Algebra Universalis, 34 (1995) 404-423 0002-5240/95/030404-20501.50 + 0.20/0 9 1995 Birkh/iuser Verlag, Basel Total tense algebras and symmetric semiassociative relation algebras P. JIPSEN, R. L. KRAMER AND R. D. MADDUX Abstract. It is well known that the lattice ARA of varieties of relation algebras has exactly three atoms. An unsolved problem, posed by B. J6nsson, is to determine the varieties of height two in ARA. This paper solves the corresponding question for varieties generated by total tense algebras. More specifically,we show that there are exactly four finitelygenerated varieties and infinitelymany nonfinitely generated varieties of height two. In the second half of the paper we show that total tense algebras are term equivalent to certain generalized relation algebras and extend our results to varieties of these algebras. 1. Introduction and definitions Tense algebras are the algebraic c o u n t e r p a r t of tense logic, which has been studied a n d used by m o d a l logicians to reason a b o u t tenses in a language a n d a b o u t t e m p o r a l processes. We first describe a typical tense algebra. Let U be a set a n d R a b i n a r y relation o n U. The relation R gives rise to two u n a r y operations f a n d g on the set of all subsets of U, n a m e l y f ( X ) = the image of X u n d e r R = {u 9 U : x R u for some x 9 X} g(J() = the preimage of X u n d e r R = {u 9 U : u R x for some x 9 X}. If U is interpreted as a set of events, between events in U, then f ( J ( ) is the X, a n d g ( X ) is the set of events in the o b t a i n e d if the operations f a n d g are a n d R is interpreted as a t e m p o r a l relation set of events in the future of some event in past of some event in X. A tense algebra is added to the Boolean algebra of all subsets Presented by B. J6nsson. Received April 8, 1994; accepted in final form December 1, 1994. 404 Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 405 of U. The resulting tense algebra is the complex algebra of ~ U, R ) , namely, era(u, R) = (Sb(U), u, ,f, g). It follows from the Extension Theorem of J6nsson and Tarski [7], Theorem 2.15, that every tense algebra can be embedded in one that arises in this concrete fashion. The abstract definition of a tense algebra runs as follows. A tense algebra is an algebra of the form 9,I = (A, + , - , f , g ) , where (i) (A, + , ) is a Boolean algebra and (ii) f and g are conjugate operations, i.e., for all x, y E A y .f(x) = 0 iff x .g(y) = 0. Let N be a tense algebra. Define the binary operation - by x . y = 2 + )5 for all x and y. By (i) we may define the distinguished elements 1 and 0 by 1 = x + )? and 0~1. Property (ii) is quite strong since it implies that f and g are additive ( f ( x + y) = f ( x ) + f ( y ) and g(x + y) = g(x) + g(y)), normal ( f ( 0 ) = 0 = g(0)), and monotone (if x _<y then f ( x ) < _ f ( y ) and g ( x ) < g ( y ) ) . By Huntington [3], [4] and J6nsson-Tarski [7], Theorem 1.15, both (i) and (ii) can be expressed by the equations listed below, so the class of all tense algebras is a variety. x + ( y +z) =(x +y) + z x +y =y +x x=2+)5+2+y f(O) = 0 g(0) = 0 f ( x ) "y < f ( x "g ( y ) ) g ( y ) " x <_g ( y . f ( x ) ) . Furthermore, the last four equations can be replaced by the following two: f(x g ( y ) ) <_f(x) " ~ g ( y "f ( x ) ) <_g ( y ) " 2. 406 P. J1PSEN ET A L. ALGEBRA UNIV. A tense algebra is called reflexive if it satisfies the identity x <_f(x). An equivalent identity is x <_ g(x). To see that these two identities are equivalent, assume x <_f(x) for every x. Then x . g ~ < _ f ( x . g(x)), so x . g ~ = x . g(x) . f ( x . g(x)). But x g(x) . f ( x . g(x)) <_ x . f ( x . g(x)) _<_x . f ( x ) . 2 = 0, hence x . g(x) = 0, i.e., x < g(x). Similarly, if 91 satisfies x <_g(x), then it also satisfies x <_f(x). A tense algebra is called t o t a l i f i t satisfies the implication x # 0 ~ f ( x ) + g(x) = 1. Of course a binary relation R on a set U is reflexive if x R x for every x ~ U, and total if x R y or y R x for all x, y e U. (Note that a total relation is reflexive.) Not surprisingly such relations give rise to correspondingly named tense algebras, that is, R is reflexive on U if and only if ~ m ( & , R ) is a reflexive tense algebra, and R is total on U if and only if (!;re(U, R ) is a total tense algebra. The a t o m structure of a tense algebra 9I is the structure At(91) = ( 5 7, R ) , where U is the set of atoms of 91 and R = {(u, v) E U x U :f(u) _> v}. If 91 is complete and atomic then 91 is isomorphic to the complex algebra of its atom structure. Finite tense algebras are complete and atomic, so they are always isomorphic to the complex algebras of their atom structures. Figures 1 and 2 show the atom structures of the ten nonisomorphic total tense algebras with at most 3 atoms and at most 8 elements. The dots and circles represent atoms. Two atoms u and v are connected by a single-headed arrow from u to v if the pair (u, v) is in R. A double-headed arrow between u and v means that both (u, v) and (v, u ) are in R. The ten atom structures in Figures 1 and 2 arise by enumerating all total relations on sets with 1, 2 or 3 elements. At(To) At(T1) At(T2) /02 /~ At(T3) At(T4) Figure 1. Necessary subalgebras of finite total tense algebras. At(Ts) /\ /\ /\ /\ At(T6) At(TT) At(Ts) At(T9) Figure 2. Remaining total tense algebras with 8 elements. Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 407 Every total tense algebra is a discriminator algebra, and hence is simple. The variety generated by all total tense algebras is finitely based. It is not hard to show that a basis is given by the equations above together with x + f ( f ( x ) + g(x)) + g ( f ( x ) + g(x)) <_f(x) + g(x). See Jipsen [5] for details. Finally, we note that if 9.1 is the complex algebra of the structure { U, R ) , then every complete subalgebra of 9/corresponds to a partition of the set U with the property that the image of each block of the partition under f and g is a union of blocks. The blocks are the atoms of the subalgebra. This observation will be very heavily used in the proof of Theorem 1 below. 2. Total tense varieties of height 2 The lattice ATA of aU varieties of tense algebras has infinitely many atoms, but only one of these atoms is a variety of reflexive tense algebras, namely the variety generated by the tense algebra T o = f f m { { x } , { { x , x ) } ) . This follows from the observation that every nontrivial reflexive tense algebra has a smallest subalgebra isomorphic to T 0. We refer to this subalgebra as the constant subalgebra, since it contains only the constants 0 and 1. The variety Vat(T0) is in tuna covered by infinitely many varieties of reflexive tense algebras. However we are interested in the varieties generated by total tense algebras, since they are later shown to correspond to varieties generated by simple subadditive semiassociative relation algebras. We now prove that only four of the varieties covering Var(To) are finitely generated, i.e. generated by finite total tense algebras. The atom structures of these four finite tense algebras are shown in Figure 1. T H E O R E M 1. I f R is a total relation on a set U and [U! >4, then ~ m ( U , R ) has a proper nonconstant subalgebra. Proof. For any u, v G U, we say that "u points at v" just in case @, v) E R. The hypothesis that R is total can then be expressed by stating that for any u, v e U, either u points at v or v points at u. Since this holds even when u = v, it follows that R is reflexive. Hence, for every X _: U, f ( X ) D_X and g(X)~_X. (1) Suppose u e U and u Cf(X). Then no element o f X points at u, equivalently, g({u}) c~X = ~ . Since R is total, u points at every element of t ; hence f ( {u }) __ X. 408 P, J I P S E N E T A L . ALGEBRA UNIV. Summarizing, we have if u Cf(X) then g({u}) c ~ X = ~ and f({u}) _~X. (2) By similar reasoning we also have if u Cg(X) thenf({u})c~X = ~ and g({u}) _~ 2". (3) Choose any X __ U such that [X[ >_ 2 and X r U. Consider the partition X and {u } for every u ~ U - 3(. (4) By the hypotheses on X and U, there are at least two sets in (4), and not all of them are singletons. Consequently, if the sets in (4) form the atoms of a subalgebra of Era(U, R ) then that subalgebra is both nonconstant and proper, hence we are done, so assume otherwise. Since f(X) and g(X) both contain X, by (1), they are clearly unions of sets in (4), so there must be some u s U - X such that either f ( { u }) or g({u}) is not a union of sets in (4). Since X is the only nonsingleton in (4), it follows that either X lies partly inside f({u}) and partly outside f ( { u }), or else X lies partly inside g({u}) and partly outside g({u}). In the former case, the second conclusion of (2) is false and the first conclusion of (3) is false, while in the latter case, the first conclusion of (2) is false and the second conclusion of (3) is false. In either case the hypotheses of (2) and (3) are false. Then u ef(X) c~g(X). Hence u peints at some element x ~ X, and some y ~ X points at u. We argue next that we may assume x and y are distinct. Suppose not. Since IX[ > 2 we may choose w e X such that w r = y . If w points at u, use w instead of y. I f u points at w, use w instead of x. This shows that we may therefore assume if X _= U, ]X[ > 2, and X r U, then these exist u e U - X and distinct x, y ~ X such that u points at x and y points at u. (5) Let tl, Y _~ U. The pair (X, Y) will be called an T2-pair if the following conditions hold: Xc~ Y = ~ , f ( X ) c~f(Y) c~g(X) c~g(Y) ~_X w Y, and ] X u Y] _> 3. I f (X, Y) is a T2-pair, then X r ~ and Y r ~ , since otherwise ~ ~_ X u Y va ~ by the second condition, and either X or Y contains at least two elements by the third condition. The reason for the name is that if (X, Y) is a T2-pair then ~ m ( X w Y, R c~ ((X u Y) x (X u Y))) is isomorphic to Ta. The collection of T2-pairs is closed under unions of chains, in the following sense. Let e be an ordinal and suppose (X~, Y~ ) is a T2-pair for every ~c < e. I f X~ _ X~ and Y~ _~ Y~ whenever ~c < Z < ~, then < U K< ~X~, U ~< ~ Y~ } is also a T2-pair. Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 409 Suppose that there is a T2-pair in (U, R ) . By Zorn's Lemma there must be some maximal T2-pair (X, Y). If X w Y = U then Era(U, R ) has a nonconstant proper subalgebra isomorphic to T2, as desired. On the other hand, if X w Y r U, then we claim that ~ m ( U , R ) has a nonconstant proper subalgebra whose atoms are X, Y, and {u} f o r u e U - X - K (6) To prove this we need only check that the f-image of every set in (6) is a union of sets in (6). Since (X, Y) is a T2-pair, f ( X ) contains both X and Y, but all other sets in (6) are singletons, so f ( X ) is clearly a union of sets listed in (6). Similarly, f (Y), g(X), and g ( Y ) are unions of sets in (6). Let u E U - X - Y. It suffices to show that f({u}) either contains or is disjoint from X, and either contains or is disjoint from Y. Since R is total and X ~ ~ , either u E f ( X ) or u ~ g(X). However, if u E f ( X ) c~ g(X), then (X, Y w {u}) is a T2-pair strictly larger than (X, Y), contradicting the maximality of (X, Y). Therefore either u o f ( X ) - g(X) or u ~ g(X) - f ( X ) . If u f ( X ) - g ( X ) then f({u}) ~ X = ~ and g({u}) ___X by (3), while if u E g(X) - f ( X ) , then g({u}) c ~ X = ~ and f({u}) _~ X by (2). In any case f({u}) and g({u}) either contain or are disjoint from X. By similar reasoning we also conclude that f({u}) and g({u}) either contain or are disjoint from Y. So far we have shown that if there is a T2-pair in (U, R ) , then Era(U, R ) has a nonconstant proper subalgebra. We may therefore assume that there is no T2-pair in (U, R). (7) This means, in particular, that there can be no "4-cycle" in R, that is, no four distinct elements u~, u2, u3, u4 e U such that (ul, u2), (u2, u3), (u3, u4), (u4, ul ) e R, for then ({ul, u3}, {u2, u4}) would be a T2-pair, contradicting (7). Suppose there is a "3-cycle" in R, that is, three distinct elements ul, u2, u3 ~ U such that uj points at u2, u2 points at u3, and u3 points at ul. Let X = {ul, u2, u3}. By (5) there exist u ~ U - X and distinct x, y e X , such that u points at x and y points at u. Let z be the third element of X, distinct from x and y. Suppose y points at x. X forms a 3-cycle under R, so x points at z and z points at y (see Figure 3(i)). It follows that there is a 4-cycle, from u to x to z to y to u. The existence of a 4-cycle contradicts (7), so we may assume x points at y. This implies that the 3-cycte in X goes from x to y to z to x. There are two final cases. If u points at z (see Figure 3(ii)), then there is a 4-cycle from u to z to x to y to u, contradicting (7), but if z points at u (see Figure 3(iii)), then there is a 4-cycle from z to u to x to y to z. 410 P. JIPSEN ET AL. Xo z0 - 4 0 I/ ,.oy (i) 0~o zO= ~ --o --oy (ii) ALGEBRA UNIV. Xo IL 9 zo 0 I/' oy (iii) Figure 3 We may therefore assume that R contains no 3-cycle. Now suppose R contains a 2-cycle, that is, two distinct elements us, u2 ~ U such that ul points at u; and u2 points at u t. By (5) there is some u s U - {ul, u2} which points at one of the elements of {us, u2} and is pointed at by the other. Either way we get a 3-cycle. We may consequently assume R has no 2-cycles, that is, R is antisymmetric. So far we may assume R is reflexive, total, and antisymmetric. We may also assume R is transitive, since the only way it can fail to be transitive is to contain a 3-cycle. Hence R is a linear ordering of U. Choose any u ~ U which is not a minimum element under this ordering (where (u, v ) e R is interpreted as "u is less than or equal to v"). T h e n f ( { u } ) # g and it is easily checked that the setsf({u}) and U - f ( { u } ) form the atoms of a nonconstant proper subalgebra of gin(U, R ) isomorphic to T~. [] C O R O L L A R Y 2. In the lattice of varieties generated by total tense algebras, Var(Ti) ( i - - 1 , 2, 3, 4) are the only finitely generated covers of Var(To). Proof Note that if 9.1 and ~ are two finite subdirectly irreducible algebras in a congruence distributive variety then it follows from J6nsson's Lemma that Var(9.I) is covered by Var(~3) if and only if 9g is isomorphic to a maximal proper subalgebra of ~. As mentioned in the introduction, a finite total tense algebra is isomorphic to the complex algebra of its atom structure, so the previous theorem implies that every finite total tense algebra with more than 8 elements has a proper nonconstant subalgebra (with at most 8 dements), and hence cannot generate a variety that covers Var(T0). Figures 1 and 2 show the atom structures