You are on page 1of 16

Local Characterization Theorems

for Some Classes of Structures

JANUSZ CZELAKOWSKI and RAIMON ELGUETA∗

August 18, 1998 (Revised version)

Abstract. Given a class of first order structures K and an algebra A of the


type of K, we define the set of A-structures in K, in symbols KA , as the
set of members of K whose underlying algebra is A. In this paper we pose
the following problem: When the closure of K under the common operations
(taking substructures, direct products,. . . ) can be expressed in terms of local
properties concerning the sets KA and global properties relating KA and
KB whenever there is an algebra homomorphism from A into B . We obtain
some results in this direction which hold, in particular, for classes of structures
axiomatized by equality-free sentences. The fundamental result says that the
classes K for which KA is an algebraic closure system, for all algebras A, are
precisely the (equality-free) strict universal Horn classes.

§0. Preliminaries. All over the paper, the symbol L stands for an
equality-free type. An equality-free type is a set of function symbols -including
constants- and a nonempty set of relation symbols which does not include the
equality symbol, all of them of finite arity. A structure of type L, or simply
an L-structure, is defined as usual. We adopt the notation according to which
structures are represented by M, N, M 0 , N 0 , . . . , with appropriate subscripts.
We write M for the domain of M and M for its underlying algebra. An algebra
of type L, or simply an L-algebra, is always the underlying algebra of some

1991 Mathematics Subject Classification. Primary 03C52, Secondary 03C30.


Keywords and phrases. Set of structures on an algebra, filter extension, local and global
properties, closure condition.

This work was partially supported by grant PB94-0920 of Spanish DGICYT.

1
2 J. CZELAKOWSKI and R. ELGUETA

L-structure; thus, if the set of function symbols of L is empty, an L-algebra


simply means a set. By a class of L-structures we always mean a nonempty
class.

We shall use a bar over a letter to represent a sequence of elements. If ā


is a sequence of elements in some set A and h is a mapping with domain A,
then h(ā) indicates the sequence of images by h of the elements in ā.

For the sake of convenience, we distinguish between equations of type L,


i.e., formulas of the form s ≈ t for some terms s, t of L (in a given number of
variables), and the remaining atomic formulas of type L. We call equality-free
first order language of L, in symbols L−
ωω , the set of all first order formulas of
type L which does not contain the equality symbol.

Given an L-structure M , a formula φ(x̄) of L−


ωω and a sequence ā of elements
in M of the same length as x̄, we write M ² φ(x̄) [ā] to indicate that ā satisfies
φ(x̄) in M in the usual sense. Also, if K is a class of L-structures, we write
K ² φ(x̄); then we mean that M ² φ(x̄) [ā] for all M in K and all sequences ā
of elements in M of the same length as x̄. Finally, if Φ is a set of sentences of
L−ωω , we write M od(Φ) for the class of L-structures M such that M ² φ.

If M and N are two L-structures, we say that M is a substructure of N , in


n
symbols M ⊆ N , if M is a subalgebra of N and RM = RN ∩ M for every
n-ary relation symbol R of L. Also, we say that N is a filter extension of M ,
and in this case we write M v N , provided the underlying algebras of M and
N coincide and RM ⊆ RN for every relation symbol R of L.

§1. Homomorphisms. If M and N are L-structures, we say that a map


h : M →N is a homomorphism from M into N if h is an algebra homomor-
phism from M into N such that for every relation symbol R of L and every
tuple ā from M , ā ∈ RM implies h(ā) ∈ RN . We write h : M → N to indicate
that h is a homomorphism from M into N , and h : M ½ N and h : M ³ N
if, in addition, it is 1-1 and onto, respectively. When h is onto we call N a
homomorphic image of M .

A homomorphism h : M → N is said to be a strict homomorphism, and in


this case we write h : M →s N , if for every relation symbol R of L and every
Local characterization theorems 3

tuple ā from M ,

(1) ā ∈ RM ⇔ h(ā) ∈ RN .

