, ,

α, β . . .

denote congruences on some algebra

A

;

Θ, Γ

are used for tolerances (reflexive, symmetric and admissible relations), while we reserve the letters

R, S

to denote reflexive (but otherwise arbitrary) binary relations. The words admissible and compatible will be used with the same meaning.

Already [J] realized the importance of dealing with reflexive admissible relations (cf. [J,Theorem2.16] ). [T] presents a very clear discussion of the interplay between congruence identities and identities involving reflexive admissible relations. Notice that our notation mainly comes from [T] , and differs from [J] .

We shall write

a R b

to mean that

(a, b) \in R

, and we will use chains of the above notation: for example,

a Θ b α c R d

means

(a, b) \in Θ

(b, c) \in α

and

(c, d) \in R

Intersection is sometimes denoted by juxtaposition; in particular

α β

denotes the meet of the congruences

α

and

β

R^{*}

denotes the transitive closure of the binary relation

R

; in particular,

Θ^{*}

is the smallest congruence which contains the tolerance

Θ

\bar{R}

denotes the smallest compatible relation containing

R

(where

R

is a binary relation on some algebra which should be clear from the context). In particular,

\bar{Θ \cup Γ}

is the smallest tolerance which contains the tolerances

Θ

and

Γ

R + S

denotes

\cup_{n \in N} {\underset{︸}{R \circ S \circ R \circ S . . .}}_{n f a c t o r s}

Thus,

R + S

is the transitive closure of

R \cup S

, and even the transitive closure of

R \circ S

, since

R

and

S

are supposed to be reflexive. In particular, if

α, β

are congruences,

α + β

is the join of

α

and

β

in the lattice of congruences, while, for

Θ

Γ

tolerances,

Θ + Γ

is the smallest congruence which contains both

Θ

and

Γ

. Notice that

Θ + Γ

is far larger than the join of

Θ

and

Γ

in the lattice of tolerances.

R^{-}

denotes the converse of

R

, that is,

a R^{-} b

if and only if

b R a

. In particular,

R + R^{-}

is the smallest equivalence relation containing

R

C g (R)

is the smallest congruence containing

R

. Notice that if

R

and

S

are compatible then

R \circ S

is compatible; and, by an induction, we get that

R + S

is compatible, too. Thus, if

R

is compatible, then

C g (R) = R + R^{-}

In general, for

R

not necessarily compatible,

C g (R) = C g (\bar{R}) = \bar{R} + {\bar{R}}^{-}

2 Intersection properties

Theorem 2.1. Suppose that

A

is an algebra such that every subalgebra of

A \times A

generated by

4

elements satisfies

β (γ \circ δ \circ γ) \subseteq β γ + δ

, for all congruences

β, γ, δ

with

δ \leq β

Then

A

satisfies

α (R + S) \subseteq α (\bar{R \cup S^{-}}) + α (\bar{R^{-} \cup S}) = α (\bar{R \cup S}) + α (\bar{R^{-} \cup S^{-}}) = α (C g (R) + C g (S))

for all reflexive relations

R

and

S

and every congruence

α

Proposition 2.2. Under the hypothesis of Theorem 2.1 ,

A

satisfies

α (R \circ S) \subseteq α (\bar{R \cup S^{-}}) + α (\bar{R^{-} \cup S})

