LATTICES OF ATOMIC THEORIES IN LANGUAGES

WITHOUT EQUALITY

TRISTAN HOLMES, DAYNA KITSUWA, J. B. NATION AND SHERI TAMAGAWA

Abstract. The structure of lattices of atomic theories in languages

without equality is described. In particular, these lattices are distribu-
tive, and both algebraic and dually algebraic.

1. Introduction
Lattices of equational theories of algebraic structures have been a rich
topic of investigation, both for specific varieties such as groups, and for
universal algebra. For the most general properties of lattices of equational
theories, see Lampe [7], [8], McKenzie [9], Newrly [13], Nurakunov [14] and
their references.
In this note, we consider lattices of atomic theories in a language that
may not contain equality. This is a continuation of the work begun in
Nation [12]. The emphasis there, however, was on implicational theories,
generalizing results for lattices of quasi-equational theories from Adaricheva
and Nation [1]. See also Gorbunov [6].
Let L be a language without equality. We will construct an ordered set
P such that the lattice of atomic theories in the language L is isomorphic
to the lattice of order ideals of P. Thus the lattice of atomic theories is
distributive, and both algebraic and dually algebraic. We will be concerned
with describing the properties of the ordered sets P that can occur in this
way.

2. Atomic theories
Let us work in a language L that has a countable set of variables X =
{x0 , x1 , x2 , . . . }, constants, function symbols, relation symbols, parentheses
and commas for punctuation, but no primitive equality relation. Constants
are regarded as nullary functions, but assume that L has no nullary relations.
Implicitly, the logic used is boolean, with its standard truth values and
functions.
Recall that terms in L are strings of symbols defined thusly.
(1) Each variable x ∈ X is a term.

Date: March 24, 2010.

This research was supported in part by U. S. Civilian Research and Development
Foundation Grant KYM1-2852-BI-07.
1
2 TRISTAN HOLMES, DAYNA KITSUWA, J. B. NATION AND SHERI TAMAGAWA

(2) Every constant c of the language is a term.

(3) If t1 , . . . , tk are terms and f is a k-ary function symbol of the lan-
guage, then f (t1 , . . . , tk ) is a term.
(4) Only strings obtained by (1)–(3) are terms.
Two terms are equal only if they are identical.
The set of terms, with the operations of L acting in the obvious way, form
the term algebra or absolutely free algebra FL(X). The term algebra is used
often, and will be denoted simply F.
No relations hold on F as an algebraic structure. An atomic formula of
L is an expression R(t1 , . . . , tk ) where R is a k-ary relation symbol of L and
t1 , . . . , tk are terms. The set of atomic formulae of L will be denoted by PL,
or just P.
Since F is an absolutely free algebra, any map σ : X → F can be extended
to a homomorphism. That is, given σ(x) for all x ∈ X, we define σ(c) = c
for constants c, and recursively
σ(f (t1 , . . . , tk )) = f (σ(t1 ), . . . , σ(tk ))
for a k-ary function symbol f . Because each term has a unique expression,
this uniquely defines the extension σ : F → F. We refer to these endo-
morphisms as substitutions, and use Sbn(F) to denote the monoid of all
substitutions.
Generally speaking, a theory is a set of sentences closed under deduction.
In this note, we consider only universally quantified atomic formulae, i.e.,
sentences of the form
∀x1 . . . ∀xm R(t1 (x1 , . . . , xm ), . . . , tk (x1 , . . . , xm ))
with R a relation symbol and t1 , . . . , tk terms. Under the circumstances, we
can suppress the quantification symbols.
The only relevant rule of deduction is then substitution. That is, inference
should be determined solely by the scheme of rules that when R is a relation
symbol, t1 , . . . , tk terms and σ a substitution, then
R(t1 , . . . , tk ) ⊢ R(σ(t1 ), . . . , σ(tk )).
Note that the relation ⊢ defined thusly on the set of atomic formulae P is
a quasi-order, i.e., reflexive and transitive. As usual, we then define two
atomic formulae Φ, Ψ to be equivalent, denoted Φ ≡ Ψ, if Φ ⊢ Ψ and Ψ ⊢ Φ.
Then ≡ is an equivalence relation on P, and ⊢ is a partial order on P/ ≡.
Let P be the ordered set hP/ ≡, ⊢i.
It is not hard to characterize this equivalence: R(s1 , . . . , sk ) ≡ R′ (t1 , . . . , tℓ )
if and only if R = R′ and there is a permutation π of the variables such that
the induced substitution has π(si ) = ti for 1 ≤ i ≤ k = ℓ.
Note in passing that the presence of a primitive equality relation, that
is, a binary relation ≈ that is assumed to be a congruence relation, would
require a more complicated deduction scheme, albeit a familiar one.
 LATTICES OF ATOMIC THEORIES IN LANGUAGES WITHOUT EQUALITY 3