We say that N is a contraction of M , and M is an expansion of N , if there


is a strict homomorphism from M onto N . A class of L-structures is said to
be an abstract class whenever it is closed under expansions and contractions,
and contains some structure with a nonempty relation. Observe that, even if
we consider equations as formulas of the language, the fact of being abstract
does not correspond to closure under isomorphic copies, because condition
(1) refers only to predicate symbols distinct from equality. Also, no class of
L-structures axiomatized by nontrivial sentences containing equality can be
abstract (see [2]).
If M and N are L-structures and h is a map from M into N , then it
is easy to see that h is a homomorphism if, and only if, for every atomic
formula φ(x̄) of L− ωω and every tuple ā from M , M ² φ(x̄) [ā] entails that
N ² φ(x̄) [h(ā)]. Similarly, h is a strict homomorphism if, and only if, for
every atomic formula φ(x̄) of L− ωω and every tuple ā in M , M ² φ(x̄) [ā] is
equivalent to N ² φ(x̄) [h(ā)]. We may use this to prove by induction the next
fundamental property, which will be applied silently several times. Notice that
this property no longer holds for formulas containing the equality symbol nor
for strict homomorphisms which are not onto.

Lemma 1. Let M, N be two L-structures and let h : M ³s N . Then for all


φ(x̄) of L−
ωω and all sequences ā of elements in M ,

M ² φ(x̄) [ā] ⇔ N ² φ(x̄) [h(ā)].

As a consequence of Lemma 1, all classes axiomatized by equality-free sen-


tences must be abstract.
If M, N are L-structures and h is an algebra homomorphism from M into
N , we define the image of M by h, in symbols h(M ), as the structure on
the algebra h(M ) where, for all relation symbols R of L, Rh(M ) = {h(ā) :
ā ∈ RM }. Also, we define the preimage of N by h, in symbols h−1 (N ), as
−1 n
the structure whose underlying algebra is h−1 (N ) and Rh (N )
= {ā ∈ M :
4 J. CZELAKOWSKI and R. ELGUETA

h(ā) ∈ RN }, for all R in L. Next lemmas follow at once from the definitions
involved and, as Lemma 1, they will be used many times without any explicit
mention.

Lemma 2. Let M , N be L-structures and let h : M →N be a homomorphism.


If h is strict, then h(M 0 ) ⊆ N for every substructure M 0 of M , and h−1 (N 0 ) ⊆
M for every substructure N 0 of N .

Lemma 3. For all L-structures M and N and all homomorphisms h : M →N ,


h defines a strict homomorphism from h−1 (N ) into N . If, in addition, h is
onto, then N is a contraction of h−1 (N ).

Recall that a binary relation θ on M is a congruence of M if θ is a congruence


of the underlying algebra which is compatible with all the relations of M in the
following sense: If R is a relation symbol of L and ā, b̄ are tuples of elements
in M ,
ā ∈ RM and ā ≡ b̄ (θ) ⇒ b̄ ∈ RM .

For any homomorphism h : M →N , the kernel of h is defined as the set


Ker h of pairs (a, b) in M such that h(a) = h(b). Trivially, if h is a strict
homomorphism, then Ker h is a congruence of M in the above sense. Also,
given a congruence θ of M , we define the quotient of M by θ, denoted M/θ,
as the L-structure whose underlying algebra is M /θ and whose relations are
given by the following condition: If R is a relation symbol of L and ā a tuple
of elements in M ,

(2) ā/θ ∈ RM/θ ⇔ ā ∈ RM .

We can readily verify that the natural map πθ : a 7−→ a/θ from M into M/θ
is a strict surjective homomorphism whose kernel is θ. Thus congruences on a
structure as defined above amounts to kernels of strict homomorphisms. The
following is the version for structures of the well known homomorphism theo-
rem in universal algebra; analogues for the remaining isomorphisms theorems
can also be proved easily.

Lemma 4. (Strict homomorphism theorem [2]) If M and N are two L-


structures and h : M ³s N , then M/Ker h ∼
= N by the map a/Ker h 7−→
Local characterization theorems 5

h(a).

