Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Structure (mathematical logic) wikipedia , lookup
History of algebra wikipedia , lookup
Birkhoff's representation theorem wikipedia , lookup
Congruence lattice problem wikipedia , lookup
Oscillator representation wikipedia , lookup
Complexification (Lie group) wikipedia , lookup
Representation theory wikipedia , lookup
Boolean algebras canonically defined wikipedia , lookup
Fundamental theorem of algebra wikipedia , lookup
proceedings of the american mathematical society Volume 111, Number 4, April 1991 SPLITTING IN RELATION ALGEBRAS H. ANDRÉKA, R. D. MADDUX, AND I. NÉMETI (Communicated by Andreas R. Blass) Abstract. We define a way, which we call splitting, of getting new relation algebras from old ones. We characterize those algebras to which splitting can be applied. We show how to split representable relation algebras in order to obtain nonrepresentable ones, and we give many examples. Introduction The method of splitting atoms, originating with L. Henkin, is well known in cylindric algebra theory and is used to obtain nonrepresentable cylindric algebras from representable ones (see [HMT, 3.2.67 and 3.2.69]). In this paper we adapt this method to relation algebras. The conditions for splittability in relation algebras seem to be more complex than in cylindric algebras, where every atom below d(a x a) can be split. We use the terminology and notation of [Mai]. In particular, we will deal with the varieties NA, WA, SA, and RA, of nonassociative, weakly associative, semiassociative, and associative relation algebras. These varieties are defined in [Ma 1] and are obtained from the variety RA of relation algebras by weakening the axiom postulating associativity of relative multiplication. If 21 is any Boolean algebra with additional operations, then At 21 denotes the set of all atoms of the Boolean reduct of 21. Finally, our convention is that ; binds more strongly than •. For example, x ; x • 0' denotes the term (x ; x) • 0'. 1. Definition and basic properties of splitting Definition 1. Let 21 and 03 be atomic NA's. We say that 21 is obtained from 03 by splitting if the following three conditions are met: (1) 213»; (2) every atom x of 21 is contained in an atom c(x) of 03, called the cover of x ; and Received by the editors October 30, 1988 and, in revised form, May 18, 1990. 1980 Mathematics Subject Classification (1985 Revision). Primary 03G15. Research supported by the Hungarian National Foundation for Scientific Research grant No. 1810. ©1991 American Mathematical Society 0002-9939/91 $1.00+ $.25 per page 1085 License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use H. ANDREKA, R. D. MADDUX, AND I. NÉMETI 1086 (3) for all x, y e At 21, if x, y < 0' then = (c(x);c(y)-0' if x ¿ y, \c(x);c(y) ifx = y. Let an and as be functions mapping At 03 to cardinals. We say that 21 is obtained from 03 by splitting along an and as if 21 is obtained from 03 by splitting, and for all x G At 03, an(x) = \{yeAt?í:y<x,y¿y}\ and as(x) = \{yeAt<A:y<x,y = y}\. The following lemma says that the result of splitting a complete atomic NA along two functions, if it exists, is unique up to isomorphism. Its proof is straightforward and is therefore omitted. Lemma 2. Let 21, 2l', and 03 be complete atomic NA's. Let an and as be functions mapping At 03 to cardinals. If 21 and 2t' are obtained from 03 by splitting along an and as, then 21 and 2l' are isomorphic by an isomorphism which leaves 03 fixed. The next theorem characterizes those functions along which an atomic NA, WA, SA, or RA can be split so as to obtain an algebra in the same class. Intuitively, in NA we have only the obvious conditions, namely, (a) in Theorem 3; in WA, the additional condition is that we cannot split "nondiscrete" identity elements; in SA, the additional condition is that we cannot split nondiscrete functional elements; and in RA we have a slightly stronger condition. If 21G NA, then Fn2t is the set of functional