A set of atomic formulae Σ ⊆ P is an atomic theory if whenever R(t1 , . . . , tk ) ∈

Σ and σ ∈ Sbn(F), then R(σ(t1 ), . . . , σ(tk )) ∈ Σ. That is, atomic theories
are just sets of atomic formulae that are closed under substitution, or equiv-
alently, deduction.
Models for atomic theories in languages without equality are discussed in
Blok and Pigozzi [2], Czelakowski [4], Elgueta [5] and Nation [12]. These
would be a digression at this point.

3. Lattices of atomic theories

By general principles, the lattice of all atomic theories of L, ordered by
set inclusion, forms an algebraic lattice ATh(L). Likewise, given an atomic
theory B of L, the set of all theories containing B forms an algebraic lattice
ATh(B). The latter is of course the principal filter ↑ B in ATh(L). The
notation is admittedly ambiguous, but the context should make clear which
is intended.
If Φ and Ψ are atomic formulae with Φ ⊢ Ψ, then for atomic theories T
we have Φ ∈ T implies Ψ ∈ T. Thus it makes sense to interpret ⊢ as ≥ on
P. With that convention, we have these characterizations.

Theorem 1. Let L be a language without equality, and let P = hP/ ≡, ⊢i

as above. Then ATh(L) is isomorphic to the lattice of order ideals O(P).

Theorem 2. Let B be an atomic theory in ATh(L), corresponding to an

order ideal I. Then ATh(B) =↑ B ∼
= O(P − I).
It follows that lattices of atomic theories are distributive, algebraic and
dually algebraic.
Now each atomic formula involves only one relation. So if R denotes the
set of relation symbols of L, then we can write P as a disjoint union P =
Ṡ
R∈RPR , where PR denotes all atomic formulae of the form R(t1 , . . . , tk ).
Ṡ
Correspondingly, P = R∈RPR .

Corollary 3. Each lattice ATh(L) or ATh(B) is isomorphic to a direct

product of the corresponding lattices for a language with only one relation
symbol.

For a fixed R, the greatest element of PR is the equivalence class of

R(x1 , . . . , xk ), each formula in PR being obtained from that one by a substi-
tution. Moreover, up to a permutation of the variables, each atomic formula
R(t1 , . . . , tk ) can be obtained from R(x1 , . . . , xk ) by a sequence of substitu-
tions of the following basic types, perhaps in more than one way.
(1) For variables x, y both appearing in Φ, β(x) = β(y) = x and β(z) =
z for all other z ∈ X.
(2) For a variable x appearing in Φ and a constant c, γ(x) = c and
γ(z) = z for all other z ∈ X.
4 TRISTAN HOLMES, DAYNA KITSUWA, J. B. NATION AND SHERI TAMAGAWA

(3) For a variable x appearing in Φ, a function symbol f , and variables

y1 , . . . , ym not appearing in Φ, δ(x) = f (y1 , . . . , ym ) and δ(z) = z for
all other z ∈ X.
Type (2) is of course technically a special case of type (3). These basic types
of substitutions give the covering relations in PR .
This analysis yields the first simple properties of PR .
Theorem 4. For each relation symbol R, the ordered set PR has these
properties.
(1) PR has a greatest element.
(2) For each p ∈ PR , the filter ↑ p is finite.
An example would serve us well at this point. Let L be the language with
one binary predicate R, one unary function symbol f , and one constant c.
The atomic formulae that can be obtained by 0, 1 or 2 basic substitutions
are
(A0 ) R(x, y) (A9 ) R(c, f (y))
(A1 ) R(f (x), y) (A10 ) R(c, c)
(A2 ) R(c, y) (A11 ) R(f (x), f (x))
(A3 ) R(x, x) (A12 ) R(f (x), f (y))
(A4 ) R(x, c) (A13 ) R(f (x), c)
(A5 ) R(x, f (y)) (A14 ) R(y, f (y))
2
(A6 ) R(f (x), y) (A15 ) R(x, f (c))
(A7 ) R(f (c), y) (A16 ) R(x, f 2 (y))
(A8 ) R(x, x)
The filter of PR consisting of these atomic formulae, ordered by ⊢, is drawn
in Figure 1.