of the ten nonisomorphic total tense algebras with at most 8 elements. Note that Ts, T6 and T7 have a subalgebra isomorphic to T I, whereas T 8, T9 have a subalgebra isomorphic to T2 (obtained by identifying the black vertices in the atom structures). On the other hand TI, T2, T3 and T4 have no proper subalgebras other than the constant one, hence they generate covers of Var(To). [] Vol. 34, 1995 Total tense algebras and symmetric semiassociativerelation algebras 411 3. A sequence of infinite total tense algebras In this section we give countably many examples of nonfinitely generated varieties of total tense algebras that cover Var(T0). These examples are based on the veiled recession frame, used by W. Blok [2] to give similar examples of varieties of modal algebras. A structure (U, R ) is called an n-recessionframe if U = { a l , . . . , a, } and R = {(ai, aj} : i >_j} w {(ae, ai+1) : 1 <. i "< n}. If we instead consider sets U = {as, az, a 3 , . . . } or U = {. . . . a_l, ao, ai, a2 . . . . } with the natural extension of R then we obtain the co-recessionframe and the Z-recession frame respectively. A veiled recession frame is the subalgebra of fire(U, R ) generated by the singletons (or finite subsets) of Sb(U). This subalgebra is denoted by ffmy(U, R). A diagram of the Z-recession frame is given in Figure 4(i). The downward pointing bold arrow represents the linear order part of R, and is intended to indicate that, in addition to the arrows shown, there are also arrows from every element to all elements below it (as well as to itself). Observe that ~ m ( U , R ) is a total tense algebra. al,n 1 ao,n~ i a-l,n~ ,1 (i) / (ii) Figure 4. Infinite total tense algebras with no proper nonconstant subalgebras. P. J I P S E N E T AL. 412 ALGEBRA UNIV. We now define a sequence of structures (U~, R~ ) that are essentially 7/copies of the n-recession frame combined like a 7/-recession frame (see Figure 4(ii)). Let U,, = {aij : i ~ 7/, 1 <_j <_n} and define R~ = {(aij, a~,t) :i > k or (i = k a n d j > / ) } w {(ai, l a ~ + l j ) : i e Z , 1 <_j<_n} w {(aij, a~j+l):i~7/, l<_j<_n}. For n = 1 the structure is (isomorphic to) the Y-recession frame, and it is easy to check that, for fixed n and i, the set V~ = {a~j : 1 <_j _< n} together with the relation R,, c~ (V i x V~) forms an n-recession frame whose complex algebra is generated by {a;,~ }. The algebras ~3, are defined as ~ m f ( U , , R,, ). THEOREM Var(T0). 3. For each n ~ co the varieties Var(~3n) are distinct and cover Proof We first show that ~3,, has no nonconstant subalgebras. Note that an element of B, is either a finite subset of L~ or a union of a finite subset together with U~= i ~ for some k ~ co or a complement of either such sets (since this collection of sets is a subalgebra of ~ m ( U n , R, )). Let x E B, and suppose 0 < x < 1. Replacing x by its complement if necessary, we may assume that f ( x ) < 1. F r o m the definition of R~ it follows that f~(x) < 1, and it is easy to see that f3(x), f~(x) is one of the I/~ for some k. C o n s e q u e n t l y f 3+ i(x), f z + ~(x) = Ilk + i f or each i ~ co. To generate the set Vk _ ~, we choose the smallest m such that Vk _ i -~ gm(Vk), define y = gm(V1.), then fZ(y) , f ( y ) = Vk_e. Finally gZ(V~).g(V~)={ai_2,~), and now we can use the operations f and g relativized to V)_2 (i.e. f ( x ) = f ( x ) 9 Vi_2 and ~(x) = g(x) 9 Vi_2) to generate each singleton subset of Vi_ 2. So we conclude that ~B~ is generated by each element x different from ~ or U~. This however does not yet imply that Var(~3~) covers Var(To). Let 9.1 be a subdirectly irreducible member of Var(~,,) that is not isomorphic to To. We need to show that Var(~3,) ___Var(9.I). Since tense algebras are congruence distributive we have ~I ~ H~P~,(~3~), and since B~ is a discriminator algebra (x r implies f ( x ) + g(x) = 1), we conclude that all members of SPr:(~B,) are simple and that (an isomorphic copy of) 91 is among them. So we may assume that ~21 is a subatgebra of ~ / F for some index set I and some nonprincipal ultrafilter F over L Consider an element x ~ A such that 0 < x < 1. Again replacing x by its complement if necessary, we assume thatf(x) < 1. Let J = {j ~ I : f ( x j ) < 1}. Then J e F a n d f ( x j ) generates all of Bn (in the j t h coordinate) uniformly for each j ~ J. But this implies that f ( x ) generates a subalgebra of 9.1 isomorphic to ~ , , hence Var(~3,) ~ Var(91). Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 413 Since total tense algebras are discriminator algebras, the varieties V a r ( ~ n ) will be distinct if we find a universal sentence ~bn that holds in ~ , and fails in ~m for m < n. Let a =f3(x).f2(x), b =f4(x)-f3(x) and c = g(b).a = g(b). Then the sentence ~ n : O < f i x ) < l ~ a=gn(c) [] has the required property. 