elements of 21. Thus Fn2l = {x G A :x ;x < 1'}. Theorem 3. Let 03 be an atomic NA. Let an, as be functions mapping At 03 to cardinals, and let a(x) = an(x) + as(x). Consider conditions (a)-(d): (a) For all x e At 03, a(x) > 1, an(x) — an(x), if x < 1 ', then an(x) = 0; if x = x, then a„(x) = Ik for some k ; and if x ± x, then as(x) = 0. (b) For all x G At 03, if a(x) > 1 and 0' ; x ^ 0, then x < 0 '. (c) For all x e At 03, if a(x) > 1 and 0 ' ; x ^ 0, then x £ Fn2l. (d) For all x,y € At®, if a(x) > 1, y ;x ¿ 0, and y < 0', then y < y;(x;x-0r). Then (i) There is an atomic 2t G NA obtained from 03 by splitting along an and as iff (a) holds. Now suppose 21G NA, 21 is atomic, and 21 obtained from 03 by splitting along an and as. Then (ii) 21G WA iff 03G WA and (b) holds, License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use SPLITTING IN RELATION ALGEBRAS 1087 (iii) 21G SA iff'Be SA and (c) holds, (iv) 21g RA //J 03g RA and (d) holds. Proof. First we prove the "only if parts. Assume 21 and 03 are atomic NA's and 21 is obtained from 03 by splitting along an and a. Then (a) holds by the elementary laws governing " in NA's (namely, by [Mai, 1.13(16)(12)(8)]). Clearly, if 21 G WA, then 03 G WA by Definition 1(1), and similarly for SA and RA. Next we show that (b), (c), or (d) holds whenever 21 is in WA, SA, or RA, respectively. Assume 21 G WA, 1' > x G At03, and 0';x ^ 0. Then there is some y e At 03 such that y < 0' and y ; x ^ 0. Then y ; x = y since x < 1' and y G At03. Recall that yr = y ;y ■V and yd = y ;y ■V (see [Mai, 5.10]). Then y = x by [Ma2, 5.12(3)]. Suppose y > y0 e At21. Then yr0= x by Definition 1(3); hence, x G At 21 since y0 e At SI, by [Ma2, 5.12(1)]. Therefore a(x) = 1 . Assume 21G SA, x G At 03, a(x) > 1, and 0' ; x ¿ 0. Then x < 0' by (b), and there are x0, xx e At21 such that x0 ^ xx and c(x0) = c(xx) = x. Then Xg = xáx= xd by Definition 1(3), so xx = xx ; xx = xd ; xx = (x0 ; x0 • 1') ; xx < (1 ; x0) ; xx = 1 ; (x0 ; x,) by [Mai, 5.11(3)], and the semiassociative law. Thus x0 ; x¡ ^ 0. We have x0 ; xx = x ; x • 0' by Definition 1(3), so x ;x £ 1'. To prove that (d) holds when 21G RA, we will need (*) below. We note that (*) is true if 21G NA. (*) If x G At03, y e At21, and x, y < 0', then x ;y = x ; c(y). To prove (*), assume x G At03, y e At21, and x, y < 0'. Clearly x ;y < x ; c(y). For the opposite direction, first note that if x = c(y), then x ; c(y) = c(y) ; c(y) = y;y < x;y. Assume x ^ c(y). Choose z e At21 so that x = c(z). Then z ^ y and x ^ c(yj, and hence x ; c(y) = c(z) ; c(y) • 0' = ^ ; y < x ; y . (*) has been proved. Assume 21GRA, x,j>GAt03, y;x^0, y < 0', and a(x) > 1. Then x < 0' by (b). Let x0, x, G At 21 be such that x0 ^ xx and c(x0) = c(Xn) = x . Then x0 ;xx = x ; x-0' by Definition 1(3), and 0 5¿y ;x = y ; c(x0)-y ; c(xx) = y ; x0 • y ; xx by (*), so 0 / y ■(y ; x0) ; xx = y ■y ; (x0 ; x,) = y ■y ; (x ; x • 0') by the associative law, and therefore y < y ; (x ; x • 0'). We now turn to proving the "if parts. Let 03, an, as, and a be as in the statement of the theorem, and assume (a) holds. We may assume that 03 is complete. We will construct a complete atomic 21 G NA that is obtained from 03 by splitting along an and as. Then we prove that 21G WA if (b) holds and 03 G WA, that 21G SA if (c) holds and 03 G SA, and that 21G RA if (d) holds and 03 G RA. First we will construct a copy of 21, namely St'. It will be the complex algebra of a certain relational structure (U, C, f, I), where C ç U x U x U, f is an involution of U, and I ç U. (See [Ma2, Definition 2.1].) For every x G At 03, let S (x) and S (x) be sets such that \Sn(x)\ = a (x), \S (x)\ - a (x), and, License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 1088 H. ANDREKA, R. D. MADDUX, AND I. NÉMETI for all i, j e {n, s} and all x, y e At03 , we have S¡(x) nS ¡(y) = 0 if i =¿j or x / y . Set U = \J{Sn(x)uSs(x) : x e At 03}. Let / be a fonction mapping U to U such that for all y e U we have ff (y) = y , f(y) = y iff y e