Mathematics Institute
University of Bern
Sidlerstrasse 5
CH-3012 Bern
Switzerland
www.math.unibe.ch
where and are min and max calculated component-wise and (a, b) = (1 b, 1 a).
These classes have therefore received considerable attention in the fuzzy logic literature (see, e.g., [8]). We note, moreover, that De Morgan algebras provide an underlying
involutive lattice structure for the algebras of substructural logics such as R-mingle,
Lukasiewicz logic, and multiplicative additive linear logic (see, e.g., [7]).
The aim of this paper is to investigate and develop characterizations of admissibility in the context of De Morgan algebras. A rule is admissible in a logic (understood
as a consequence relation) if every substitution that makes each premise of the rule
into a theorem of the logic, also makes the conclusion into a theorem. Equivalently, a
rule is admissible if it can be added to the logic without producing any new theorems.
The admissible rules of classical propositional logic are also derivable (that is, classical propositional logic is structurally complete), but this is not the case in general for
non-classical logics (see [20, 912, 18, 3, 4, 13, 14]). In algebra, rules correspond (roughly
speaking) to quasi-identities and the admissible quasi-identities of a quasivariety may
be understood as the quasi-identities that hold in the free algebra on countably innitely many generators.
For algebraizable logics, admissible rules may be translated into admissible quasiidentities and vice versa. De Morgan algebras provide semantics for Belnaps fourvalued logic (which, note, has no theorems), but do not form the equivalent algebraic
semantics for any algebraizable logic (see [5] for details) and we therefore focus only
on the algebraic notion. In particular, we give characterizations here of admissible
quasi-identities for the classes of Kleene lattices KL, Kleene algebras KA, De Morgan
lattices DML, and De Morgan algebras DMA. For KL, KA, and DML, axiomatizations of
the admissible quasi-identities make use of a single additional quasi-identity. However,
in the case of DMA, we make use of not only a quasi-identity but also a universal
formula. We conclude by giving a brief overview of recent admissibility and structural
completeness results for some related classes of algebras.
2 Admissibility in Quasivarieties
Let us begin by briey recalling some basic notions from universal algebra, referring
to [2] for further details. For a language L, we denote the formula algebra over countably
innitely many variables by FmL and let the metavariables , , stand for arbitrary
members of FmL called L-formulas. An L-identity is a pair of L-formulas, written
, and we let the metavariables , stand for arbitrary nite sets of L-identities.
An L-quasi-identity is identied with an ordered pair consisting of a nite set of Lidentities and a single L-identity , written (often dropping the
external brackets in ). We denote sets of L-quasi-identities using the metavariable
. As usual, if the language is clear from the context we may omit the prex L when
referring to these concepts.
Let K be a class of algebras of the same language L and let { } be a
nite set of L-identities. We write |=K to denote that for every A K
and homomorphism h : FmL A, ker h implies ker h. We abbreviate
|=K by |=K , and |={A} by |=A , saying in the latter
case that the quasi-identity holds in the algebra A. K is said to be a
quasivariety if there exists a set of L-quasi-identities such that A K if and only if
(henceforth, i) all quasi-identities in hold in A. If there exists such a consisting
only of identities, then K is called a variety. The variety V(K) and quasivariety Q(K)
generated by K are, respectively, the smallest variety and quasivariety containing K.
Now let Q be a quasivariety for a language L. An L-quasi-identity will
be called admissible in Q, if for every homomorphism : FmL FmL :
|=Q ( ) ( ) for all
implies
|=Q () ().
In fact, quasi-identities admissible in Q are simply the quasi-identities that hold in the
free algebra on countably innitely many generators of Q, denoted FQ . To establish this
well known but crucial fact, we make use of the canonical homomorphism hQ : FmL
FQ that maps each formula to its equivalence class in FQ , recalling (see [2] for details)
that for each L-identity :
|=Q
|=FQ
hQ () = hQ ().
|=FQ .
x0
(which dene the quasivariety of torsion-free abelian groups) hold in all free abelian
groups and are hence admissible in the variety, but do not hold in all abelian groups.
A quasivariety Q is called structurally complete if each of its proper subquasivarieties generates a proper subvariety of V(Q), i.e., for each quasivariety Q Q, we
have V(Q ) V(Q). Equivalently (proved by Bergman in [1], Proposition 2.3), Q is
structurally complete i Q = Q(FQ ). Hence, combining with Lemma 1:
Corollary 1 The following are equivalent for any quasivariety Q for a language L:
(a) Q is structurally complete
(b) Q = Q(FQ )
(c) An L-quasi-identity is admissible in Q i |=Q .