for all reflexive relations

R

and

S

and every congruence

α

Proof. Let $a, c \in A$ and $(a, c) \in α \cap (R \circ S)$ . Thus $a α c$ , and there is $b \in A$ such that $a R b S c$ .
Consider the subalgebra $B$ of $A \times A$ generated by the four elements $(a, a), (a, b), (c, b), (c, c)$ .
First, observe that if $(x, y) \in B$ then $(x, y) \in \bar{R \cup S^{-}}$ , since all the generators of $B$ are in $R \cup S^{-}$ , and, by definition, $\bar{R \cup S^{-}}$ is compatible.
We have that $((a, a), (a, b))$ and $((c, b), (c, c))$ belong to $(0 \times 1)_{| B}$ , $((a, b), (c, b))$ belongs to $(α \times 0)_{| B}$ and $((a, a), (c, c))$ belongs to $(α \times α)_{| B}$ .
The above relations imply that $((a, a), (c, c))$ belongs to $(α \times α)_{| B} \cap ((0 \times 1)_{| B} \circ (α \times 0)_{| B} \circ (0 \times 1)_{| B})$ Since $(α \times 0)_{| B} \leq (α \times α)_{| B}$ , by the hypothesis of the Theorem,
$(α \times α)_{| B} \cap ((0 \times 1)_{| B} \circ (α \times 0)_{| B} \circ (0 \times 1)_{| B}) \subseteq ((α \times α)_{| B} \cap (0 \times 1)_{| B}) + (α \times 0)_{| B} = (0 \times α)_{| B} + (α \times 0)_{| B}$
In conclusion, $((a, a), (c, c))$ belongs to $(0 \times α)_{| B} + (α \times 0)_{| B}$ This implies that there is some $n$ , and there are pairs $(x_{i}, y_{i}) \in B$ $(0 \leq i \leq n)$ such that $(a, a) = (x_{0}, y_{0}) (x_{n}, y_{n}) = (c, c)$ $(x_{i}, y_{i}) \equiv (x_{i + 1}, y_{i + 1}) m o d (0 \times α)_{| B} f o r i e v e n$ $(x_{i}, y_{i}) \equiv (x_{i + 1}, y_{i + 1}) m o d (α \times 0)_{| B} f o r i o d d$ In other words, $a = x_{0} = y_{0} c = x_{n} = y_{n}$ $x_{i} = x_{i + 1}, y_{i} α y_{i + 1} f o r i e v e n$ $x_{i} α x_{i + 1}, y_{i} = y_{i + 1} f o r i o d d$ In particular, $a = x_{0} = x_{1} α x_{2} = x_{3} α x_{4} . . .$ , and $a = y_{0} α y_{1} = y_{2} α y_{3} = y_{4} . . .$ , hence $x_{i} α y_{j}$ for all $i$ 's and $j$ 's, since $α$ is a congruence, and both $x_{i}$ and $y_{j}$ are congruent to $a$ modulo $α$ .
Moreover, since $(x_{i}, y_{i}) \in B$ , then $(x_{i}, y_{i}) \in \bar{R \cup S^{-}}$ for all $i$ 's, by the remark made after the definition of $B$ .
Hence, for all $i$ 's, $(x_{i}, y_{i}) \in α (\bar{R \cup S^{-}})$ , and $(y_{i}, x_{i}) \in {(α (\bar{R \cup S^{-}}))}^{-} = α (\bar{R^{-} \cup S})$ .
In conclusion, the sequence $a = x_{0} = x_{1} y_{1} = y_{2} x_{2} = x_{3} y_{3} = x_{4} . . . x_{n} = y_{n} = c$ witnesses that $(a, c) \in α (\bar{R \cup S^{-}}) + α (\bar{R^{-} \cup S})$ . □

Corollary 2.3. [L] Under the hypothesis of Theorem 2.1 ,

A

satisfies

(w T I P) α Θ^{*} = (α Θ)^{*}

for every tolerance

Θ

and every congruence

α

Proof. One inclusion is trivial. By taking $R = S = Θ$ in Proposition 2.2 , we get $α (Θ \circ Θ) \subseteq (α Θ)^{*}$ . The conclusion follows by induction: see [L,Lemma3.3] for details; actually, the argument comes from [CH] and [CH1] . □

Corollary 2.4. Under the Hypothesis of Theorem 2.1 ,

A

satisfies

α (R + R^{-}) \subseteq α (\bar{R} + {\bar{R}}^{-}) = α \bar{R} + α {\bar{R}}^{-} = α C g (R)

