+1 6 a
⇐⇒ c ← (b1 ∗ · · · ∗ bk+1 ) 6 a.

The above equivalences can be justified as follows. The first holds because ∗ is idempotent
and commutative, the second follows from the induction hypothesis for the case n = k,
the third from the induction hypothesis for the case n = 1, and the last from the definition
of c ← (b1 ∗ · · · ∗ bk+1 ). 
We are now ready to prove Theorem 5.23.

Proof. To prove the “only if” part, suppose that K has EDPRC and consider an algebra A ∈
K. Moreover, let ∨ be the join operation of the lattice ConK ( A). Recall from Proposition
5.16 that hC(ConK ( A)); ∨i is a join semilattice, whose order is the inclusion relation.
Furthermore, in view of Proposition 5.19, the elements of C(ConK ( A)) are precisely the
finitely generated K-congruences of A. Therefore, the semilattice hC(ConK ( A)); ∨i is
generated by the principal K-congruences of A, because for every finite X ⊆ A × A,

CgKA ( X ) = CgKA ( a, c).


_

h a,ci∈ X

70
5.2. Quasi-varieties with EDPRC

By Lemma 5.24, to conclude the proof it suffices to show that for every a, b, c, d ∈
A, there exists an element CgKA ( a, b) ← CgKA (c, d) ∈ C(ConK ( A)) such that for all θ ∈
C(ConK ( A)),

CgKA ( a, b) ⊆ θ ∨ CgKA (c, d) ⇐⇒ CgKA ( a, b) ← CgKA (c, d) ⊆ θ. (5.6)

To this end, consider a, b, c, d ∈ A and let Φ( x, y, z, v) be the finite set of equations


witnessing EDPRC for K. We will work under the identification of Φ A (c, d, a, b) with the
following subset of A × A:

{h ϕ A (c, d, a, b), ψ A (c, d, a, b)i : ϕ ≈ ψ ∈ Φ}.

Bearing this in mind, we define

CgKA ( a, b) ← CgKA (c, d) := CgKA (Φ A (c, d, a, b)).

Since Φ A (c, d, a, b) is finite, CgKA ( a, b) ← CgKA (c, d) is a finitely generated K-congruence of


A and, therefore, a compact element of ConK ( A), by Proposition 5.19. In brief,

CgKA ( a, b) ← CgKA (c, d) ∈ C(ConK ( A)).

Now, for every θ ∈ C(ConK ( A)), we have

CgKA ( a, b) ⊆ θ ∨ CgKA (c, d) ⇐⇒ h a, bi ∈ θ ∨ CgKA (c, d)


⇐⇒ h a, bi ∈ CgKA (θ ∪ {hc, di})
⇐⇒ h a/θ, b/θ i ∈ CgKA/θ (c/θ, d/θ )
⇐⇒ Φ A (c, d, a, b) ⊆ θ
⇐⇒ CgKA/θ (Φ A (c, d, a, b)) ⊆ θ
⇐⇒ CgKA ( a, b) ← CgKA (c, d) ⊆ θ.

The above equivalences are justified as follows. The first, second and fifth hold because
CgKA is a closure operator. The third is a consequence of Lemma 5.12, while the fourth
holds because Φ witnesses EDPRC for K. Finally, the last one is a consequence of the
definition of CgKA ( a, b) ← CgKA (c, d). This establishes (5.6), whence C(ConK ( A)) can be
turned into a dually Brouwerian semilattice.
To prove the “if” part of the statement, let θ be the minimum of ConK ( T (Var )) and
consider the algebra F := T (Var )/θ. Clearly, F ∈ K. By assumption the join semilat-
tice hC(ConK ( F )); ∨i can be endowed with a binary operation ← that turns it into a
F
dually Brouwerian semilattice. As the K-congruences CgK ( x/θ, y/θ ) and CgKF (z/θ, v/θ )
are finitely generated, they belong to C(ConK ( F )), by Proposition 5.19. Thus,
F
CgK (z/θ, v/θ ) ← CgKF ( x/θ, y/θ )

is a well-defined compact K-congruence of F. By Proposition 5.19, it also finitely generated,


i.e., there is a finite set of pairs of formulas

Φ = {hε i ( x, y, z, v, w1 , . . . , wn ), δi ( x, y, z, v, w1 , . . . , wn )i : i < m}

71
5. D EDUCTION THEOREMS

such that
F
CgK (z, v) ← CgKF ( x, y) = CgKF ({hε i /θ, δi /θ i : i < m}).
We will prove that the set of equations

Ψ := {ε i ( x, y, z, v, x . . . x ) ≈ δi ( x, y, z, v, x, . . . , x ) : i < m}

witnesses EDPRC for K.


First, in view of Lemma 5.11, it suffices to prove that for every finitely generated A ∈ K
and a, b, c, d ∈ A,
h a, bi ∈ CgKA (c, d) ⇐⇒ A  Ψ(c, d, a, b).
Accordingly, consider a finitely generated A ∈ K and a, b, c, d ∈ A. The proof will proceed
through a series of claims.
Claim 5.25. There exists a surjective homomorphism g : F → A such that

g( x/θ ) = c g(y/θ ) = d g(z/θ ) = a g(v/θ ) = b

and g(wi /θ ) = c, for every i 6 n.

Proof of the Claim. Let {e1 , . . . , ek } be a set of generators for A. Then consider a function
f : Var → A such that {e1 , . . . , en } ⊆ f [Var ],

f (x) = c f (y) = d f (z) = a f (v) = b.

and f (wi ) = c, for every i 6 n. By Proposition 1.4, f can be extended to a homomor-


phism f ∗ : T (Var ) → A. Since T (Var )/Ker( f ∗ ) ∈ IS( A) ⊆ K, we obtain Ker( f ∗ ) ∈
ConK ( T (Var )). As θ is the minimum of ConK ( T (Var )), this implies θ ⊆ Ker( f ∗ ). Because
of this, the map g : F → A, defined by the rue

ϕ/θ 7−→ f ∗ ( ϕ),

is a well-defined homomorphism such that

g( x/θ ) = c g(y/θ ) = d g(z/θ ) = a g(v/θ ) = b

and g(wi /θ ) = c, for every i 6 n. Furthermore, g is surjective because g[Var/θ ] contains


the generators e1 , . . . , ek of A. 