Ss(x) for some x G At03, and for all x G At03, if y e Sn(x), then f(y) e Sn(x). Such a fonction / exists by (a). Let c be the fonction mapping U to At03 defined by c(y) = x if y e Sn(x) U Ss(x). For all x, y, z e U ,let [x,y, z] = {(x,y, z), (fx, z, y), (y, fz, fx), (fy, fx, fz), (fz,x, fy), (z, fy, x)}. Let C=\J{[x,y,z]: c(x) ; c(y) > c(z) and c(x), c(y), c(z) < 0'} U (J {[x, y, y] : c(x) ; c(y) > c(y), c(x) < V, and either c(y) < 0'or y = x) , and let I = {x:xeU and c(x) < 1'} . Set 2i' = €m(U, C, f, I). It follows directly from the definitions and [Mai, 2.2(2)] that Si' G NA. For every b e B let h(b) = {y : y e U and c(y) < b}. It is easy to check that « is an embedding of 03 into 2l', and that 2l' is obtained from the image of 03 by splitting along an and as. We may then obtain 21 by replacing the image of 03 in Si' with 03 itself. Now assume (b) and 03 G WA. To show Si' G WA it suffices to check condition [Mai, 2.2(d)]. Assume w e I and (w, x, x) e C and (x,y, z) e C. Then c(w) ; c(x) > c(x) and c(x) ; c(y) > c(z ) in 03 ; hence, c(w) ; c(z) > c(z), since 03 G WA . If c(z) < 0', then (w, z, z) e C, and we are done. Assume c(z) < V. Then c(w) = c(z). If c(x) < V, then x = y = z, since (x, y, z) e C. Hence (w, z, z) e C, and we are done. If c(x) < 0', then c(xj < 0' and 0 ^ c(xj < c(xj ; c(w), so a(c(w)) = 1 by (b). Hence z = w , since c(z) = c(w), so (w , z, z) e C, and we are done. Now assume (c) and 03 G SA. We want to check [Mai, 2.2(e)]. Let (v , w , x), (x, y, z) e C. We have to find some u e U for which (v , u, z) e C. We get (b) from (c), since for every x G At 21, either x < 1' or x < 0', but if x < V then x G Fn2l. From what has been proved so far, we have 2l' G WA, and [Mai, 2.2(d)] holds. If v e I, then w = x by the definition of C, so we have (v , x, x), (x, y, z) e C, and hence (v , z, z) e C by 2.2(d), so u = z will do. From (v , w , x), (x, y, z) e C, we get (z, f(y), x), (x, f(w), v) e C. If z e I, then f(y) = x, so (z, x, x), (x, f(w), v) e C; hence, (z, v, v) e C by 2.2(d), and finally (v, f(v), z) e C, so u = f(v) will do. Therefore we may assume v £ I and z £ I. Since (v, w, x), (x, y, z) e C, we have c(v);c(w) > c(x) and c(x) ; c(y) > c(z). Since 03 G SA, c(z) < (c(v) ; c(w)) ; c(y) < (c(v) ; 1) ; 1 = c(v) ; 1, so there is some c(u) G At03 such that c(v);c(u) > c(z). If c(u) < 0', then (v,u, z) e C, and we are done. Assume c(u) < V. Then c(v) = c(z). If v = z, then (v ,u, z) = (v,u,v) G [u, f(v), f(v)] Ç C. If v ¿ z, then a(c(v)) > 1 , License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 1089 SPLITTING IN RELATION ALGEBRAS and c(u) < c(z)~ ; c(v) < 0' ; c(v), so c(v) £ Fn2l by (c). Hence there is some c(u) G At 03 such that c(u') < 0' and c(v) ; c(u) > c(z). Now (v, u , z) e C, and we are done. Assume (d) and 03 G RA. To check [Mai, 2.2(f)], let (v, w, x), (x, y, z) e C. We have to find some u e U for which (v, u, z), (w , y, u) e C. First note that (b) holds, since clearly (d) implies (c), so 2l' G WA and 2.2(d) holds. If v , w , y , or z is in / then let u be z , y , w or f(v ), respectively. In each of these four cases it is easy to show, by arguments similar to those in the previous paragraph, that (v , u, z), (w , y, u) e C, using the fact that 2.2(d) holds. Therefore assume v , w, y, z £ I, and hence c(v), c(w), c(y), c(z) < 0'. Because 03 G RA, there is some c(u) e At03 such that c(v);c(u) > c(z) and c(w);c(y) > c(u). If c(u) < 0', then we are done. Assume c(u) < V. Then c(v) = c(z) and c(w) = c(yj. If v = z and w = f(y), then (v , u, z), (w , y, u) e C and we are done. Assume v ^ z . (The case w ^ f(y) is completely analogous, and therefore we omit it.) Then a(c(v)) > 1 . By (v , w , x) G C and (d), we have c(y) < c(w)~ ; (c(vj ; c(z) • 0'), so there is some c(u') G At 03 such that c(y) < c(wj ; c(u) and c(u') < c(vj ; c(z) • 0'. Then c(u') < c(w) ; c(y) and c(z) < c(v) ; c(u), so (v , u , z), (w , y, u) e C and we are done. D We now turn to investigating the condition of splittability