§2. Operating with structures. Let κ ≥ ω be a cardinal, I a set and


F is a κ-complete filter (or ultrafilter) on I. If Mi is an L-structure for each
Q
i ∈ I, we define the direct product i∈I Mi and the κ-reduced product (or
Q
ultraproduct) i∈I Mi /F as usual; in particular, ω-reduced products are the
well-known reduced products. For the sake of convenience we allow I to be
Q
empty; in this case, i∈I Mi is by definition the L-structure which consists of
a one element underlying algebra and where all relations are nonempty. We
Q
write πj for the natural projection of i∈I Mi onto the jth component, for
Q Q
all j ∈ I, and πF for the natural projection of i∈I Mi onto i∈I Mi /F. In
general, none of these projections is a strict homomorphism. The preimage of
Q QF
i∈I Mi /F by πF , in symbols i∈I Mi , is called the κ-filtered direct product
of {Mi : i ∈ I} by F, and satisfies that, for all relation symbols R of L and
Q
all tuples ā of elements in i∈I Mi ,
QF
Mi
(3) ā ∈ R i∈I ⇔ {i ∈ I : ā(i) ∈ RMi } ∈ F,

where ā(i) indicates the n-tuple of elements in Mi which consists of the ith
components of the elements in ā. We adopt the notation according to which
a/F represents the equivalence class of a modulo the congruence relation
Q
determined by F, for all a in the domain of i∈I Mi .
If M is an L-structure, we say that M is a subdirect product of the family
Q
{Mi : i ∈ I}, and in this case we write M ⊆sd i∈I Mi , if M is a substructure
Q
of i∈I Mi and πi restricted to M is surjective for every i ∈ I. Also, we
Q
say that a 1-1 homomorphism h : M ½ i∈I Mi is a subdirect embedding, in
Q Q
symbols h : M ½sd i∈I Mi , if h(M ) ⊆sd i∈I Mi . It follows readily from
the definition that subdirect embeddings are always strict homomorphisms.
If all the Mi , for i ∈ I, are structures on the same underlying algebra, let us
say A, we define the κ-filtered intersection of {Mi : i ∈ I} by F, in symbols
TF
i∈I Mi , as the structure on A where, for all relation symbols R in L and all
tuples ā of elements in A,
TF
Mi
(4) ā ∈ R i∈I ⇔ {i ∈ I : ā(i) ∈ RMi } ∈ F.
6 J. CZELAKOWSKI and R. ELGUETA

TF
If F = {I}, then i∈I Mi turns out to be the usual intersection of the struc-
tures Mi , for i ∈ I. In view of (3) and (4), it is clear that
Y \
(5) F
Mi = F
πi−1 (Mi ).
i∈I i∈I

Hence, κ-reduced products are obtained by contracting κ-filtered intersections.


An upward κ-directed poset is a partially ordered set (P, ≤) with the prop-
erty that if X ⊂ P is of cardinality less than κ, then there is an r in P with
p ≤ r for all p in X. A (κ, v)-directed system in L consists of an upward
κ-directed poset (P, ≤) and an L-structure Mp for each p ∈ P such that for
all elements p, q ∈ P , if p ≤ q then Mp v Mq . Since all the structures in a
(κ, v)-directed system must have the same underlying algebra, we define the
S
union of the system, in symbols p∈P Mp , as the L-structure on this algebra
S
where a relation symbol R of L is interpreted as p∈P RMp . If κ = ω, we
simply say that {Mp : p ∈ P } is a v-directed system in L.

Lemma 5. Let κ ≥ ω be a cardinal and let K be a class of L-structures, all


of them on the same underlying algebra. Then K is closed under κ-filtered
intersections if, and only if, it is closed under intersections and unions of
(κ, v)-directed systems.

Proof. Fix an index set I. It is straightforward to see that, for all filters F
over I and all families {Mi : i ∈ I} of L-structures on the same underlying
algebra,
\ [ \
F
(6) Mi = Mi .
i∈I X∈F i∈X
T
So the backward implication follows from the fact that the structures i∈X Mi ,
X ∈ F, form a (κ, v)-directed system whenever F is κ-complete. But this is
clear after endowing F with the following partial ordering: X ≤ Y ⇔ Y ⊆ X.
For the reverse implication, let us verify that unions of (κ, v)-directed sys-
tems can be expressed as κ-filtered intersections. Let (P, ≤) be an upward
κ-directed poset and let Mp be an L-structure for each p ∈ P such that
Mp v Mq whenever p ≤ q. Let FP be the filter over P generated by the sets
S TF P
[p) = {q ∈ P : p ≤ q}. We claim that p∈P Mp = p∈P Mp .
Local characterization theorems 7