Claim 5.26. The following equivalence holds:

h a, bi ∈ CgKA (c, d) ⇐⇒ hz/θ, v/θ i ∈ Ker( g) ∨ConK ( F ) CgKF ( x/θ, y/θ ).

Proof of the Claim. Since g : F → A is a sujective homomorphism, from Corollary 1.19 it


follows that the map h : F/Ker( g) → A, defined by the rule

( ϕ/θ )/Ker( g) 7−→ g( ϕ/θ ),

72
5.2. Quasi-varieties with EDPRC

is an isomorphism. As a consequence, h a, bi ∈ CgKA (c, d) if and only if


F/Ker( g)
h(z/θ )/Ker( g), (v/θ )/Ker( g)i ∈ CgK (( x/θ )/Ker( g), (y/θ )/Ker( g)).
In view of Lemma 5.12, the above display is equivalent to
hz/θ, v/θ i ∈ CgKF (Ker( g) ∪ {h x/θ, y/θ i}).
Hence, we conclude that
h a, bi ∈ CgKA (c, d) ⇐⇒ hz/θ, v/θ i ∈ CgKF (Ker( g) ∪ {h x/θ, y/θ i})
⇐⇒ hz/θ, v/θ i ∈ Ker( g) ∨ConK ( F ) CgKF ( x/θ, y/θ ),
as desired. 
Claim 5.27. The following equivalence holds:
F
CgK (z/θ, v/θ ) ← CgKF ( x/θ, y/θ ) ⊆ Ker( g) ⇐⇒ A  Ψ(c, d, a, b).
Proof of the Claim. We have
F
CgK (z/θ, v/θ ) ← CgKF ( x/θ, y/θ ) ⊆ Ker( g)
⇐⇒ CgKF ({hε i /θ, δi /θ i : i < m}) ⊆ Ker( g)
⇐⇒ {hε i /θ, δi /θ i : i < m} ⊆ Ker( g)
⇐⇒ g(ε i ( x, y, z, v, w1 , . . . , wn )/θ ) = g(δi ( x, y, z, v, w1 , . . . , wn )/θ ), for all i < m
⇐⇒ δiA (c, d, a, b, c, . . . , c) = ε A
i ( c, d, a, b, c, . . . , c ), for all i < m
⇐⇒ A  Ψ(c, d, a, b).
The above equivalences are justified as follows. The first holds by the definition of
F
CgK (z/θ, v/θ ) ← CgKF ( x/θ, y/θ ), the second because CgKF is a closure operator, the third
follows from the definition of g, the fourth from Claim 5.25, and the last from the definition
of Ψ. 
In view of Claims 5.26 and 5.27, to conclude the proof, it suffices to show that
hz/θ, v/θ i ∈ Ker( g) ∨ CgKF ( x/θ, y/θ ) ⇐⇒ CgKF (z/θ, v/θ ) ← CgKF ( x/θ, y/θ ) ⊆ Ker( g).
This follows from the following series of equivalences:
hz/θ, v/θ i ∈ Ker( g) ∨ CgKF ( x/θ, y/θ )
⇐⇒ CgKF (z/θ, v/θ ) ⊆ Ker( g) ∨ CgKF ( x/θ, y/θ )
⇐⇒ CgKF (z/θ, v/θ ) ⊆ CgKF (Σ) ∨ CgKF ( x/θ, y/θ ), for some finite Σ ⊆ Ker( g)
⇐⇒ CgKF (z/θ, v/θ ) ← CgKF ( x/θ, y/θ ) ⊆ CgKF (Σ), for some finite Σ ⊆ Ker( g)
⇐⇒ CgKF (z/θ, v/θ ) ← CgKF ( x/θ, y/θ ) ⊆ Ker( g).
F
The first equivalence follows from the fact that CgK is a closure operator, the second and
F F
the last from the compactness of CgK (z/θ, v/θ ) and CgK (z/θ, v/θ ) ← CgKF ( x/θ, y/θ ).
Lastly, the third follows from the fact that hC(ConK ( F )); ∨, ←i is a dually Brouwerian
semilattice. 

73
5. D EDUCTION THEOREMS

Remark 5.28. The proof of Claim 5.25 can be slightly simplified by observing that F is the
free algebra of K with a denumerable set of free generators. However, since we did not
introduce the notion of a free algebra, we opted for spelling the details in full. 
Remark 5.29. In view of Theorem 5.23, EDPRC can be viewed as a property of lattices
of K-congruences. Since their structure is preserved by category equivalences between
quasi-varieties, this implies that EDPRC is preserved by category equivalences too. As a
consequence, if two algebraizable logics ` and `0 have categorically equivalent algebraic
semantics, then ` has a DDT if and only if so does `0 . This kind of observations have been
exploited, for instance, in [62, 63]. 

5.3 Sketches of structure theory


We shall now review some basic properties of quasi-varieties with EDPRC.

Definition 5.30. A quasi-variety K is said to be relatively congruence distributive when


ConK ( A) is a distributive lattice, for every A ∈ K. When K is a variety, we drop the adverb
“relatively” and say that K is congruence distributive.

The following result was established in [87].

Theorem 5.31 (Köhler & Pigozzi). Quasi-varieties with EDPRC are relatively congruence
distributive.

Proof. Let K be a quasi-variety with EDPRC and consider an algebra A ∈ K. By Proposition


5.19, the lattice ConK ( A) is algebraic. Since the proof of Proposition 5.17 shows that every
algebraic lattice B is isomorphic to I(C( B)), we obtain

ConK ( A) ∼
= I(C(ConK ( A))).

Therefore, it will be enough to show that the lattice I(C(ConK ( A))) is distributive.
We will prove a more general result, namely, that if B is a dually Brouwerian semilattice,
then I( B) is a distributive lattice. Since C(ConK ( A)) is a dually Brouwerian semilattice by
Theorem 5.23, this will conclude the proof.
Accordingly, let B be a dually Brouwerian semilattice. As B has a minimum (namely,
a ← a, for any a ∈ B), the poset of ideals I( B) is a well-defined lattice. To prove that it is
distributive, it suffices to show that, for every I, J, L ∈ I( B),

( I ∨ J ) ∩ ( I ∨ L ) ⊆ I ∨ ( J ∩ L ),

where the join ∨ is computed in I( B). To this end, consider an element a ∈ ( I ∨ J ) ∩ ( I ∨ L).
Then there are b ∈ I, c ∈ J and d ∈ L such that

a 6 b ∗ c and a 6 b ∗ d.