A0

A1 A2 A3 A4 A5

A6 A7 A8 A9 A10 A12 A13 A14 A15 A16

A11

Figure 1. Top of PR

In general, the structure at the top of PR is more restricted than indicated

by Theorem 4. Define the depth of an element b in an ordered set with 1 to
be the shortest length of a maximal chain from b to 1.
 LATTICES OF ATOMIC THEORIES IN LANGUAGES WITHOUT EQUALITY 5

Theorem 5. Let R be a relation symbol in a language L. If b is an element

of depth two in PR , then the interval [b, 1] is isomorphic to one of the ordered
sets in Figure 2.
Note that each of these possibilities occurs in our example. The proof is
by considering all the ways we can construct an atomic formula of depth 2.

Figure 2. Intervals at top of PR

A similar analysis would yield all upper intervals of depth k for other
small values of k. The point, though, is that only a few types of intervals
can occur at the top of PR .

4. The semilattice property

A more general restriction on the structure of PR is given by the next
result.
Theorem 6. For each relation symbol R, the ordered set PR is a join semi-
lattice.
Recall that the order on PR is given by substitution maps. That is, for
s, t ∈ Fn where n is the rank of R, the inclusion R(s) ≤ R(t) holds if and
only if there is a σ ∈ Sbn(F) with σ(t) = s. It is convenient to extend this
quasi-order and the induced equivalence to Fn itself, and write s ≤ t. To
prove the theorem, it then suffices to show that Fn is a semilattice. Let x
denote the top element of this ordered set, x = (x1 , . . . , xn ).
Our objective, given two elements s and t of Fn , is to find the least u
from which both s and t can be obtained by substitutions. First we find
a standard intermediate element s◦ between x and s, and likewise t◦ with
t ≤ t◦ ≤ x. Using these elements and the associated substitutions, there is
a straightforward method to determine the join.
Consider an arbitrary s ∈ Fn . Let s◦ be the expression obtained by
replacing all occurrences of variables in s by distinct variables. This is
unique up to a permutation, which is our equivalence in Fn . Note that
s ≤ u implies s◦ ≤ u◦ . Moreover, the substitution map σ : x → s can be
factored through s◦ using substitutions, say as σ = γτ where τ (x) = s◦ and
γ(s◦ ) = s.
For example, with n = 2, let s = (f (x1 , g(x1 , f (x1 , x2 ))), h(x3 , f (x1 , x2 ))).
Then s◦ = (f (x1 , g(x2 , f (x3 , x4 ))), h(x5 , f (x6 , x7 ))). Starting from x =
6 TRISTAN HOLMES, DAYNA KITSUWA, J. B. NATION AND SHERI TAMAGAWA

(x1 , x2 ) we have τ (x1 ) = f (x1 , g(x2 , f (x3 , x4 ))) and τ (x2 ) = h(x5 , f (x6 , x7 ))).
Then γ(x1 ) = γ(x2 ) = γ(x3 ) = γ(x6 ) = x1 and γ(x4 ) = γ(x7 ) = x2 and
γ(x5 ) = x3 .
En route to determining when s ≤ u, let us generalize the preceding
observation.
Lemma 7. Assume that σ(w) = s witnesses s ≤ w and that w◦ = w. Then
s◦ ≤ w, and there exist τ , γ ∈ Sbn(F) such that τ (w) = s◦ , γ(s◦ ) = s and
σ = γτ .
Proof. The condition w◦ = w means that all the variables in w are distinct.
For this case, there is a reduction scheme for testing s ≤ w that, when the
inclusion holds, reconstructs the substitution with σ(w) = s. By making
all the variables used in the substitution distinct, we then obtain τ with
τ (w) = s◦ .
First note that, when w = w◦ , then s ≤ w holds iff si ≤ wi for 1 ≤ i ≤ n.
This, in turn, is determined recursively by the rules
(1) t ≤ x for any term t and variable x;
(2) f (t1 , . . . , tm ) ≤ g(u1 , . . . , un ) iff f = g and tj ≤ uj for all j.
Note that, in (2), each uj has no repeated variables, and different uj ’s involve
different variables. The reduction process either yields
(i) x ≤ uj with x a variable and uj a proper term, for some branch, in
which case the inclusion fails, or
(ii) an expression t ≤ x for every branch, in which case defining x 7→ t
reconstructs σ.