Indeed, if X ∈ FP , then there are p1 , . . . , pm ∈ P such that [p1 )∩· · ·∩[pm ) ⊆


X. Hence, since P is upward κ-directed, [pX ) ⊆ X for some pX ∈ P . Then
T
pX satisfies that p∈X Mp v MpX . Therefore, by the definition of FP ,
[ [ \ [ \ [ [
Mp = Mq v Mp v MpX v Mp .
p∈P p∈P q∈[p) X∈FP p∈X X∈FP p∈P

TF P S
So, using (6), we conclude that p∈P Mp = p∈P Mp .
We must still prove that FP is a κ-complete filter. This also follows rather
easily from the fact that P is upward κ-directed. Let D ⊆ FP with |D| < κ.
For every X ∈ D, there is a finite number of elements p1 , . . . , pm in P such
that [p1 ) ∩ · · · ∩ [pm ) ⊆ X. Since κ ≥ ω, the set of all these p1 , . . . , pm for X
ranging over D is of cardinality less than κ. So there is a q ∈ P which is over
T
all of them. Hence [q) ⊆ X for all X ∈ D, and thus [q) ⊆ D. Therefore
T
D is in FP . 

We define a κ-directed diagram in L, or simply a directed diagram in L if


κ = ω, as an upward κ-directed poset (P, ≤) together with an L-structure Mp
for each p ∈ P and a homomorphism hpq : Mp →Mq for each pair of elements
p, q in P , such that

(7) for all p ∈ P, hpp : Mp →Mp is the identity map;


(8) if p ≤ q ≤ r in P , then hpr = hqr ◦ hpq .

If each homomorphism hpq is onto, we say that the κ-directed diagram is


surjective. We define as usual the (surjective) κ-direct limit of this (surjective)
κ-directed diagram, in symbols limp∈P Mp . Namely, if A denotes the disjoint
union of the sets M p , for p ∈ P , we define a relation ∼ on A as follows:

(9) If a ∈ M p and b ∈ M q , then a ∼ b iff there is r


such that p ≤ r, q ≤ r and hpr (a) = hqr (b).

Using (7) and (8), ∼ is an equivalence relation, and the domain of limP Mp
is the quotient set A/ ∼. Moreover, for each p ∈ P , we define the map
hp : M p →A/ ∼ by putting hp (a) = a/ ∼. Then hq ◦ hpq = hp whenever
p ≤ q in P , since if a ∈ M p then hpq (a) ∼ a by (9). So we may interpret the
8 J. CZELAKOWSKI and R. ELGUETA

symbols of L as follows. If F is a function symbol of L and ā/ ∼ is a tuple in


A/ ∼, then put

(10) F limP Mp (ā/ ∼) = hp (F Mp (ā)),

where p ∈ P is such that ā is a tuple in Mp (it exists because (P, ≤) is an


upward κ-directed poset). Similarly, if R is a relation symbol of L, we put

(11) ā/ ∼∈ RlimP Mp ⇔ there is p ∈ P and b̄ in Mp such that


b̄ ∼ ā and b̄ ∈ RMp .

In this situation, the maps hp : Mp → limP Mp are homomorphisms satisfying


hp = hq ◦ hpq whenever p ≤ q in P . They are called limit homomorphisms
and they uniquely determine, up to isomorphism, the direct limit, in the usual
sense. We can also verify that if the maps hpq are strict (resp. 1-1, onto),
then so are the maps hp . The proof of these facts are left to the reader.

§3. Main Theorems. Let K be a class of L-structures and A an algebra


of type L. We define the set of A-structures in K, denoted by KA , as the class
of members of K on the algebra A. Obviously, KA is partially ordered by the
filter extension relation v, for all algebras A. In this section we characterize
the closure of abstract classes under certain algebraic operations in terms of
simple combinatorial properties concerning the sets KA -or equivalently, in
terms of the order structure of KA -, for A ranging over all algebras of type L.
The first result isolates those algebraic constructions which correspond to
the fundamental operations of intersection and union of (κ, v)-directed sys-
tems in KA . Part (i) goes back at least to [8].

Theorem 1. Let K be an abstract class of L-structures.


(i) KA is closed under arbitrary intersections for all L-algebras A if, and
only if, K is closed under subdirect products.
(ii) If κ ≥ ω is a cardinal, then KA is closed under unions of (κ, v)-directed
systems for all L-algebras A if, and only if, K is closed under surjective
κ-direct limits.