As B is dually Brouwerian, this implies

a ← b 6 c and a ← b 6 d,

74
5.3. Sketches of structure theory

whence a ← b ∈ J ∩ L. Since b ∈ I, this yields

b ∗ ( a ← b ) ∈ I ∨ ( J ∩ L ).

As B is dually Brouwerian, from a ← b 6 a ← b it follows a 6 b ∗ ( a ← b). Together with


the above display, this implies a ∈ I ∨ ( J ∩ L), as desired. 
Relative congruence distributivity has a number interesting consequences related to
axiomatization problems.

Definition 5.32. A quasi-variety is said to be finitely based when it can be axiomatized by


a finite set of quasi-equations.

Remark 5.33. In view of the Compactness Theorem of first order logic, if a variety is finitely
based as a quasi-variety, then it can be axiomatized by a finite set of equations too. 
Given a finite set K of finite algebras of finite type, there is no guarantee that V(K)
or Q(K) are finitely based. For varieties, this is known since [92]. In fact a transparent
characterization of the finite algebras A of finite type for which V( A) is finitely based
seems out of reach, in part because the problem of determining whether V( A) is finitely
based is undecidable [96]. This makes the following result appealing [110] .

Theorem 5.34 (Pigozzi). Let K be a finite set of finite algebras of finite type. If Q(K) is relatively
congruence distributive, then it is finitely based.

Remark 5.35. The above result subsumes Baker’s Finite Basis Theorem [6] for varieties.
Both results were generalized in [109] and [7]. 
In view of Theorem 5.31, we obtain the following.

Corollary 5.36. Let K be a finite set of finite algebras of finite type. If Q(K) has EDPRC, then it
is finitely based.

This result admits a logical reading. A logic ` is said to be finitely axiomatizable if it can be
axiomatized by a finite Hilbert calculus.

Corollary 5.37. Let ` be an algebraizable logic of finite type and K its equivalent algebraic
semantics. If ` has a DDT and K = Q(M) for a finite set of finite algebras M, then ` is finitely
axiomatizable.

Proof. From Theorem 5.7 it follows that K has EDPRC. Therefore, we can apply Corollary
5.36, obtaining that K is axiomatized by a finite set Σ of quasi-equations. Let τ and ∆ be
the sets of equations and formulas that, together with K, witness the algebraizability of `.
It is easy to show that ` can be axiomatized by the Hilbert calculus consisting of the rules

x  ∆[τ ( x )]
∆[τ ( x )]
∆( ϕ1 , ψ1 ) ∪ · · · ∪ ∆( ϕn , ψn )  ∆(ε, δ),

for every quasi-equation ( ϕ1 ≈ ψ1 & . . . & ϕn ≈ ψn ) =⇒ ε ≈ δ in Σ. 

75
5. D EDUCTION THEOREMS

Exercise 5.38. Prove that the Hilbert calculus in the above proof axiomatizes `. 

Example 5.39 (Tabular superintuitionistic logics). A superintuitionistic logic is said to be tab-


ular if its equivalent algebraic semantics is a variety of the form V( A), for a finite Heyting
algebra A. We will prove that every tabular intermediate logic is finitely axiomatizable. Ac-
cordingly, let ` be tabular and V( A) its equivalent algebraic semantics (where A is a finite
Heyting algebra). In view of Jónsson’s Lemma [84], the subdiretcly irreducible members
of V( A) form a finite set K of finite algebras. Therefore, V( A) = PSD (K) ⊆ Q(K). As the
inclusion Q(K) ⊆ V( A) is obvious, we conclude that V( A) = Q(K). Since ` has a DDT
(Exercise 5.2), we can apply Corollary 5.37, obtaining that ` is finitely axiomatizable. 

Exercise 5.40. Prove that if K is a finite set of finite algebras, there exists a finite algebra
A such that V(K) = V( A). On the other hand, exhibit a finite set K of finite algebras for
which there is no algebra A (finite or infinite) such that Q(K) = Q( A). 
Theorem 5.31 can be used to disprove that certain quasi-varieties have EDPRC.

Example 5.41 (Groups). As an exemplification, we will prove that the variety Gr of groups
lacks EDPC. Let A be the Klein four-group, i.e., the direct product Z2 × Z2 , where Z2 is the
groups of integers modulo two. The congruence lattice of A is isomorphic to the following
nondistributive diamond.

• • •


Hence, the variety Gr is not congruence distributive. By 5.31, this implies that Gr lacks
EDPC, as desired. In view of the bridge theorem between the DDT and EDPRC, the
algebraizable logic of groups G, defined in the proof of Corollary 4.30, lacks any DDT. 

Principal K-congruences (and, therefore, EDPRC) are also related to the building blocks
of quasi-varieties, as we proceed to explain.

Definition 5.42. Let K be a quasi-variety. A member A of K is said to be subdirectly


irreducible relative to K when for every subdirect emebedding f : A → ∏i∈ I Bi with { Bi :
i ∈ I } ⊆ K, there exists some i ∈ I such that the composition pi ◦ f : A → Bi is an
isomorphism. The class of all subdirectly irreducible algebras relative to K will be denoted
by KRSI and its elements are called the RSI members of K.

An algebra A is said to be subdirectly irreducible (in the absolute sense) when it is subdirectly
irreducible relative to the quasi-variety of all algebras of its type.
The following classical result was discovered by Birkhoff in the setting of varieties and
generalized to quasi-varieties by Maltsev [68, Thm. 3.1.1].

Subdirect Decomposition Theorem 5.43. If K is a quasi-variety, then K = IPSD (KRSI ).

76
5.3. Sketches of structure theory

As we mentioned, the RSI members of a quasi-variety K admit a description in terms


of principal K-congruences.

Proposition 5.44. Let K be a quasi-variety and A ∈ K. The following conditions are equivalent:

(i) A is an RSI member of K;

(ii) id A is completely meet irreducible in ConK ( A);

(iii) There are distinct a, b ∈ A such that h a, bi ∈ CgKA (c, d), for every pair of distinct c, d ∈ A.

Exercise 5.45. Prove the above result. Hint: you might wish to use Proposition 1.25. 
A special class of RSI algebras is the following.

Definition 5.46. Let K be a quasi-variety. An algebra A ∈ K is simple relative to K if it has


exactly two K-congruences. The class of all simple algebras relative to K will be denoted
by KRS and its elements are called the RS members of K.