4. Application to symmetric semiassociative relation algebras We n o w show that tense algebras are term-equivalent to reducts o f certain semiassociative relation algebras. This allows us to extend the previous results to varieties of semiassociative relation algebras. A nonassociative relation algebra is of the f o r m 9.1 = ~A, + , , ;, ~, 1'), where (A, + , - ) is a Boolean algebra, ~ is a unary operation, 1' is an identity element with respect to the binary operation; and for all x, y, z e A x.y;z=O iff y . x ; ~ = O iff z.y;x=O. (8) ( T h r o u g h o u t we use the convention that ; has precedence over - which in turn takes precedence over + ) . The above equivalence can be expressed by equations, so the class of all nonassociative relation algebras is a variety ( M a d d u x [10], Corollary 1.5, or [ 12], T h e o r e m 2). A nonassociative relation algebra is symmetric if it satisfies the equation 2 = x. Every nonassociative relation algebra satisfies ( x ; y ) ~ = y ; 2 ( M a d d u x [10], T h e o r e m 1.13(13)), so every symmetric nonassociative relation algebra is commutative, that is, the operation ; is commutative. A nonassociative relation algebra is reflexive if it satisfies x < x ; x. I f we omit the requirement o f an identity element for a symmetric nonassociative relation algebra, then we obtain reducts (A, + , -, ;) which we will refer to as symmetric r-algebras (symmetric residuated algebras). Thus (a, + , - , ;) is a symmetric r-algebra just in case (A, + , - ) is a Boolean algebra and for all x, y, z e A. x.y;z=O iff y.x;z=O iff z.y;x=O. (9) Symmetric r-algebras are also commutative. Listed below are some other properties o f nonassociative relation algebras and symmetric r-algebras that we will use later. The first two equations assert that ; is normal, the next two that ; distributes over 414 P. J I P S E N E T A L . ALGEBRA UNIV. + , and the last two properties assert that ; is m o n o t o n e in each variable. x;0=0;x=0 x;(y+z)=x;y+x;z (x + y ) ;z = x ;z + y ; z i f x < y then x ; z < y ; z ifx<ythenz;x_<z;y. The following two results establish a term-equivalence between the variety of reflexive tense algebras and a subvariety of symmetric r-algebras. T H E O R E M 4. Let 9.1 = ( A , + , - , f , g ) be a tense algebra and define a binary operation ; on A by x ; y = f ( x . y) + x . g(y) + y . g(x). Then ( A , +, , ;) is a symmetric r-algebra that satisfies the equation x ; (x . y) < x + y . Ifg, l is reflexive then ~A, +, -, ;) is reflexive, and the o p e r a t i o n s f a n d g can be recovered f r o m ; by the term functions f ( x ) = x ; x and g(x) = x + x ; ft. z.y. P r o o f N o t e that z . x ; y = O if g(x) = 0, which is equivalent to x.y.g(z)=O and z.x.g(y)=0 and and only if z.f(x.y)+z.x.g(y)+ z.y.g(x)=O. This last statement is unchanged by permuting the variables x, y, and z, so the statements z. x ; y = 0, x . z ; y = 0 and y 9 x ; z = 0 are equivalent. To check that (A, + , -, ;) satisfies the equation x ; (`2 - y) _< x + y , we c o m p u t e x ; (2.y) =f(x. ,2.y) + x . g ( 2 . y ) +`2.y.g(x) < O+x +x.y =x +y. Assume that N is reflexive. Then x <_f(x) and x < g(x), so x ; x = f ( x " x) + x g(x) + x g(x) = f ( x ) § x" g(x) = f ( x ) + x = f ( x ) , x + x ; "2 = x + f ( x . x) + x g("2) + "2 "g(x) = x +f(0) + "2. g ( x ) = x . g ( x ) + "2 . g ( x ) = g(x), and ~A, + , -, ;) is reflexive since x < f ( x ) = x ; x. [] Vol. 34, t995 Total tense algebras and symmetric semiassociative relation algebras 415 Since the equation x;(f.y)<<_ x + y is satisfied by all symmetric r-algebras obtained in this way, we can only expect a converse for such algebras. A symmetric r-algebra that satisfies this equation is called subadditive because for disjoint x and y the relative product x ; y is below x + y. (Indeed, from x 9y = 0 and subadditivity we get x ; y = x ; ( ~ - y ) + x ; ( x . y ) = x ;(37.y) + x ; 0 = x ; (5~,y) _<x + y . ) T H E O R E M 5. Let 9.1 = (A, +, -, ;) be a subadditive symmetric r-algebra and define f(x)=x+x;x and g(x)=x+x;2. Then (A, + , - , f , g ) is a reflexive tense algebra. I f ~ is reflexive then ; can be recovered from f and g by x ; y = f ( x . y) + x . g(y) + y . g(x). Proof. We show that y .f(x) = 0 iff x . g ( y ) = O. It follows that (A, + , - , f , g ) is a tense algebra. Furthermore, (A, + , - , f , g ) is reflexive by the definitions o f f and g. Suppose y -f(x) = y 9(x + x ; x) = O. Then y - x = 0 and y . x ; x = O, hence, by distributivity, monotonicity, and subadditivity, y . x ;.f = y - x ;(x-p + 2.p) =y. x ;(x.y) + y . x ; ( 2 . y) <_y" x ;x + y . ( x +37) = 0 . This shows that x -y ; )7 = 0 by (9) and therefore x- g(y) = x . (y + y ; )7) = 0. For the converse, suppose x . g ( y ) = 0 . Then x - y = 0 and x . y ; f = 0 . The latter equation implies x ;y = y ; x <_y by commutativity and (9). It follows that x - x ; y _ < x - y = 0 , hence also y . x ; x = 0 . Consequently, y . f ( x ) = y 9(x + x ; x) = 0. So far we know that (A, + , - , f , g ) is a reflexive tense algebra. Next we show that f(x (lO) . y) + x. g(y) + y . g(x) = x ; y + x . y. First we expand the left side of (10). f ( x .y) + x . g ( y ) + y .g(x) = x .y + ( x .y) ; ( x - y ) + x , ( y + y ;2) +y.