Proof. (i) Assume first that KA is closed under intersections for all A. Let
Q
h : M ½sd I Ni with Ni in K, for i ∈ I. We must prove that M is in K. For
Local characterization theorems 9

this, define Mi = (πi ◦ h)−1 (Ni ), for i ∈ I. Each Mi is an expansion of Ni


and, hence, a member of K, since by hypothesis K is abstract. So it suffices
T
to see that M = I Mi . Let R be an n-ary relation symbol of L and ā an
n-tuple of elements of A. Then
\
ā ∈ RMi ⇔ (πi ◦ h)(ā) ∈ RNi for all i ∈ I
i∈I
Q
Ni
⇔ h(ā) ∈ R i∈I ⇔ ā ∈ RM ,

because of the definition of subdirect embedding.


Conversely, suppose that K is closed under subdirect products. If A is an
T
algebra of type L and Mi is in KA for i ∈ I, then the intersection M = i∈I Mi
is still on A and the mapping a 7−→ ha : i ∈ Ii defines a subdirect embedding
Q
from M into I Mi . Thus M is in K, as required.
(ii) As before, let us prove first the backward implication. Assume that
KA is closed under unions of (κ, v)-directed systems, for all algebras A of
type L. Let (P, ≤) be an upward κ-directed poset and let Mp be in K for all
p ∈ P . Also, let hpq be an onto homomorphism from Mp onto Mq , for all
p, q ∈ P , and assume (7) and (8). We must show that the surjective κ-direct
limit limp∈P Mp is in K. Let
[
Np = h−1
pr (Mr ),
r≥p

and abbreviate M for limp∈P Mp . We claim that, for all p ∈ P , Np = h−1


p (M ).
Clearly, both the structures Np and h−1 p (M ) have the same underlying
algebra, which is the one of Mp . Let R be an n-ary relation symbol of L and
−1
ā an n-tuple of elements in Mp . By (11), if ā ∈ Rhp (M ) then there is a q ∈ P
and an n-tuple b̄ in Mq such that ā ∼ b̄ and b̄ ∈ RMq . Also, by (9), there is
an r ∈ P with p ≤ r and q ≤ r satisfying hpr (ā) = hqr (b̄). So, using that hqr
is a homomorphism, we obtain that hpr (ā) ∈ RMr for all r ≥ p. This proves
the inclusion h−1p (M ) v Np . The reverse inclusion follows easily from the fact
that ā ∼ hpr (ā) whenever p ≤ r. Thus the desired equality holds.
Now h−1p (M ) is an expansion of M , because the limit homomorphism hp is
onto. So, since K is abstract by assumption, it suffices to see that Np is in K.
10 J. CZELAKOWSKI and R. ELGUETA

For this, let us show that the set of structures h−1


pr (Mr ), r ≥ p, forms a
(κ, v)-directed system in L; then, we will apply the hypothesis and conclude
that their union is in K, as required. Let X be a set of elements in P of
cardinality less than κ, and suppose that p ≤ r for all r ∈ X. Since P is
upward κ-directed, we can find a q ∈ P such that r ≤ q for all r ∈ X. But
hpq = hpr ◦ hrq for all r ∈ X, and so we have that h−1 −1
pr (Mr ) v hpq (Mq ), for
r ∈ X. Hence, h−1 −1
pq (Mq ) is a filter extension of hpr (Mr ), for all r ∈ X.
To see the reverse implication, observe that the union of a (κ, v)-directed
system is a special case of surjective κ-direct limit. The details are left to the
reader. 

It is known by [7] (see [6]) that substructures and reduced products in


an abstract class can be captured by appropiate surjective direct limits of
subdirect products. So, by the previous theorem, closure under intersections
and unions of v-directed systems suffices to guarantee that a class is closed
under substructures and reduced products. It is an open problem, however,
to isolate those local conditions concerning the posets KA for the different
algebras A which characterize separately that K is closed under these two
operations. We only have the following characterizations, which involve global
properties relating KA and KB whenever there is an algebra homomorphism
from A into B.

Theorem 2. Let K be an abstract class of L-structures.