Corollary 5.47. If K is a quasi-variety, then KRS ⊆ KRSI .

Proof. Let A be an RS member of K. Then ConK ( A) is the two-element chain with mini-
mum id A and maximum A × A. Therefore, id A is completely meet irreducible in ConK ( A).
By Proposition 5.44, we conclude that A is also RSI. 
As for the case of subdirect irreducibility, the notion of a simple algebra admits an
absolute variant. More precisely, an algebra A is simple (in the absolute sense) if Con( A)
has precisely two elements.
As we will see, the definability of principal K-congruences influences the model
theoretic properties of the class of RSI members of a quasi-variety K. To make this precise,
it is convenient to introduce the following concept.

Definition 5.48. A quasi-variety K is said to have definable principal relative congruences


(DPRC) if there exists a first order formula φ( x, y, z, v) such that, for every A ∈ K and
a, b, c, d ∈ A,
h a, bi ∈ CgKA (c, d) ⇐⇒ A  φ(c, d, a, b).
When K is a variety, it is common to use the expression definable principal congruences
(DPC), as opposed to (DPRC).

Clearly, every quasi-variety with EDPRC has DPRC, while the converse need not be
true in general. For instance, an argument analogous to the one detailed in Example 5.41
shows that the variety CR of commutative rings with unit lacks EDPC. On the other hand,
the formula
φ( x, y, z, v) := ∃w(w( x − y) ≈ z − v)
witnesses DPC for CR.
A class of structures is said to be elementary when it can be axiomatized by a set of
sentences.

Theorem 5.49. If K is a quasi-variety with DPRC, then KRSI and KRS are elementary classes.

77
5. D EDUCTION THEOREMS

Proof. We detail the proof for the case of KRSI and leave that of KRS as an exercise. Let
φ( x, y, z, v) be the formula witnessing DPRC for K. Then consider the sentence
ϕ := ∃ x, y( x 6≈ y & ∀z, v(z 6≈ v → φ(z, v, x, y))). (5.7)
Let also Σ be the set of universal the closures of the quasi-equations valid in K. We will
prove that Σ ∪ { ϕ} axiomatizes KRSI .
In view of Maltsev’s Theorem, Σ axiomatizes K. Therefore, as ϕ witnesses DPRC,
Σ ∪ { ϕ} axiomatizes the members A of K that contain two distinct a, b such that h a, bi ∈
CgKA (c, d), for every pair of distinct c, d ∈ A. In view of Proposition 5.44, these are precisely
the RSI members of K. 
As a consequence of Löwenheim-Skolem Theorem, we obtain the following.
Corollary 5.50. Let K be a quasi-variety with DPRC whose language has cardinality κ. If K has
an infinite RSI member, then it also has an RSI member of cardinality λ, for every λ > κ + ℵ0 .
Exercise 5.51. Complete the proof of Theorem 5.49, by showing that if a quasi-variety K
has DPRC, then KRS is elementary. Prove also that if K has EDPRC, when extended with
all trivial algebras of the suitable type, KRS becomes a universal class. 
Theorem 5.31 can be used to disprove that certain quasi-varieties have EDPRC.
Example 5.52 (Modal algebras). As an exemplification, we will prove that the variety MA
of modal algebras lacks DPC and, therefore, EDPC. Notably, in view of Theorem 5.7, this
implies that the logic Kg lacks any DDT. By Theorem 5.49, it suffices to prove that the
class of simple modal algebras is not elementary. We will do this, by showing that it is
not closed under ultraproducts (see Corollary 3.6, if necessary). To this end, recall that a
modal algebra A is simple when it is nontrivial and for every element a ∈ A r {1} there
is n ∈ N such that n a = 0.
For every n ∈ N, consider the Kripke frame hWn , Rn i with universe Wn = { x1 , . . . , xn }
and accessibility relation Rn defined as follows:
hy, zi ∈ Rn ⇐⇒ either y = z or (y = xi and z = xi+1 ) or (y = xn and z = x1 ).
Furthermore, let An be the complex algebra of hWn , Rn i. It is easy to prove that An is
simple. Furthermore, for every n ∈ N, there exists an element an ∈ An+2 r {1} such that
nAn+2 an 6= 0 An+2 .
Suppose, with a view to contradiction, that the class simple modal algebras is closed
under ultraproducts. Then consider a free ultrafilter U on N and take the ultrapower
B := ∏ An /θU .
n ∈N

By assumption, B is a simple modal algebra. Then consider an element ~b ∈ ∏n∈N An such


that ~b(n + 2) = an , for every n ∈ N. Since B is simple, there is m ∈ N such that
mB~b/θU = 0B .
By Łoś’ Theorem, this implies that U contains a finite subset of N. But this contradicts the
assumption that U is free. 

78
5.3. Sketches of structure theory

At this stage, it is natural to wonder how restrictive is EDPRC with respect to DPRC.
To answer this question, it is convenient to introduce the following concept.

Definition 5.53. A quasi-variety K is said to have the relative congruence extension property
(RCEP) if for every B 6 A ∈ K and θ ∈ ConK ( B) there exists φ ∈ ConK ( A) such that
θ = φ ∩ ( B × B). When K is a variety, ConK ( A) and ConK ( B) can be replaced, respectively,
by Con( A) and Con( B), and K is said to have the congruence extension property (CEP).

The following result was established in [60].

Theorem 5.54 (Fried, Grätzer & Quackenbush). A quasi-variety has EDPRC if and only if it is
relatively congruence distributive and it has DPRC and the RCEP.

As the majority of quasi-varieties in the algebra of logic have the RCEP and are relatively
congruence distributive, in this setting it is often that case that EDPRC is equivalent to
DPRC. For instance, as all varieties of modal algebras have the CEP and are congruence
distributive, EDPC and DPC coincide for them.
We conclude our journey with a description of varieties of modal algebras with EDPC.

Theorem 5.55 (Blok & Pigozzi). A variety of modal algebras has EDPC if and only if it is weakly
transitive.