For notation, let Xv denote the variables occurring in v.
Now if s◦ ≤ u◦ , then one can define substitutions such that
µ(u◦ ) = s◦
γ(s◦ ) = s
δ(u◦ ) = u
where dom µ = dom δ = Xu◦ and dom γ = Xs◦ . So in order to be able to
define ν with ν(u) = s witnessing the inclusion s ≤ u, it is necessary and
sufficient that ker δ ⊆ ker γµ. If so, we can find ν so that νδ = γµ. See
Figure 3.
Lemma 8. The inclusion s ≤ u holds in Fn if and only if s◦ ≤ u◦ and, for
the maps defined above, ker δ ⊆ ker γµ.
Let us continue the example, with s, s◦ and γ as before. Let u =
(f (x1 , g(x1 , x2 )), h(x3 , x2 )), so that u◦ = (f (x1 , g(x2 , x3 )), h(x4 , x5 )). Then
s◦ ≤ u◦ is witnessed by µ with µ(x1 ) = x1 , µ(x2 ) = x2 , µ(x3 ) = f (x3 , x4 ),
µ(x4 ) = x5 and µ(x5 ) = f (x6 , x7 ). Likewise u ≤ u◦ is witnessed by δ with
δ(x1 ) = x1 , δ(x2 ) = x1 , δ(x3 ) = x2 , δ(x4 ) = x3 and δ(x5 ) = x2 . Now
 LATTICES OF ATOMIC THEORIES IN LANGUAGES WITHOUT EQUALITY 7

u◦
µ δ
s◦ u
γ ν

Figure 3

ker δ = ker γµ = [x1 , x2 ] [x3 , x5 ] [x4 ], so we can find ν witnessing s ≤ u, viz.,

ν(x1 ) = x1 , ν(x2 ) = f (x1 , x2 ) and ν(x3 ) = x3 .
Now when s◦ = s and t◦ = t, it is easy to compute s ∨ t. With the
understanding that distinct variables are used throughout, s ∨ t = (s1 ∨
t1 , . . . , sn ∨ tn ) where recursively
(1) x ∨ t = x when x is a variable and t any term;
(2) f (u) ∨ g(v) = f (u ∨ v) if f = g, and
(3) f (u) ∨ g(v) = x if f 6= g.
To determine s ∨ t for arbitrary s, t ∈ Fn , first find s◦ ∨ t◦ as above.
By construction, (s◦ ∨ t◦ )◦ = s◦ ∨ t◦ . Then, given maps µ, γ, µ′ , γ ′ with
µ(s◦ ∨ t◦ ) = s◦ , γ(s◦ ) = s, µ′ (s◦ ∨ t◦ ) = t◦ and γ ′ (t◦ ) = t, we want to find
u ∈ Fn and maps δ, ν, ν ′ such that δ(s◦ ∨ t◦ ) = u, ν(s) = s and ν ′ (u) = t
with the least possible u, so that u = s ∨ t. See Figure 4.

s◦ ∨ t◦
µ δ µ′

s◦ u t◦
γ ν ν′ γ′

s t

Figure 4

We need to define δ(x) for x ∈ Xs◦ ∨t◦ . Suppose v is any element such
that s, t ≤ v ≤ s◦ ∨ t◦ . Now s ≤ v implies s◦ ≤ v◦ , and similarly t◦ ≤ v◦ .
Thus v◦ = s◦ ∨ t◦ . Hence it must be that δ(x) ∈ X for x ∈ Xs◦ ∨t◦ ; the only
question is to decide which variables have the same image. By Lemma 8, as
long as ker δ ⊆ ker γµ and ker δ ⊆ ker γ ′ µ′ , we can find ν, ν ′ with ν(v) = s
and ν ′ (v) = t. For the least such v, take ker δ = ker γµ ∩ ker γ ′ µ′ , and that
determines δ up to a permutation of the variables. Let u = δ(s◦ ∨ t◦ ).
To show that u is indeed the join, consider an arbitrary element w ≥ s, t.
Now w◦ ≥ s◦ , t◦ , so w◦ ≥ s◦ ∨ t◦ , On the variables Xw◦ define substitutions
8 TRISTAN HOLMES, DAYNA KITSUWA, J. B. NATION AND SHERI TAMAGAWA

α, β such that α(w◦ ) = w and β(w◦ ) = s◦ ∨ t◦ . We want to define ξ on Xw

so that ξα = δβ. See Figure 5.