This corollary supplies a method for establishing structural completeness for quasivarieties. A quasivariety Q is structurally complete if each member of a class of algebras
generating Q as a quasivariety can be embedded into FQ , since then any quasi-identity
failing in one of the generating algebras must fail in FQ . More precisely:
For convenience, in the remainder of this paper, we will use the symbols Ll and Lb
to denote, respectively, the languages of De Morgan lattices and De Morgan algebras.
3 Kleene Algebras
Recall from the introduction that a Kleene algebra is a De Morgan algebra satisfying
x x y y
and that the variety of Kleene algebras is denoted by KA. The following nite algebras
are particularly useful members of this variety (where 1 m N):
C2m = {m, m + 1, . . . , 1, 1, . . . , m 1, m}, min, max, , m, m
C2m+1 = {m, m + 1, . . . , 1, 0, 1, . . . , m 1, m}, min, max, , m, m .
The fuzzy algebra [0, 1], min, max, 1 x, 0, 1 and also each Cn for any odd n 3,
generates KA as a quasivariety. In particular, KA = Q(C3 ) (see, e.g. [15, 19]).
Now consider the quasi-identity
x x
x y.
(1)
Following almost exactly the corresponding proof of [19], Proposition 4.7 for Kleene
lattices (see also [6], page 239), Q(C4 ) is axiomatized relative to KA by the quasiidentity
x x, x y x y y y.
(2)
Hence we obtain:
Theorem 1 {(2)} is a basis for the admissible quasi-identities of KA.
Proof Q(C4 ) is structurally complete (Lemma 4) and axiomatized relative to KA by
{(2)} (see [19], Proposition 4.7). Moreover, C3 is a homomorphic image of C4 , so
V(C4 ) = V(C3 ) = KA. Hence, since (2) holds in C4 , it is admissible in KA, and the
result follows by Lemma 3.
Note that the quasi-identity (1) does not provide a basis for the admissible quasiidentities of KA. In fact, it axiomatizes the quasivariety Q(C3 C2 ) relative to KA
(see [19], Proposition 4.5). We remark also that by almost exactly the same reasoning, we can show that {(2)} provides a basis for the admissible quasi-identities of the
class KL of Kleene lattices. The only dierence lies in the proof of Lemma 4 that the
quasivariety of Kleene lattices generated by the four element chain is structurally complete: in this case we simply change the mapping g to g(1) = x x, g(1) = x x,
g(2) = (x x) y, and g(2) = (x x) y.
4 De Morgan Algebras
The class DMA of De Morgan algebras is generated as a quasivariety by the four-valued
diamond algebra
D4 = {, a, b, }, , , , ,
where {, a, b, }, , , , is the diamond bounded lattice of Fig. 1 and is dened
by = , = , a = a, and b = b. That is, DMA = Q(D4 ) (see [15]). However,
to obtain a characterization of the admissible quasi-identities of DMA, it will be helpful
to rst consider the (easier) case of the class DML of De Morgan lattices. Let us write
AL to denote the De Morgan lattice reduct of a De Morgan algebra A, noting that
DML = Q(DL ).
4
and
= x2 x2 ,
and note that < and < in FQ(DL ) . Now we dene e : DL FQ(DL ) by
42
42
42
e = hQ(DL ) g where
42
g(, 1)
g(a, 1)
g(b, 1)
g(, 1)
=
=
=
=
( ) ( )
( )
( )
g(, 1)
g(a, 1)
g(b, 1)
g(, 1)
=
=
=
=
( )
( )
( ) ( ).
It is straightforward to check that none of these formulas are equivalent to each other
in FQ(DL ) and hence that the mapping is one-to-one. It remains to check that e is
42
a homomorphism. First note that is preserved by the mapping; e.g., e((a, 1)) =
e(a, 1) = e(a, 1) follows from the fact that
x (x y) ((x y) x)
7
Q(DL ) = DML
4
Q(DL , CL )
3
42
Q(CL ) = KL
3
Q(DL )
42
Q(CL CL )
2
3
Q(CL )
4
Q(CL ) = BL
2
holds in FQ(DL ) . To see that is preserved by the mapping (the case of is dual),
42
we note that
(x y) (x y) < x (x y) < x y
(x y) (x y) < (x y) y < x y
y x < (x y) x < (x y) (x y)
y x < y (x y) < (x y) (x y)
hold in FQ(DL ) . Moreover, using the fact that < and < in FQ(DL
42 )
42
( ) ( ) <
( ) < ( )
( ) < ( )
< ( ) ( )
hold in FQ(DL ) , as do the following
42
( ) ( ) ( ( )) ( ( )) ( ) ( )
( ( )) ( ( )) ( ) ( )
(( ) ) (( ) ) ( ) ( )
so is preserved by e.