Proof. Recall from Exercise 5.6 that weakly transitive varieties have EDPC. Therefore, it
only remains to prove the converse. Accordingly consider variety K of modal algebras
with EDPC. By Theorem 4.13, it is the equivalent algebraic semantics of an axiomatic
extension ` of Kg . As any other axiomatic extension of Kg , the logic ` has the following
property: for every Γ ∪ { ϕ, ψ} ⊆ T (Var ),

Γ, ϕ ` ψ ⇐⇒ there is n ∈ N such that Γ ` n ϕ → ψ. (5.8)

Furthermore, ` has a DDT, by Theorem 5.7. Let then I ( x, y) be the set witnessing
the DDT for `. Since I is finite, we can assume that it has cardinality at most one, for
if I = { ϕ1 , . . . , ϕn }, we can replace I with I ∗ = { ϕ1 ∧ · · · ∧ ϕn }. Furthermore, we can
also assume that I contains a tautology and, therefore, that it is nonempty. Accordingly,
we will assume that I = { x ( y}, for some formula x ( y in two variables x and y.
Therefore, the DDT amounts to the statement that, for every set of formulas Γ ∪ { ϕ, ψ},

Γ, ϕ ` ψ ⇐⇒ Γ ` ϕ ( ψ.

Now, observe that x ( y ` x ( y. In view of the DDT, this yields x, x ( y ` y. By


(5.8) there exists n ∈ N such that

∅ ` n ( x ∧ ( x ( y)) → y.

Observe that in the class of modal algebras MA the equation n ( x ∧ y) ≈ n x ∧ n y holds.


Therefore,
   
MA  n ( x ∧ ( x ( y)) → y → (n x ∧ n ( x ( y)) → y ≈ 1.

79
5. D EDUCTION THEOREMS

By Theorem 2.15, this yields


   
∅ `Kg n ( x ∧ ( x ( y)) → y → (n x ∧ n ( x ( y)) → y . (5.9)

Since ` is an extension of Kg , by applying modus ponens to the above display and (5.9),
we obtain
∅ ` (n x ∧ n ( x ( y)) → y.
Recall that in CPC every formula of the form ((α ∧ β) → γ) → ( β → (α → γ)) is
a tautology. Therefore, the same is true in `. As the formula above have the form
(α ∧ β) → γ, by modus ponens we obtain

n ( x ( y) ` n x → y.

As ` in substitution invariant, we can replace y by n+1 x in the above deduction and


obtain
n ( x ( n+1 x ) ` n x → n+1 x. (5.10)
Lastly, since ` is an extension of Kg , we have x ` n+1 x. By the DDT, this yields
∅ ` x ( n+1 x. Together with x ` x and substitution invariance, this yields

∅ ` n ( x ( n +1 x ).

By (5.10), we conclude that


∅ ` n x → n+1 x.
Since the algebraizability of ` is witnessed by τ = { x ≈ 1}, ∆ = { x ↔ y} and K, from
the above display it follows K  n x → n+1 x ≈ 1, that is, K  n x 6 n+1 x. Hence, we
conclude that Kg is weakly transitive. 
Corollary 5.56. An axiomatic extension of Kg has the DDT if and only if its equivalent algebraic
semantics is a weakly transitive variety of modal algebras.

Remark 5.57. EDPRC can also be used to formulate the theory of Jankov’s formulas [77, 78, 79]
in universal algebraic terms [19]. This is made possible by the lattice theoretic notion of a
splitting [128] and its application to subvariety lattices [94]. 
Remark 5.58. In this course, we focused on the bridge theorem connecting the DDT with
EDPRC. Similar theorems holds for a variety of other metalogical properties, including
weaker variants of the DDT [22, 40, 41, 45, 114] and of the proof by cases [39, 44, 108],
interpolation [47], Beth definability [14, 75, 76], inconsistency lemmas [31, 111, 116] and
relevance principles [50, 91, 106]. For information of the algebraic counterparts of these
properties, the reader might consult [86], see also [99]. 

80
Bibliography

[1] J. Adámek. How many variables does a quasivariety need? Algebra Universalis,
(27):44–48, 1990.

[2] P. Aglianò and A. Ursini. Ideals and other generalizations of congruence classes. J.
Austral. Math. Soc. Ser. A, 53(1):103–115, 1992.

[3] P. Aglianò and A. Ursini. On subtractive varieties II: general properties. Algebra
Universalis, 36:222–259, 1996.

[4] P. Aglianò and A. Ursini. On subtractive varieties III: from ideals to congruences.
Algebra Universalis, 37(3):296–333, 1997.

[5] P. Aglianò and A. Ursini. On subtractive varieties IV: definability of principal ideals.
Algebra Universalis, 38(3):355–389, 1997.

[6] K. Baker. Finite equational bases for finite algebras in a congruence distributive
class. Advances in Mathematics, 24:207–243, 1977.

[7] K. A. Baker and J. Wang. Definable principal subcongruences. Algebra Universalis,


47:145–151, 2002.

[8] J. T. Baldwin and J. Berman. The number of subdirectly irreducible algebras in a


variety. Algebra Universalis, 5(3):379–389, 1975.

[9] J. T. Baldwin and J. Berman. A model theoretic approach to Malcev conditions. The
Journal of Symbolic Logic, 42(2):277–288, 1977.

[10] J. L. Bell and A. B. Slomson. Models and ultraproducts: An introduction. North-Holland,


Amsterdam, 1971. Second revised printing.

[11] C. Bergman. Structural completeness in algebra and logic. In Algebraic logic (Budapest,
1988), volume 54 of Colloq. Math. Soc. János Bolyai, pages 59–73. North-Holland,
Amsterdam, 1991.

[12] C. Bergman. Universal Algebra: Fundamentals and Selected Topics. Chapman & Hall
Pure and Applied Mathematics. Chapman and Hall/CRC, 2011.

81
B IBLIOGRAPHY

[13] N. Bezhanishvili and T. Moraschini. Citkin’s description of hereditarily structurally


complete intermediate logics via Esakia duality. Submitted manuscript, available online,
2020.

[14] W. J. Blok and E. Hoogland. The Beth property in Algebraic Logic. Studia Logica,
83(1–3):49–90, 2006.

[15] W. J. Blok and B. Jónsson. Algebraic structures for logic. A course given at the 23rd
Holiday Mathematics Symposium, New Mexico State University, 1999.

[16] W. J. Blok and B. Jónsson. Equivalence of consequence operations. Studia Logica,


83(1–3):91–110, 2006.

[17] W. J. Blok and P. Köhler. Algebraic semantics for quasi-classical modal logics. The
Journal of Symbolic Logic, 48:941–964, 1983.