(i) K is closed under substructures if, and only if, for all algebra homo-
morphisms h : A →B, if N is in KB then h−1 (N ) is in KA .
(ii) K is closed under substructures and direct products if, and only if, it
satisfies both of the following conditions.
a. For all algebra homomorphisms h : A →B, if N ∈ KB then
h−1 (N ) ∈ KA .
b. KA is closed under arbitrary intersections (i.e., (KA , v) is a
complete lattice) for all L-algebras A.

Proof. (i) follows easily from Lemma 2, whereas (ii) follows from (i) and
Theorem 1(i). 

If in Theorem 2(ii) we replace closure under direct products by closure under


Local characterization theorems 11

κ-reduced products, for a cardinal κ ≥ ω, then condition a can be replaced


by a purely local one, as it is expected by the previous comment. We provide
here a direct proof of this fact. At this point the author is specially indebted
to Janusz Czelakowski, because he suggested him the idea to prove that (ii)
implies (i).

Theorem 3. Let K be an abstract class of L-structures and κ ≥ ω a regular


cardinal. Then the following are equivalent.
(i) K is closed under substructures and κ-reduced products.
(ii) KA is closed under intersections and unions of (κ, v)-directed systems,
for all L-algebras A.
(iii) KA is closed under κ-filtered intersections, for all L-algebras A.

Proof. (i) ⇒ (ii) Suppose that K is closed under substructures and κ-reduced
products. Then K is also closed under subdirect products and hence, by
Theorem 1(i), KA includes all its intersections, for all L-algebras A. To prove
that it is closed under unions of (κ, v)-directed systems, let (P, ≤) be an
upward κ-directed poset and Mp an A-structure in K for each p ∈ P such
that, if p, q ∈ P and p ≤ q, then Mp v Mq . The set of all ā ∈ AP for which
there exists an r ∈ P such that ap = aq whenever p, q ∈ P and p, q ≥ r is
clearly the universe of a subalgebra of AP . Denote by B this subalgebra. As
in the proof of Lemma 5, let FP be the filter over P generated by the sets
[p) = {q ∈ P : p ≤ q}, for p ∈ P , and write ∼ for the equivalence relation
Q Q
on P Mp determined by FP . We claim that the substructure of P Mp /FP
S
generated by B/ ∼ is isomorphic to the union P Mp .
Denote by N this substructure. First observe that, for all ā ∈ B, there is
a unique a ∈ A such that, for some r ∈ P , ap = a whenever p ≥ r. Besides,
since P is upward κ-directed, we have that

(12) for all n ∈ ω and all ā1 , . . . , ān ∈ B, there is an r ∈ P


such that a1p = a1 , . . . , anp = an whenever p ≥ r.

So define the map h from B/ ∼ into A by h(ā/ ∼) = a, and let us see that h
is the desired isomorphism. Since the poset P is upward κ-directed, ā ∼ b̄ iff
a = b for all ā, b̄ ∈ B, and thus h is a bijection. Let ā1 , . . . , ān be elements of
12 J. CZELAKOWSKI and R. ELGUETA

B. Using (12), if F is an n-ary function symbol of L, then

h(F N (ā1 / ∼, . . . , ān / ∼)) = h(F B (ā1 , . . . , ān )/ ∼)


= F A (h(ā1 ), . . . , h(ān )).

Also, given an n-ary relation symbol R of L,

N ² R(x1 , . . . , xn ) [ā1 / ∼, . . . , ān / ∼]


Y
⇔ Mp /FP ² R(x1 , . . . , xn ) [ā1 / ∼, . . . , ān / ∼]
p∈P

⇔ {p ∈ P : ha1p , . . . , anp i ∈ RMp } ∈ FP


\m
⇔ [pi ) ⊆ {p ∈ P : ha1p , . . . , anp i ∈ RMp } for some p1 , . . . , pm ∈ P
i=1
⇔ ∃r ∈ P such that Mp ² R(x1 , . . . , xn ) [a1p , . . . , anp ] for all p ≥ r
[
⇔ Mp ² R(x1 , . . . , xn ) [h(ā1 / ∼), . . . , h(ān / ∼)].
p∈P