w◦
α β

w s◦ ∨ t◦
ξ
δ µ′
µ
s◦ u t◦

γ ν ν′ γ′

s t

Figure 5

Note that α(x) ∈ X for x ∈ Xw◦ . So we can find the desired ξ exactly
when ker α ⊆ ker δβ. Now for x, y ∈ Xw◦ , we have
(x, y) ∈ ker δβ iff (βx, βy) ∈ ker δ
iff (βx, βy) ∈ ker γµ and (βx, βy) ∈ ker γ ′ µ′
iff (x, y) ∈ ker γµβ and (x, y) ∈ ker γ ′ µ′ β.
But w ≥ s implies ker α ⊆ ker γµβ, while w ≥ t implies ker α ⊆ ker γ ′ µ′ β,
so this holds. Thus w ≥ u.
Let us continue the example, with s = (f (x1 , g(x1 , f (x1 , x2 ))), h(x3 , f (x1 , x2 )))
as before, and the element t = (f (x1 , g(x1 , x2 )), h(g(x3 , x3 ), x2 )). Then t◦ =
(f (x1 , g(x2 , x3 )), h(g(x4 , x5 ), x6 )), and s◦ ∨ t◦ = (f (x1 , g(x2 , x3 )), h(x4 , x5 )).
This is the element that was labelled u◦ before, and thus we obtain the
same maps γ, µ with γµ(u◦ ) = s and ker γµ == [x1 , x2 ] [x3 , x5 ] [x4 ]. On the
t-side, we have µ′ (u◦ ) = t◦ and γ ′ (t◦ ) = t, where µ′ (x1 ) = x1 , µ′ (x2 ) = x2 ,
µ′ (x3 ) = x3 , µ′ (x4 ) = g(x4 , x5 ), µ′ (x5 ) = x6 while γ ′ (x1 ) = x1 , γ ′ (x2 ) = x1 ,
γ ′ (x3 ) = x2 , γ ′ (x4 ) = x3 , γ ′ (x5 ) = x3 , γ ′ (x6 ) = x2 . Now we calculate
ker γ ′ µ′ == [x1 , x2 ] [x3 , x5 ] [x4 ], which is also then ker γµ ∩ ker γ ′ µ′ . Thus
the map δ from above yields
s ∨ t = δ(u◦ ) = u = (f (x1 , g(x1 , x2 )), h(x3 , x2 )).

5. Conclusion
Theorems 1, 2 and Corollary 3 show that a lattice of atomic theories
ATh(B) is isomorphic to the lattice O(P) of order ideals of an ordered set
P. Theorems 4–6 then give some restrictions on the ordered sets P that
occur in this way. These results tend to support idea that the complexities
of lattices of universal theories in general arise, not from the atomic sentences
 LATTICES OF ATOMIC THEORIES IN LANGUAGES WITHOUT EQUALITY 9

in the axiom system (e.g., reflexivity for equality), but rather from the quasi-
identities (e.g., symmetry and transitivity). This is hardly surprising!

References
[1] K. Adaricheva and J.B. Nation, Lattices of quasi-equational theories as congru-
ence lattices of semilattices with operators, Parts I and II, preprint available at
www.math.hawaii.edu/∼jb.
[2] W.J. Blok and D. Pigozzi, Algebraic semantics for universal Horn logic without
equality, in Universal Algebra and Quasigroup Theory (Jadwisin, 1989), vol. 19 of
Res. Exp. Math., Heldermann, Berlin, 1992, pp. 1–56.
[3] A. Church, Introduction to Mathematical Logic, Princeton Univ. Press, 1956.
[4] J. Czelakowski, Protoalgebraic Logics, vol. 10 of Trends in Logic - Studia Logica
Library, Kluwer, Dordrecht, 2001.
[5] R. Elgueta, Characterizing classes defined without equality, Studia Logica 58 (1997),
357–394.
[6] V. Gorbunov, Algebraic Theory of Quasivarieties, Siberian School of Algebra and
Logic, Plenum, New York, 1998.
[7] W.A. Lampe, A property of the lattice of equational theories, Alg. Univ. 23 (1986),
61–69.
[8] W.A. Lampe, Further properties of lattices of equational theories, Alg. Univ. 28
(1991), 459–486.
[9] R. McKenzie, Finite forbidden lattices, Universal Algebra and Lattice Theory, Lecture
Notes in Mathematics 1004(1983), Springer-Verlag, Berlin, 176–205.
[10] D. Monk, Mathematical Logic, Springer-Verlag, 1976.
[11] J. Nation, Notes on Lattice Theory, online at www.math.hawaii.edu.
[12] J. Nation, Lattices of theories without equality, preprint.
[13] N. Newrly, Lattices of equational theories are congruence lattices of monoids with one
additional unary operation, Alg. Univ. 30 (1993), 217–220.
[14] A.M. Nurakunov, Equational theories as congruences of enriched monoids, Alg. Univ.
58 (2008), 357–372.

Department of Mathematics, University of Hawaii, Honolulu, HI 96822,

USA
E-mail address: [email protected], [email protected], [email protected],
[email protected]