[18] W. J. Blok, P. Köhler, and D. Pigozzi. On the structure of varieties with equationally
definable principal congruences II. Algebra Universalis, 18:334–379, 1984.

[19] W. J. Blok and D. Pigozzi. On the structure of varieties with equationally definable
principal congruences I. Algebra Universalis, 15:195–227, 1982.

[20] W. J. Blok and D. Pigozzi. Protoalgebraic logics. Studia Logica, 45:337–369, 1986.

[21] W. J. Blok and D. Pigozzi. Algebraizable logics, volume 396 of Mem. Amer. Math. Soc.
A.M.S., Providence, January 1989.

[22] W. J. Blok and D. Pigozzi. Local deduction theorems in algebraic logic. In


H. Andréka, J. D. Monk, and I. Németi, editors, Algebraic Logic, volume 54 of
Colloq. Math. Soc. János Bolyai, pages 75–109. North-Holland, Amsterdam, 1991.

[23] W. J. Blok and D. Pigozzi. Algebraic semantics for universal Horn logic without
equality. In A. Romanowska and J. D. H. Smith, editors, Universal Algebra and
Quasigroup Theory, pages 1–56. Heldermann, Berlin, 1992.

[24] W. J. Blok and D. Pigozzi. On the structure of varieties with equationally definable
principal congruences III. Algebra Universalis, 32:545–608, 1994.

[25] W. J. Blok and D. Pigozzi. On the structure of varieties with equationally definable
principal congruences IV. Algebra Universalis, 31:1–35, 1994.

[26] W. J. Blok and D. Pigozzi. Abstract algebraic logic and the deduction theorem. Avail-
able in internet http: // orion. math. iastate. edu/ dpigozzi/ , 1997. Manuscript.

[27] W. J. Blok and J. G. Raftery. Assertionally equivalent quasivarieties. International


Journal of Algebra and Computation, 18(4):589–681, 2008.

[28] W. J. Blok and J. Rebagliato. Algebraic semantics for deductive systems. Studia
Logica, Special Issue on Abstract Algebraic Logic, Part II, 74(5):153–180, 2003.

82
Bibliography

[29] S. Burris. Boolean constructions. In Universal algebra and lattice theory (Puebla, 1982),
volume 1004 of Lecture Notes in Mathematics, pages 67–90. Springer, Berlin, 1983.

[30] S. Burris and H. P. Sankappanavar. A course in Universal Algebra. Available on-


line https://www.math.uwaterloo.ca/~snburris/htdocs/ualg.html, the millen-
nium edition, 2012.

[31] M. A. Campercholi and J. G. Raftery. Relative congruence formulas and decomposi-


tions in quasivarieties. Algebra Universalis, 78(3):407–425, 2017.

[32] E. Casanovas, P. Dellunde, and R. Jansana. On elementary equivalence for equality-


free logic. Notre Dame Journal of Formal Logic, 37(3):506–522, 1996.

[33] A. Chagrov and M. Zakharyaschev. Modal Logic, volume 35 of Oxford Logic Guides.
Oxford University Press, 1997.

[34] C. C. Chang and H. J. Keisler. Model Theory, volume 73 of Studies in Logic. North-
Holland, Amsterdam, third edition, 1990.

[35] P. Cintula and C. Noguera. Logic and Implication: An Introduction to the General
Algebraic Study of Non-Classical Logics. Trends in Logic. Springer, To appear.

[36] A. Citkin. On structurally complete superintuitionistic logics. Soviet Mathematics


Doklady, 19:816–819, 1978.

[37] A. Citkin. Structurally complete superintuitionistic logics and primitive varieties of


pseudoBoolean algebras. Mat. Issled. Neklass. Logiki, 98:134–151, 1987.

[38] J. Czelakowski. Equivalential logics, I, II. Studia Logica, 40:227–236 and 355–372,
1981.

[39] J. Czelakowski. Filter distributive logics. Studia Logica, 43:353–377, 1984.

[40] J. Czelakowski. Algebraic aspects of deduction theorems. Studia Logica, 44:369–387,


1985.

[41] J. Czelakowski. Local deductions theorems. Studia Logica, 45:377–391, 1986.

[42] J. Czelakowski. Protoalgebraic logics, volume 10 of Trends in Logic—Studia Logica


Library. Kluwer Academic Publishers, Dordrecht, 2001.

[43] J. Czelakowski. Equivalential logics (after 25 years of investigations). Reports on


Mathematical Logic, (38):23–36, 2003.

[44] J. Czelakowski and W. Dziobiak. Congruence distributive quasivarieties whose


finitely subdirectly irreducible members form a universal class. Algebra Universalis,
27(1):128–149, 1990.

[45] J. Czelakowski and W. Dziobiak. A deduction theorem schema for deductive systems
of propositional logics. Studia Logica, Special Issue on Algebraic Logic, 50:385–390,
1991.

83
B IBLIOGRAPHY

[46] J. Czelakowski and R. Jansana. Weakly algebraizable logics. The Journal of Symbolic
Logic, 65(2):641–668, 2000.
[47] J. Czelakowski and D. Pigozzi. Amalgamation and interpolation in abstract algebraic
logic. In X. Caicedo and Carlos H. Montenegro, editors, Models, algebras and proofs,
number 203 in Lecture Notes in Pure and Applied Mathematics Series, pages 187–
265. Marcel Dekker, New York and Basel, 1999.
[48] P. Dellunde and R. Jansana. Some characterization theorems for infinitary universal
Horn logic without equality. The Journal of Symbolic Logic, 61(4):1242–1260, 1996.
[49] J. J. Dukarm. Morita equivalence of algebraic theories. Colloquium Mathematicum,
55:11–17, 1988.
[50] W. Dziobiak, A. V. Kravchenko, and P. J. Wojciechowski. Equivalents for a quasiva-
riety to be generated by a single structure. Studia Logica, 91(1):113–123, 2009.
[51] R. Elgueta. Characterizing classes defined without equality. Studia Logica, 58(3):357–
394, 1997.
[52] R. Elgueta and R. Jansana. Definability of Leibniz equality. Studia Logica, 63:223–243,
1999.
[53] L. Esakia. Topological Kripke models. Soviet Math. Dokl., 15:147–151, 1974.
[54] L. Esakia. Heyting Algebras. Duality Theory. Springer, English translation of the
original 1985 book. 2019.
[55] J. M. Font. Abstract Algebraic Logic - An Introductory Textbook, volume 60 of Studies in
Logic - Mathematical Logic and Foundations. College Publications, London, 2016.
[56] J. M. Font and R. Jansana. A general algebraic semantics for sentential logics, volume 7
of Lecture Notes in Logic. A.S.L., Second edition edition, 2017. Available online at
projecteuclid.org/euclid.lnl/1235416965.
[57] J. M. Font, R. Jansana, and D. Pigozzi. A survey on abstract algebraic logic. Studia
Logica, Special Issue on Abstract Algebraic Logic, Part II, 74(1–2):13–97, 2003. With an
“Update” in 91 (2009), 125–130.
[58] J. M. Font and V. Verdú. Algebraic logic for classical conjunction and disjunction.
Studia Logica, Special Issue on Algebraic Logic, 50:391–419, 1991.
[59] P. Freyd. Algebra valued functors in general and tensor products in particular.
Colloq. Math., 14:89–106, 1966.
[60] E. Fried, G. Grätzer, and R. Quackenbush. Uniform congruence schemes. Algebra
Universalis, 10:176–189, 1980.
[61] N. Galatos and J. Gil-Férez. Modules over quantaloids: Applications to the iso-
morphism problem in algebraic logic and π-institutions. Journal of Pure and Applied
Algebra, 221(1):1–24, January 2017.