(x+x;2) = x . y + ( x - y ) ;(x .y) + x . y ;)5 + y . x ;2. Next, by distributivity, monotonicity, and subadditivity, x.y ;p =x.y ;(p.x) +x.y ; ( y . 2 ) _<y ; x + x - ( y + 2 ) = x ; y + x . y , (11) 416 P. JIPSEN ET AL. ALGEBRA UNIV. and, similarly, y- x ;2 _<x ;y + x . y , SO x.y;2+y,x;2 <x;y+x.y. Combining the previous equation with (11) (x- y) ; (x- y) _< x ; y, we get half of (10), namely, and the observation that f ( x . y) + x . g ( y ) + y . g(x) <_x ; y + x . y. For the inclusion in the other direction, we first expand x ;y. x ;y =(x.y +x.y) ;(x.y +2.y) = (x. y) ; (x. y) ~2 (x.)7) ; (x. y) + (x. y) ; ()~. y) + (x. fi) ; ()~. y). Next we show the last three terms in the expansion of x ; y x- y ; 2 + Y" x ; 2. From monotonicity we have (x. 2) ; (x. y) _<2 ; y and by subadditivity, (x2) ;(x-y) _<x.y +x.y =x SO (x "2) ; (x. y) _<x 2 ;y. In a similar way we also get (~-y) ; (~. y) ~ y . x ;~. By monotonicity, (x-)9) ; ( i f . y ) _<x;)?.37 ; y and, by subadditivity, (x 2 ) ;(~ y ) _<x-2 + - ~ y _<x + y are included in Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 417 SO (x-y) ; ( ~ - y ) <_(x + y ) . x ; x . y ; y =x.x ;x. y ;y + y.x ;Yc..f ;y <_x. ~ ; y + y . x ;92. We may now conclude that x;y~(x.y);(x-y)+x-yLg+y.x;:2. From this last equation and (11) we get the second half of (10), so (10) holds. Now if 91 is reflexive, then x - y _< (x. y) ; (x- y) _< x ; y, hence x ; y + x - y = x ;y, so by (10), f ( x . y) + x . g ( y ) + y . g ( x ) = x ; y. as desired. [] C O R O L L A R Y 6. The variety o f reflexive tense algebras is term-equivalent to the variety o f reflexive subadditive s y m m e t r i c r-algebras. Proof Let 9.I = (A, + , - , f , g ) be a reflexive tense algebra. Define ; by x ; y = f ( x . y) + x . g ( y ) + y - g(x). By Theorem 4, (A, + , -, ;) be a reflexive subadditive symmetric r-algebra in which the operations f, g are definable by f ( x ) = x ; x and g(x) = x + x ; Yc. For the converse, let ~I = (A, + , - , ;) be a reflexive subadditive symmetric r-algebra. Define f and g by f(x) = x ; x and g ( x ) = x + x ; 92. Reflexivity implies f ( x ) = x + x ; x , so by Theorem 5, (A, + , - , f , g3 is a reflexive tense algebra in which ; can be recovered from f and g by x ; y = f ( x . y ) + x- g(y) + y- g(x). D To establish a connection between subadditive symmetric r-algebras and total tense algebras we require the semiassociative identity (x;l);l=x;1 introduced in Maddux [9]. 418 P. JI[PSEN ET AL. A L G E BRA UNIV. T H E O R E M 7. The variety generated by all total tense algebras is term-equivalent to the variety of reflexive subadditive semiassociative symmetric r-algebras. Proof. Let ~A, + , - , J ~ g > be an algebra in the variety generated by all total tense algebras, so that ~A, + , - , f , g ) is a tense algebra which satisfies the identity x + f ( f ( x ) + g(x)) + g ( f ( x ) + g(x)) <_f(x) + g(x). (12) Let h(x) = f ( x ) +g(x) for every x e A. Then (12) is equivalent to x + h(h(x)) _< h(x). (13) A consequence of (t3) is that ~A, + , - , f , g ) satisfies x <_h(x). From the latter identity we can prove that ~A, + , - , f , g ) is reflexive. To see this, let z = x .f(x). Then z = x .f(x) < h(x .f(x)) = f ( x -f(x)) + g(x .f(x)) < f ( x ) + ff = 2. From z _< ~ it follows that z = 0, hence x <_f(x). Define the binary operation ; on A by x ; y = f ( x . y) + x . g ( y ) + y . g(x). (14) By Theorem 4, (A, + , - , ;) is a reflexive subadditive symmetric r-algebra with f ( x ) = x ; x and g ( x ) = x + x ; ~ . We will show (A, + , - , ;) also satisfies the semiassociative identity. From (14) and the definition of h we have x ; 1 = f ( x . 1) + x . g ( 1 ) + 1 .g(x) =h(x) + x - g ( t ) . Since (A, + , - , f g ) is a tense algebra, the operations f and g are monotone and distribute over +. It follows that the same properties apply to h, so we can derive one half of the semiassociative identity from a consequence of (13), namely h(h(x)) <_h(x), as follows: (x ; 1) ; 1 = h(h(x) + x . g ( 1 ) ) + (h(x) + x . g(1)) . g ( 1 ) = h(h(x)) + h(x" g(1)) + h(x)" g(1) + x . g(l) < h(x) + x . g(1) =x;1. Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 419 For the other direction, note first that x < h(x) by (13), hence x <<_h(x) + x. g(1) = x ; 1 for every x in A. Replacing x by x ; 1 yields x ; 1 < (x ; 1) ; 1. Now suppose that {A, + , - , ;) is a reflexive subadditive semiassociative symmetric r-algebra. Define f and g by f(x) = x ; x and g ( x ) = x + x ; x. Reflexivity implies f ( x ) = x + x ; x, so (A, + , -, f, g ) is a reflexive tense algebra and x ; y = f ( x . y ) + x . g ( y ) + y . g ( x ) by Theorem 5. We use the semiassociative identity to derive (13). First note that h(x) = f ( x ) + g ( x ) = x ; x + x + x ; 2 = x+x;(x+Y)=x+x;1. Then x + h(h(x)) = x +h(x + x ; 1) =X +(x+x;1)+(x+x; zX +x; 1);1 1 + ( x ; 1); 1 =x+x;1 = h(x). Since (A, + , - , f , g ) is a reflexive tense algebra satisfying (13), which is equivalent to (12), it is in the variety generated by all total tense algebras. [] Before we extend our results about total tense algebras to subadditive symmetric semiassociative relation algebras, we recall what is known about the bottom of the lattice ASA of semiassociative relation algebra varieties. J6nsson and Tarski [8] proved that there are exactly three atoms, generated by the algebra 211,212 and 213, in the lattaice ARA of varieties of relation algebras. Later Andrhka, J6nsson, and N6meti [ 1] pointed out that the proof given there requires only the semiassociative identity, so the same result holds in AsA. Furthermore, Var(211) has no join-irreducible covers and Var(212) has exactly one join-irreducible cover, generated by the relation algebra of all binary relations on a 2-element set. T H E O R E M 8. L e t 9.1 = ( A , +, -, ;, ~, 1') be a f i n i t e simple subadditive s y m m e t ric semiassociative relation algebra with m o r e than 4 atoms. Then 21 has a proper n o n c o n s t a n t subalgebra. A subalgebra of 21 is "proper", of course, if it is not the same as 21, but the meaning of "nonconstant" is slightly different in this case, because semiassociative relation algebras have a distinguished identity element 1'. Indeed, every simple Proof 420 P. J I P S E N E T AL. ALGEBRA UNIV. finite semiassociative relation algebra has a subalgebra whose only elements are 0, 1, 1', and 1', and it is this subalgebra that we refer to as the "constant subalgebra" of 91. A nonassociative relation algebra is said to be integral if x ; y r 0 implies either x r 0 or y ~ 0. If 9.1 is not integral, then 91 has a proper nonconstant subalgebra by M a d d u x [11], T h e o r e m 3. We m a y therefore assume that 91 is integral. By M a d d u x [11], T h e o r e m 4, this is equivalent to assuming that 1' is an atom. Next we show that we m a y also assume 91 is reflexive. Let 9 1 ' = (A, + , -, ;', ~, 1') where, for all x, y ~ A', x;'y=x;y+x.y. We claim that 91' is a reflexive finite integral symmetric semiassociatve relation algebra with exactly the same subalgebras as 91. Obviously 91' satisfies 2 = x since 91 does so. Hence to prove that (8) holds for 91' we need only show (9) holds for 9.1', namely, x . y ;'z =O iff y . x ;'z =O iff z . y ; ' x =O. Suppose x . y ; ' z = O . Then O = x . ( y ; z + y . z ) = x . y ; z + x . y . z , so 0 = x . y ; z and 0 = x 9y 9z. Using (8) and the s y m m e t r y of 91, we get 0 = y . x ; z, so 0=y'x;z+x'y'z=y'(x;z+x'z)=yx;'z. The other equivalences can be proved similarly. Next, x;'l'=x;l'§ l'=x§ l'=x, and, similarly, 1' ;' x = x, so 1' is an identity for ;'. To show that semiassociative identity holds for 91', use the assumption that it holds for 91 to get (x;'l) ;'1 = ( x ; l+x. 1);'l=(x;1 +x);l+(x;l+x)- 1 =(x;1);l+x;l+x=x;l+x=x;'l. So far we have shown that 9.1' is a symmetric semiassociative relation algebra. Obviously 9.I' is finite. 91 and 91' have the same Boolean algebra as reduct, so 1' is an a t o m in both, hence 91' is integral. 91' is reflexive since x < _ x ; x + x = x ; x + x 9 x = x ;' x. W h a t remains is to show 91 and 91' have the same subalgebras. The operations of 91' are term-definable from those of 91, so every subalgebra of 91 is a subalgebra of 91'. F o r the converse, assume ~ is a subalgebra of 91'. We wish to show the universe B of ~ is closed under the operations of 91. The only Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 421 operation o f 92 which is n o t also an operation o f 92' is ;. Hence we need only show that B is closed under ;. Let x, y E B. First expand x ; y as follows. x ;y = (x .y) ; (x. y) + ( x - y ) ; (x. y) + (x -y) ; (2. y) + (x. y) ; (~ .y). N o t e that ; and ;' coincide on disjoint arguments, that is, if u . v = 0, then u;'v=u;v+u.v=u;v+0=u;v. We can apply this observation to the last three terms in the expansion o f x ; y, obtaining x ; y = (x. y) ; (x .y) + (x. y) ;' (x- y) + (x. y) ;' (~. y) + (x 2 ) ;' (x "y). Thus we need only show ( x . y ) ; ( x - y ) s B. Let z = x ' y . W e claim that either z<z;z, in which case z ; z = z ; ' z s B , or else z - z ; z = 0 , in which case z ; z = z ;' z . 