Thus h is an isomorphism.
We should still prove that FP is a κ-complete filter. But this was done in
S
the proof of Lemma 5. So p∈P Mp is a substructure of a κ-reduced product
of members in K and, therefore, it is in K.
(ii) ⇒ (i) Assume that KA is closed under intersections and unions of (κ, v)-
directed systems, for all algebras A. Let F be a κ-complete filter over some set
I and Mi a structure in K for each i ∈ I. We must prove that the κ-reduced
Q
product I Mi /F is in K. By Lemma 3, this structure is a contraction of
QF
I Mi . Moreover, we know by (5) and (6) that
Y [ \
F
Mi = πi−1 (Mi ),
i∈I X∈F i∈X

But, since K is abstract and πi−1 (Mi ) is an expansion of Mi , πi−1 (Mi ) is in


T
K for all i ∈ I. So it suffices to verify that the structures i∈X πi−1 (Mi ),
X ∈ F, form a (κ, v)-directed system. For this, we use that F is κ-complete.
Define a partial ordering ≤ on F by X ≤ Y iff Y ⊆ X. Then, if D ∈ F with
T T T
|D| < κ, D is in F and X ≤ D for all X ∈ D. So i∈∩D πi−1 (Mi ) is a
T
filter extension of i∈X πi−1 (Mi ) for every X in D, as required.
Local characterization theorems 13

Let us now prove that K is closed under substructures. Let M ⊆ N with


N ∈ K, and write A for the underlying algebra of N . Let B denote the set of
all ā ∈ Aκ for which there exists a ξ ∈ κ such that aν = aµ whenever ν, µ ≥ ξ.
Obviously, B is the universe of a subalgebra of Aκ . Moreover,

(13) for all ā ∈ B, there is a unique a ∈ A such that


|{ν ∈ κ : aν = a}| = κ.

Otherwise, there would exist a ξ ∈ κ such that |{ν ∈ κ : ν ≤ ξ}| = κ. But


this is impossible, because κ is infinite. So write a for the unique element in
A which corresponds to ā ∈ B. Define

F = {X ⊆ κ : |κ \ X| < κ}.

It is clear that F is a κ-complete filter over κ. Write ∼ for the equivalence


relation on Aκ determined by F, and let M 0 denote the substructure of N κ /F
generated by the quotient set B/ ∼.
We claim that M 0 is isomorphic to M . For this, we can reason as in the
proof of the forward implication with some minor changes. If ā, b̄ ∈ B, then
ā ∼ b̄ is equivalent to |{ν ∈ κ : aν 6= bν }| < κ, and by (13) this occurs if,
and only if, a = b. So the map h : ā/ ∼7−→ a is a bijection from B/ ∼ onto
A. There remains to see that h preserves functions and relations. The proof
that h is an algebra homomorphism has no problem. Next, if R is an n-ary
relation symbol of L and ā1 , . . . , ān are elements of B, then

M 0 ² R(x1 , . . . , xn ) [ā1 / ∼, . . . , ān / ∼]


⇔ N κ /F ² R(x1 , . . . , xn ) [ā1 / ∼, . . . , ān / ∼]
⇔ {ν ∈ κ : ha1ν , . . . , anν i ∈ RN } ∈ F
⇔ ∃ξ ∈ κ such that N ² R(x1 , . . . , xn ) [a1ν , . . . , anν ] for all ν ≥ ξ
⇔ N ² R(x1 , . . . , xn ) [h(ā1 / ∼), . . . , h(ān / ∼)].

(In the last two equivalences we use again that κ is infinite). So h is an


isomorphism.
To finish the proof, let πB denote the natural projection from B onto B/ ∼.
One can readily verify that
[ \
−1
πB (M 0 ) = πν−1 (N ),
X∈F ν∈X
14 J. CZELAKOWSKI and R. ELGUETA

where now πν indicates the projection of B onto its νth component, for all
T
ν ∈ κ. Also, the structures ν∈X πν−1 (N ), X ∈ F , form a (κ, v)-directed
system, because F is a κ-complete filter. So the assumptions on K say again
−1
that πB (M 0 ) is in K, and therefore, using the claim and the hypothesis that
K is abstract, M is in K.
(ii) ⇔ (iii) This is the content of Lemma 5. 

If κ = ω, we obtain the next special case of Theorem 3.

Corollary. Let K be an abstract class of L-structures. The poset (KA , v)


is an algebraic closure system on the poset of all L-structures on A if, and
only if, K is closed under substructures and reduced products. 