84
Bibliography

[62] N. Galatos and J. G. Raftery. A category equivalence for odd sugihara monoids and
its applications. Journal of Pure and Applied Algebra, 216(10):2177–2192, 2012.

[63] N. Galatos and J. G. Raftery. Idempotent residuated structures: some category


equivalences and their applications. Trans. Amer. Math. Soc., 367:3189–3223, 2015.

[64] N. Galatos and C. Tsinakis. Equivalence of consequence relations: an order-theoretic


and categorical perspective. The Journal of Symbolic Logic, 74(3):780–810, 2009.

[65] O. C. Garcı́a and W. Taylor. The lattice of interpretability types of varieties, volume 50.
Mem. Amer. Math. Soc., 1984.

[66] A. J. Gil and J. Rebagliato. Protoalgebraic Gentzen systems and the cut rule. Studia
Logica, Special Issue on Abstract Algebraic Logic, 65(1):53–89, 2000.

[67] V. I. Glivenko. Sur quelques points de la logique de M. Brouwer. Academie Royal de


Belgique Bulletin, 15:183–188, 1929.

[68] V. A. Gorbunov. Algebraic theory of quasivarieties. Siberian School of Algebra and


Logic. Consultants Bureau, New York, 1998. Translated from the Russian.

[69] G. Grätzer and E. T. Schmidt. Characterizations of congruence lattices of abstract


algebras. Acta Sci. Math. (Szeged), 24:34–59, 1963.

[70] A. Grzegorczyk. Undecidability of some topological theories. Fundamenta Mathe-


maticae, 38:137–152, 1951.

[71] H. P. Gumm and A. Ursini. Ideals in universal algebras. Algebra Universalis, 19(1):45–
54, 1984.

[72] B. Herrmann. Algebraizability and Beth’s theorem for equivalential logics. Bulletin
of the Section of Logic, 22(2):85–88, 1993.

[73] B. Herrmann. Equivalential and algebraizable logics. Studia Logica, 57:419–436, 1996.

[74] B. Herrmann. Characterizing equivalential and algebraizable logics by the Leibniz


operator. Studia Logica, 58:305–323, 1997.

[75] E. Hoogland. Algebraic characterizations of various Beth definability properties.


Studia Logica, Special Issue on Abstract Algebraic Logic, 65:91–112, 2000.

[76] J. R. Isbell. Epimorphisms and dominions, chapter of Proceedings of the Conference on


Categorical Algebra, La Jolla, California, 1965, pages 232–246. Springer, New York,
1966.

[77] V. A. Jankov. On the relation between deducibility in intuitionistic propositional


calculus and finite implicative structures. Doklady Akademii Nauk SSSR, 151:1293–
1294, 1963.

[78] V. A. Jankov. Three sequences of formulas with two variables in positive proposi-
tional logic. Izv. Akad. Nauk SSSR Ser. Mat., 32:880–883, 1968.

85
B IBLIOGRAPHY

[79] V. A. Jankov. Conjunctively irresolvable formulae in propositional calculi. Izvestiya


Akademii Nauk SSSR. Seriya Matematicheskaya, 33:18–38, 1969.

[80] R. Jansana and T. Moraschini. The poset of all logics I: interpretations and lattice
structure. The Journal of Symbolic Logic, 2021. To appear.

[81] R. Jansana and T. Moraschini. The poset of all logics II: Leibniz classes and hierarchy.
The Journal of Symbolic Logic, 2021. To appear.

[82] R. Jansana and T. Moraschini. The poset of all logics III: Finitely presentable logic.
Studia Logica, 109:49–76, 2021.

[83] S. Jaśkowski. Recherches sur le systeme de la logique intuitioniste. In Actes du


Congrès International des Philosophie Scientifique, Paris, volume VI, pages 58–61, 1936.

[84] B. Jónsson. Algebras whose congruence lattices are distributive. Mathematica


Scandinavica, 21:110–121, 1967.

[85] K. A. Kearnes and E. W. Kiss. The shape of congruences lattices, volume 222 of Mem.
Amer. Math. Soc. American Mathematical Society, 2013. Monograph.

[86] E. W. Kiss, L. Márki, P. Pröhle, and W. Tholen. Categorical algebraic properties.


a compendium on amalgamation, congruence extension, epimorphisms, residual
smallness, and injectivity. Studia Sci. Math. Hungar., 18:79–141, 1983.

[87] P. Köhler and D. Pigozzi. Varieties with equationally definable principal congru-
ences. Algebra Universalis, 11:213–219, 1980.

[88] T. Kowalski, F. Paoli, and M. Spinks. Quasi-subtractive varieties. The Journal of


Symbolic Logic, 76(4):1261–1286, 2011.

[89] M. Kracht. Tools and techniques in modal logic, volume 142 of Studies in Logic and the
Foundations of Mathematics. North-Holland Publishing Co., Amsterdam, 1999.

[90] M. Kracht. Modal consequence relations, volume 3, chapter 8 of the Handbook of


Modal Logic. Elsevier Science Inc., New York, NY, USA, 2006.