We now turn our attention to De Morgan algebras. Here the picture is not so clear
since the quasivariety lattice is innite (see [6]). In particular, unlike the case of DML,
the quasi-identity (1) does not provide a basis for the admissible quasi-identities of
DMA. It follows from results of Pynko [19] that {(1)} axiomatizes the quasivariety
Q(D42 ) relative to DMA. However, the quasi-identity
(x x) y
is admissible in DMA but does not hold in the De Morgan algebra D42 . So {(1)} cannot
suce as a basis for the admissible quasi-identities of DMA.
Let us consider instead the De Morgan algebra D obtained from D42 by adding
42
an extra top element and bottom element (see Fig. 1). Note that D4 is a homomorphic image of D under the composition of f : D D42 , f () = (, 1),
42
42
f () = (, 0), f ((x, y)) = (x, y) for all (x, y) {, } and the projection p : D42
D4 , p(x, y) = x. Hence V(Q(D )) = DMA. Moreover:
42
Lemma 6 Q(D ) is structurally complete.
42
Proof We extend the embedding given in the proof of Lemma 5 by dening
g() =
and
g() =
42
It follows that the admissible quasi-identities of DMA consists of those quasiidentities that hold in Q(D ). However, unlike the cases of Kleene algebras and De
42
Morgan lattices, we have been unable to nd an axiomatization of this quasivariety
using just quasi-identities. Instead we make use also of a universal formula. More precisely, for a language L, we identify universal formulas consisting of an ordered pair of
nite sets , of L-formulas, written (often dropping brackets). For a class
of algebras K for L, we write |=K to denote that for every A K and homomorphism h : FmL A, ker h implies ker h = . As for quasi-identities, we drop
brackets when considering just one algebra and say that the universal formula holds
in this algebra.
Observe that the following universal formula holds in D and hence also in FDMA :
42
xy
x , y .
(3)
Let us dene DMA to be the class of all De Morgan algebras A such that the quasiidentity (1) and the universal formula (3) both hold in A. We will show that a quasiidentity is admissible in DMA i it holds in all members of DMA . The main idea of
the proof will be to reduce the question of the admissibility of a quasi-identity in DMA
to the question of the admissibility of certain quasi-identities in DML. The following
lemma, proved by an easy induction on c(), the number of occurrences of connectives
, , and in a formula , will be useful in this respect.
Lemma 7 For any FmLb , one of the following holds:
1. |=DMA
2. |=DMA
3. |=DMA for some FmLl with c() c().
Let us say that an Lb -identity is in normal form if and are either , , or
members of FmLl .
Theorem 3 For any Lb -quasi-identity :
is admissible in DMA
|=DMA .
Proof Suppose rst that |=DMA . Both the quasi-identity (1) and the universal
formula (3) hold in FDMA , so FDMA DMA . Hence |=FDMA and by Lemma 1,
is admissible in DMA. For the other direction, it suces, using Lemmas 7
and 1, to prove the following:
For any nite set { } of Lb -identities in normal form:
|=FDMA
implies
|=DMA .
()
Let c() be the number of occurrences of connectives , , and in and let s()
be the number of identities in containing or . We prove () by induction on
the lexicographically ordered pair c(), s() . The idea is to successively eliminate
occurrences of and in by reducing c(), s() .
Base case. Suppose that there are no occurrences of and in , i.e., s() = 0. If
= or {, } {, }, then we are done. Moreover, if FmLl and {, },
then |=FDMA : just consider a homomorphism from FmLb to D4 that maps
all the variables to a. Finally, consider , FmLl . Suppose that |=FDMA .
By Lemma 1, is admissible in DMA. But for any , FmLl , we have
|=DMA i |=D4 i |=DL i |=DML . So
4
10
2. c( ) c( ) and s( ) = s( )
3. |=DMA implies {x } |=DMA .
By the induction hypothesis, using 1. and 2., |=DMA . But then also
by 3., {x } |=DMA as required.
The cases = {1 2 }, = {1 2 }), = { },
and = {x } are treated symmetrically to the preceding cases.
We remark that this result leaves open two interesting questions: Can we nd a
similarly elegant basis of quasi-equations for the admissible quasi-identities of DMA?
And does {(1), (3)} axiomatize the universal theory of FDMA relative to DMA (i.e., is
it the case that holds in FDMA i holds in all De Morgan algebras in
which (1) and (3) both hold)?
5 Related Work
De Morgan lattices can be regarded as the algebraic counterpart of Belnaps four-valued
logic (see [5] for details). However, since there exists no faithful translation of equations
into formulas of this logic, DML is not the equivalent algebraic semantics of this or of
any algebraizable logic ([5], Proposition 2.12). Indeed, Belnaps logic has no theorems so
admissibility is trivial: every rule with at least one premise is admissible. Nevertheless,
the algebras of many notable (substructural and many-valued) logics have De Morgan
algebras or De Morgan lattices as reducts, in particular, the algebras of multiplicative
additive linear logic, the relevant logics R and R-Mingle, and Lukasiewicz logics (see,
e.g., [7]). In this nal section, we briey survey the state of the art regarding questions
of structural completeness and admissible rules for these and related classes of algebras.