5 e B. Suppose b o t h cases fail. Then u r 0 ~ v, where u = z 9 z ; z and v = z.z;z. Since 9.1 is integral and subadditive, we have 0 ~ u ; v < u + v. But u ; v < < _ z ; z and u < z ; z , so O # u ; v < _ ( u - z ; z + v ) . z ; z = v . z ; z < v . Thus 0 r u ; v- v. By (8) and sDnmetry, this yields 0 r v ; v- u. But v _< z and u _< z ; z, so v ; v 9 u _< z ; z 9z ; z = 0, a contradiction. This completes the p r o o f o f our claim that 92' is a reflexive finite integral symmetric semiassociative relation algebra and 9.I' has the same subalgebras as 92, so we might as well assume that 92 is reflexive. Let 9 2 ' = ( A ' , + , - ' , ;') be the algebra whose universe A ' is defined by A ' = {x ~ A : x < 1'}, where, for all x, y ~ A', x-'=2. x, 1', y=x;y We claim that 92' is a subadditive reflexive symmetric r-algebra. First o f all, (A, + , - ' ) is a Boolean algebra. Since 9.1 is symmetric, (9) holds. W h e n (9) is applied to 9.1', that is, with ; replaced by ;', it is equivalent to x-y;z=l'=O iff y . x ; z . 1'=0 m iff _ _ z.y;x- i'=0. _ _ But for x, y, z ~ A ' we have x = x 9 1', y = y 9 1', and z = z 9 1', so the latter condition actually coincides with (9). Thus 92' is a symmetric r-algebra. The subadditivity and reflexivity o f 92' follows immediately f r o m the subadditivity and reflexivity o f 92. By T h e o r e m 5, 92' is term-eqtfivalent to a reflexive tense algebra 9 2 " = (A',+ -,f,g), wheref(x)=x+x, x=x+x x . 1' and g ( x ) = x + x ; x = 422 P. J I P S E N ET AL. ALGEBRA UNIV. m x+x;(21'). 1'. N o t e that i f 0 r thenx<l' andx;l=l since 9.1 was assumed to be integral, so f ( x ) + g ( x ) = x + x ; ( x . 1--). l ' + x ; ( 2 . 1'). 1 ' = x + x ; 1'- l ' = x ; l ' l'+x;l', l'=x;1. 1 ' = 1'. But 1' is the unit element o f 9.I', so 9.1" is total. 9.I' has more than 3 atoms, so we can apply T h e o r e m 1 to the a t o m structure of ~I" and conclude that 9,1" has a proper n o n c o n s t a n t subalgebra. O f course the same conclusion holds for 91', since 91' is term-equivalent to 96". Let B' be the universe of this subalgebra of 9.1', and define B = B' t2 {x + 1' ~ A 9 x ~ B}. Then, for all x, y e B', we have x ; y = x ; ' y + x ; y . 1', (x+l');y=x;'y+y+x;y. l', (x + 1') ; ( y + 1') = x ; ' y + x + y + 1'. These equations, together with the assumption that 1' is an atom, show that (B, + , -, ;, 1') is a proper nonconstant subalgebra of ~1. [] Since the preceding theorem did not assume reflexivity, there are more than four finitely generated varieties covering Var(gX3). The exact n u m b e r is f o u n d counting the n u m b e r o f nonisomorphic structures that can be obtained by deleting loops (x, x ) from the a t o m structures o f T1, T2, T3 and T 4. Nonfinitely generated covers o f Var(9.13) can be constructed from the total tense algebras ~3n o f T h e o r e m 3 via the term-equivalence, similar to the p r o o f above. C O R O L L A R Y 9. In the lattice o f varieties o f subadditive symmetric semiassociatire relation algebras there are exactly 19 finitely generated varieties, and infinitely many nonfinitely generated varieties, that cover Var(~13). REFERENCES [I] ANDRI~KA,H., JONSSON, B. and N~METI, I., Free algebras in discrimina:gr varieties, Algebra Universalis 28 (1991), 401 447. [2] BLOK,W. J., The lattice of modal logics: an algebraic investigation, J. of Symb. Logic 45 (2) (1980), 22l -236. [3] HUNTINGTON,E. V., New sets of independent postulates for the algebra of logic, with special reference to Whitehead and Russell's Principia Mathematica, Transactions of the American Mathematical Society 35 (1933), 274 304. [4] HUNTINGTON, E. V., Boolean algebra. A correction, Transactions of the American Mathematical Society 35 (1933), 557-558. [5] JIPSEN, P., Discriminator varieties of Boolean algebras with residuated operators, in: Cecylia Rauszer (ed.), Algebraic Methods" in Logic and in Computer Science, Banach Center Publication, Vol. 28, Institute of Math., Polish Academy of Sciences, Warszawa, 1993. Vol. 34, 1995 Total tense algebras and symmetric semiassociative relation algebras 423 [6] JlPSEN, P. and LUK~CS, E., Minimal relation algebras, Algebra Universatis 32, no. 2 (1994), 189 203. [7] J6NSSON, B. and TARSKI, A., Boolean algebras wtih operators, Part L American Journal of Mathematics 7.3 (1951), 891-939. [81 J6NSSON, B. and TARSKI, A., Boolean algebras with operators, Part II, American Journal of Mathematics 74 (1952), 127-162. [9] MADDUX, R. D., 7bpies in relation algebras, Doctoral dissertation, University of California, Berkeley, 1978. [10] MADDUX, R. D., Some varieties containing relation algebras, Transactions of the American Mathematical Society 272, no. 2 (1982), 501-526. [tl] MADDUX, R, D., Necessary subatgebras of simple nonintegral semiassociative relation algebras', Algebra Universalis 27 (1990), 544-558. [ 121 MADDUX,R. D., Pair-dense relation algebras, Transactions of the .American Mathematical Society 328, no. 1, (199t)~ 83-t3L [ 13] TUZA, Z., Representations of relation algebras and patterns of coloured triplets, in: Algebraic Logic (Proc. Conf. Budapest 1988, ed. by H. Andrbka, J. D. Monk, and I. N6meti) Colloq. Math. Soc. J. Bolyai Vol. 54, North-Holland, Amsterdam (1991), 671-693, Department of Mathematics Iowa State University Ames, IA 50011-2066 U.S.A.