[91] J. Łoś and R. Suszko. Remarks on sentential logics. Indagationes Mathematicae,


20:177–183, 1958.

[92] R. C. Lyndon. Identities in finite algebras. Proceedings of the Americal Mathematical


Society, 5:8–9, 1954.

[93] A. I. Mal’cev. The metamathematics of algebraic systems, collected papers: 1936-1967.


Amsterdam, North-Holland Pub. Co., 1971.

[94] R. McKenzie. Equational bases and nonmodular lattice varieties. Transactions of the
Americal Mathematical Society, 174:1–43, 1972.

86
Bibliography

[95] R. McKenzie. An algebraic version of categorical equivalence for varieties and more
general algebraic categories. In P. Aglianò and R. Magari, editors, Logic and algebra,
volume 180 of Lecture Notes in Pure and Appl. Math., pages 211–243. Dekker, New
York, 1996.

[96] R. McKenzie. Tarski’s finite basis problem is undecidable. International Journal of


Algebra and Computation, 6(1):49–104, 1996.

[97] J. C. C. McKinsey. A solution of the decision problem for the Lewis systems S2 and
S4, with an application to topology. The Journal of Symbolic Logic, 6:117–134, 1941.

[98] J. C. C. McKinsey and A. Tarski. The algebra of topology. Annals of Mathematics,


45:141–191, 1944.

[99] G. Metcalfe, F. Montagna, and C. Tsinakis. Amalgamation and interpolation in


ordered algebras. Journal of Algebra, 402:21–82, 2014.

[100] T. Moraschini. A computational glimpse at the Leibniz and Frege hierarchies. Annals
of Pure and Applied Logic, 169(1):1–20, January 2018.

[101] T. Moraschini. A logical and algebraic characterization of adjunctions between


generalized quasi-varieties. Journal of Symbolic Logic, 83(3):899–919, 2018.

[102] T. Moraschini. A study of the truth predicates of matrix semantics. Review of Symbolic
Logic, 11(4):780–804, 2018.

[103] T. Moraschini. On the complexity of the Leibniz hierarchy. Annals of Pure and Applied
Logic, 170(7):805–824, July 2019.

[104] T. Moraschini. On equational completeness theorems. Journal of Symbolic Logic, 2021.


To appear.

[105] T. Moraschini and J. G. Raftery. On prevarieties of logic. Algebra Universalis, (80),


2019.

[106] T. Moraschini, J. G. Raftery, and J. J. Wannenburg. Singly generated quasivarieties


and residuated stuctures. Mathematical Logic Quarterly, 66(2):150–172, 2020.

[107] W. Neumann. On Mal’cev conditions. Journal of the Australian Mathematical Society,


17:376–384, 1974.

[108] C. Noguera and P. Cintula. The proof by cases property and its variants in structural
consequence relations. Studia Logica, 101:713–747, 2013.

[109] A. M. Nurakunov and M. M. Stronkowski. Quasivarieties with definable relative


principal subcongruences. Studia Logica, 92(1):109–120, 2009.

[110] D. Pigozzi. Finite basis theorem for relatively congruence-distributive quasi-


varieties. Transactions of the Americal Mathematical Society, 310:499–533, 1988.

87
B IBLIOGRAPHY

[111] A. Přenosil and T. Lávička. Semisimplicity, the excluded middle, and glivenko
theorems. Available online, 2020.

[112] J. G. Raftery. Correspondences between Gentzen and Hilbert systems. The Journal of
Symbolic Logic, 71(3):903–957, 2006.

[113] J. G. Raftery. The equational definability of truth predicates. Reports on Mathematical


Logic, (41):95–149, 2006.

[114] J. G. Raftery. Contextual deduction theorems. Studia Logica, 99(1):279–319, 2011.

[115] J. G. Raftery. A perspective on the algebra of logic. Quaestiones Mathematicae,


34:275–325, 2011.

[116] J. G. Raftery. Inconsistency lemmas in algebraic logic. Mathematical Logic Quaterly,


59(6):393–406, 2013.

[117] J. G. Raftery. Order algebraizable logics. Annals of Pure and Applied Logic, 164(3):251–
283, 2013.

[118] J. G. Raftery. Admissible Rules and the Leibniz Hierarchy. Notre Dame Journal of
Formal Logic, 57(4):569–606, 2016.

[119] H. Rasiowa. An algebraic approach to non-classical logics, volume 78 of Studies in Logic


and the Foundations of Mathematics. North-Holland, Amsterdam, 1974.

[120] J. Rebagliato and V. Verdú. Algebraizable Gentzen systems and the deduction
theorem for Gentzen systems. Mathematics Preprint Series 175, University of
Barcelona, June 1995.

[121] G. Reyes and H. Zolfaghari. Bi-Heyting Algebras, Toposes and Modalities. Journal
of Philosophical Logic, 25(1):25–43, 1996.

[122] V. V. Rybakov. Hereditarily structurally complete modals logics. The Journal of


Symbolic Logic, 60(1):266–288, March 1995.

[123] V. V. Rybakov. Admissibility of logical inference rules, volume 136 of Studies in Logic.
Elsevier, Amsterdam, etc., 1997.

[124] T. Skolem. Logisch-kombinatorische untersuchungen über die erfüllbarkeit und


beweisbarkeit mathematischen sätze nebst einem theoreme über dichte mengen.
Videnskabsakademiet i Kristiania, Skrifter I, (4):1–36, 1920.

[125] W. Taylor. Characterizing Mal’cev conditions. Algebra Universalis, 3:351–397, 1973.

[126] A. Ursini. On subtractive varieties I. Algebra Universalis, 31:204–222, 1994.

[127] A. Ursini. On subtractive varieties. V. Congruence modularity and the commutators.


Algebra Universalis, 43(1):51–78, 2000.

[128] P. Whitman. Splittings of a lattice. American Journal of Mathematics, 65:179–196, 1943.

88
Bibliography

[129] R. Wójcicki. Theory of logical calculi. Basic theory of consequence operations, volume 199
of Synthese Library. Reidel, Dordrecht, 1988.

[130] F. Wolter. On logics with coimplication. Journal of Philosophical Logic, 27:353–387,


1998.

[131] A. Wroński. On cardinalities of matrices strongly adequate for the intuitionistic


propositional logic. Reports on Mathematical Logic, 2:55–62, 1974.

89

You might also like