We are still able to prove another result in this direction.

Theorem 4. An abstract class K of L-structures is closed under substruc-


tures, direct products and homomorphic images if, and only if, KA is a prin-
cipal filter of the poset of all L-structures on A, for all L-algebras A.

Proof. Assume first that K is closed under substructures, direct products


and homomorphic images. Fix an algebra A of type L. By Theorem 1(i), KA
is closed under arbitrary intersections. So the poset KA has a least element.
Write M for this element. Then every filter extension N of M is a homomor-
phic image of M , and hence it is in KA . Thus KA is the principal filter of
the poset of all L-structures on A generated by M . The proof of the converse
does not entail any difficulties, and so it is left as an exercise. 

Addenda. The referee kindly pointed out to the authors the connec-
tion between their work in this paper and [7] (or [6]). This connection is
as follows. Let M be an L-structure. Following to [7], we define a congru-
ence on M as an ordered pair Θ = hΘ(L), Θ(≈)i where Θ(L) is a sequence
n
hθ(R) : R relation symbol of Li such that RM ⊆ θ(R) ⊆ M (n means the
arity of R) and Θ(≈) is a congruence on the structure MΘ = hM , Θ(L)i in
the sense of §1 above (cf. the notion of filter-congruence given in [3]). Also,
we define the quotient of M by Θ as the structure M/Θ := MΘ /Θ(≈). We
write Con M for the set of congruences on M in the sense defined right now
Local characterization theorems 15

and, given a class of L-structures K, we put

ConK M := {Θ ∈ Con M : M/Θ ∈ K}.

Then [7] (see also [6]) shows that K is closed under subdirect products if, and
only if,

(*) ConK M is closed under intersections for any L-structure M ;

and K is closed under surjective direct limits if, and only if,

(**) ConK M is closed under unions of chains fo any L-structure M .

In the present paper, conditions (*) and (**) are relaxed in the sense that the
posets ConK M , for M ranging over all L-structures, are replaced by the posets
KA , for A ranging over the class of L-algebras. The relationship between these
posets is clear: if MA denotes the L-structure on A where all relations are
identically false, then

KA ∼
= {Θ ∈ ConK MA : Θ(≈) = {ha, ai : a ∈ A}}.

Observe that the assumption that K is abstract is essential for this weakening
of the hypothesis. To get some insight about the convenience of using the
posets ConK M or the posets KA for studiyng the structural properties of
classes, we refer the reader to [3].

References
[1] S. Burris and H.P. Sankappanavar, “A course in universal algebra”, Graduate
Texts in Mathematics, Vol. 78, Springer-Verlag, New York, 1981.
[2] R. Elgueta, Characterizing classes defined without equality, Studia Logica 58
(1997), 357–394.
[3] , An approach to the algebraic study of equality-free classes, to appear in
Algebra Universalis.
[4] , Freeness in classes defined without equality, to appear in the Journal of
Symolic Logic.
16 J. CZELAKOWSKI and R. ELGUETA

[5] , Algebraic characterizations for universal fragments of logic, to appear in


Mathematical Logic Quarterly.
[6] V.A. Gorbunov, The structure of the lattice of quasivarieties, Algebra Universalis
32 (1994), 493–530.
[7] V.A. Gorbunov and V.I. Tumanov, The structure of lattices of quasivarieties, in
“Mathematical logic and the theory of algorithms”, Trudy Inst. Mat. Sib. Otd.
Akad. Nauk SSSR 2, Novosibirsk, Nauka, 1982; pp. 12–44.
[8] A.I. Mal’cev, “The metamathematics of algebraic systems”, Collected Papers:
1936– 1937, Studies in Logic and the Foundations of Mathematics, Vol. 66,
North- Holland, Amsterdam, 1971.

DEPARTMENT OF MATHEMATICS, UNIVERSITY OF OPOLE, OPOLE, POLAND.

E-mail address: jczel@sparc-1.uni.opole.pl


DEPARTMENT OF APPLIED MATHEMATICS II, POLYTECHNIC UNIVERSITY OF CAT-

ALONIA, Pau Gargallo 5, 08028 BARCELONA, SPAIN.

E-mail address: elgueta@ma2.upc.es

You might also like