An involutive commutative residuated lattice (involutive CRL for short) is an algebra A = A, , , , , , t with binary operations , , , , a unary operation , and
a constant t such that (i) A, , , is a De Morgan lattice; (ii) A, , t is a commutative monoid; (iii) x y = (x y) for all x, y A. A bounded involutive CRL is an
algebra A = A, , , , , , t, , such that A, , , , , , t is an involutive CRL
and A, , , , , is a De Morgan algebra. We also dene x0 = t and xn+1 = x xn
for n N.
It is easy to see (following a similar proof by Cintula and Metcalfe in [3]) that the
variety of involutive CRLs is not structurally complete. For 3 n N, let
Ln = {0, 1/(n 1), . . . , (n 2)/(n 1), 1}
and
Ln = Ln , min, max, L , L , L , 1
xy
does not hold in Q since it does not hold in Ln (just let x = (n 2)/(n 1) and y = 1).
On the other hand, note that if n1 and hold in all members of Q
for some formula , then = holds in the two-valued algebra L2 . But this is not
possible, so the quasi-identity is admissible in Q.
11
t y.
References
1. Bergman, C.: Structural completeness in algebra and logic. In: H. Andrka, J. Monk,
e
I. Nemeti (eds.) Algebraic Logic (Proc. Conf., Budapest, 8-14 August 1988), Colloquia
Mathematica Societatis Jnos Bolyai, vol. 54, pp. 5973. North-Holland, Amsterdam
a
(1991)
2. Burris, S., Sankappanavar, H.P.: A Course in Universal Algebra, Graduate Texts in Mathematics, vol. 78. Springer-Verlag, New York (1981)
3. Cintula, P., Metcalfe, G.: Structural completeness in fuzzy logics. Notre Dame Journal of
Formal Logic 50(2), 153183 (2009)
4. Cintula, P., Metcalfe, G.: Admissible rules in the implication-negation fragment of intuitionistic logic. Annals of Pure and Applied Logic 162(10), 162171 (2010)
5. Font, J.M.: Belnaps four-valued logic and De Morgan lattices. Logic Journal of the IGPL
5(3), 129 (1997)
6. Gaitn, H., Perea, M.H.: A non-nitely based quasi-variety of De Morgan algebras. Studia
a
Logica 78(1-2), 237248 (2004)
7. Galatos, N., Jipsen, P., Kowalski, T., Ono, H.: Residuated Lattices: An Algebraic Glimpse
at Substructural Logics. Elsevier (2007)
12
8. Gehrke, M., Walker, C.L., Walker, E.A.: Normal forms and truth tables for fuzzy logics.
Fuzzy Sets and Systems 138, 2551 (2003)
9. Ghilardi, S.: Unication in intuitionistic logic. Journal of Symbolic Logic 64(2), 859880
(1999)
10. Ghilardi, S.: Best solving modal equations. Annals of Pure and Applied Logic 102(3),
184198 (2000)
11. Iemho, R.: On the admissible rules of intuitionistic propositional logic. Journal of Symbolic Logic 66(1), 281294 (2001)
12. Jeabek, E.: Admissible rules of modal logics. Journal of Logic and Computation 15,
r
411431 (2005)
13. Jeabek, E.: Admissible rules of Lukasiewicz logic. Journal of Logic and Computation
r
20(2), 425447 (2010)
14. Jeabek, E.: Bases of admissible rules of Lukasiewicz logic. Journal of Logic and Compur
tation 20(6), 11491163 (2010)
15. Kalman, J.A.: Lattices with involution. Trans. Amer. Math. Soc. 87, 485491 (1958)
16. Moisil, G.: Recherches sur lalgbre de la logique. Ann. Sci. Univ. Jassy 22(3), 1117
e
(1935)
17. Olson, J.S., Raftery, J.G.: Positive Sugihara monoids. Algebra Universalis 57, 7599 (2007)
18. Olson, J.S., Raftery, J.G., Alten, C.J.V.: Structural completeness in substructural logics.
Logic Journal of the IGPL 16(5), 453495 (2008)
19. Pynko, A.P.: Implicational classes of De Morgan lattices. Discrete Math. 205(1-3), 171181
(1999)
20. Rybakov, V.V.: Admissibility of Logical Inference Rules, Studies in Logic and the Foundations of Mathematics, vol. 136. Elsevier, Amsterdam (1997)