for every reflexive relation

R

Proof. The first inclusion, as well as the inclusion $α (\bar{R} + {\bar{R}}^{-}) \supseteq α \bar{R} + α {\bar{R}}^{-}$ are trivial.
Since $\bar{R} \circ {\bar{R}}^{-}$ is a tolerance, we can apply Corollary 2.3 with $\bar{R} \circ {\bar{R}}^{-}$ in place of $Θ$ , thus getting $α (\bar{R} + {\bar{R}}^{-}) = α (\bar{R} \circ {\bar{R}}^{-})^{*} = {(α (\bar{R} \circ {\bar{R}}^{-}))}^{*} \subseteq α \bar{R} + α {\bar{R}}^{-}$ , where the last inclusion follows from Proposition 2.2 , with ${\bar{R}}^{-}$ in place of $S$ , since ${\bar{R}}^{- -} = \bar{R}$ , and since $(α \bar{R} + α {\bar{R}}^{-})^{*} = α \bar{R} + α {\bar{R}}^{-}$ .
$C g (R) = \bar{R} + {\bar{R}}^{-}$ holds in every algebra, as mentioned at the end of Section 1 , hence $α C g (R) = α (\bar{R} + {\bar{R}}^{-})$ . □

Proof of Theorem 2.1 . Since $(\bar{R \cup S^{-}})^{-} = \bar{R^{-} \cup S}$ , we can apply Corollary 2.4 with $\bar{R \cup S^{-}}$ in place of $R$ , getting $α (R + S) \subseteq α ((\bar{R \cup S^{-}}) + (\bar{R^{-} \cup S})) = α ((\bar{R \cup S^{-}}) + (\bar{R \cup S^{-}})^{-}) = α (\bar{R \cup S^{-}}) + α (\bar{R \cup S^{-}})^{-} = α (\bar{R \cup S^{-}}) + α (\bar{R^{-} \cup S})$ .
Since $\bar{R \cup S^{-}} \subseteq \bar{R} \circ {\bar{S}}^{-}$ , $α (\bar{R \cup S^{-}}) \subseteq α (\bar{R} \circ {\bar{S}}^{-}) \subseteq α (\bar{R \cup S}) + α (\bar{R^{-} \cup S^{-}})$ , by Proposition 2.2 with $\bar{R}$ in place of $R$ and ${\bar{S}}^{-}$ in place of $S$ , and since $\bar{\bar{R} \cup \bar{S}} = \bar{R \cup S}$ . Similarly, $α (\bar{R^{-} \cup S}) \subseteq α (\bar{S} \circ {\bar{R}}^{-}) \subseteq α (\bar{R \cup S}) + α (\bar{R^{-} \cup S^{-}})$ , hence $α (\bar{R \cup S^{-}}) + α (\bar{R^{-} \cup S}) \subseteq α (\bar{R \cup S}) + α (\bar{R^{-} \cup S^{-}})$ . By replacing $S$ with $S^{-}$ in the inclusion just obtained, we get the reverse inclusion.
For the last identity, $α (C g (R) + C g (S)) = α (C g (R \cup S)) = α (\bar{R \cup S}) + α (\bar{R \cup S})^{-} = α (\bar{R \cup S}) + α (\bar{R^{-} \cup S^{-}})$ by the last identity in Corollary 2.4 , with $R \cup S$ in place of $R$ . □

References

Gábor Czédli and Eszter K. Horváth, Congruence distributivity and modularity permit tolerances, Acta Universitatis Palackianae Olomucensis. Facultas Rerum Naturalium. Mathematica 41 (2002), 39–42. MR MR1967338 (2004b:08010)
, All congruence lattice identities implying modularity have Mal $^{'}$ tsev conditions, Algebra Universalis 50 (2003), no. 1, 69–74. MR MR2026828 (2004j:08006)
Bjarni Jónsson, Congruence varieties, Algebra Universalis 10 (1980), no. 3, 355–394. MR MR564122 (81e:08004)
Paolo Lipparini, Tolerance intersection properties and subalgebras of squares, submitted (available at the author's web page).
Steven T. Tschantz, More conditions equivalent to congruence modularity, Universal algebra and lattice theory (Charleston, S.C., 1984), Lecture Notes in Math., vol. 1149, Springer, Berlin, 1985, pp. 270–282. MR MR823021 (87e:08009)

Dipartimento di Matematica, Viale della Ricerca Scientifoca, II Universita di Roma (Tor Vergata), ROME ITALY E-mail address : lipparin@axp.mat.uniroma2.it URL: http://www.mat.uniroma2.it/~lipparin

Paolo Lipparini