in RA's. Definition 4. Suppose 21G RA, 21 is atomic, and x G At 21. We say that x ¿s splittable in 21 if x < 0' and y < y ; (x ; x • 0') whenever 0' > y G At 21 and y ;x ^ 0, and y < (x ; x • 0') ; y whenever 0' > y e At2l and x ; y ^ 0. We say that 21 is splittable iff 21 has a splittable atom. By Theorem 3, x is splittable in 21 iff there is some 03 G RA obtained from 21 by splitting such that x ^ At 03 . Note that if x is splittable then there is no y e Fn2lDAt2l such that y < 0' and either y = xd or yö = xr. The following lemma says that the converse holds if x = x ; 1 ; x . Lemma 5. Let 21G NA, x G At SI. Consider statements (l)-(4): ( 1) for every yGAt2l, if y ; x ¿ 0 and y < 0 ', then y < y ; (x ; x ■0 "); (1) for every y e Fn2ln At SI, (0'-y) ;x = 0; (3) for every y e Fn2l n At21, if y < 0' then y # xd ; and (4) for every y e Fn2l n At 21, if y = xd, then y = xd. Then (1) implies (1), (3), and (4). If 21G SA and x = x ; 1 ; x, then (l)-(4) are all equivalent. Proof. It is easy to show that (1) implies (2) in every NA. Assume SI e SA and x = x ; 1 ; x . Assume that (2) holds. want to show that y < y ; (x ; x • 0'). since y e At 21. From y e At 21 and Using these equations, 21G SA, and y;y<y;lA;y<x;l;lA;(l;x) Let y G At 21, y ; x ^ 0, and y < 0'. We We have y £ Fn2l by (2), so y < y ; 0', y ; x / 0 we get y < 1 ; x and y < x ; I. x = x ; 1 ; x, we get = x;lA;x<(x;l;x);x License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use = x;x, 1090 H. ANDRÉKA, R. D. MADDUX, AND I. NEMETI so y = y ■y ; 0' < y ; (y ; y • 0') < y ; (x ; x • 0'), as desired. It is easy to see that if 21G SA then (2)-(4) are equivalent by the elementary laws governing x and xr [Mai, 5.11 and 5.12]. D 2. Examples Examples 1, 2, and 3 show that the condition x = x ; 1 ; x cannot be omitted in Lemma 5. Examples 4, 5, and 6 show how to split atoms in order to obtain nonrepresentable RA's from representable ones. Example 1. Let 21 be the following RA: AtSl = {1\ a, ¿}, a = a, b = b, a; a = b + 1', b ; b = a + V, and a;b = b;a = a + b . It is fairly easy to prove that 21 has only one representation, the one given in Figure 1. (More precisely, 21 is embeddable in 9\tU, the algebra of all relations on U, just in the case that |t/| = 5 and the relations correlated with a and b look like those in Figure 1.) Now 21 is not splittable, because a -¿ a ; (b ; b • 0') and b jÇb ; (a ; à • 0'). Example 2. Nonminimal RA's in which all the atoms are equivalence elements are not splittable, because of the following: Suppose 21 G RA, 21 is atomic, |At2l| > 3, and for every a G At 21, à = a and a;a = a + V. (It follows that 1' G At 21.) Let 0' > a e At SI. Then there is some b e At 21 such that b < 0' and a ^ b. Now a ; ä • 0' = a, but b £ b ;a, showing that a is not splittable. Examples of such algebras are the Lyndon algebras (see [L] or [J]). In a Lyndon algebra, if a and b are atoms distinct from each other and from 1', then a = a , a ; a = a + V, and a ; b = (a + b + V)~ . Example 3. We now examine the three minimal RA's. We will see that two of them are splittable and that all the algebras obtained from them by splitting are representable. Let 21 be a minimal RA. Then At2l ç {V, 0'} and 0'; 0' G {0, 1', 1}. If 0' = 0, then 1' = 1 . Hence 21 is splittable. The algebras obtained from 21 by splitting are all Boolean relation algebras, and hence are representable. Assume 0';0' = 1'. Then 0' G FnSl; hence, 1' and 0' are not splittable in 21 since a / •.-./■• \ a. 6A.•;.• W VI a ! b b \ a \/AXA\/ •'-a-• Figure 1. Representation of 21 License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 1091 SPLITTING IN RELATION ALGEBRAS RA ¡ \ kL R" •¡A.■/■■■ S'X.••• \\ ?' R' /"'S' jR'Í •1y-.f»S""\.■:.• k ..••••""/ S'"'\ / \ v "•••s" k' .■■•""/ R' R" \zXy ¡"'S". S'r'\ \//\y -R'-'■'• R" ••■-R' Figure 2 0' ; 0V 0 and 0' ; 1' ¿ 0. Thus SI is not splittable. Assume 0' ; 0' = 1. Then 0' is splittable in 21 and the algebras 03 obtained from 21 by splitting are the following: At© = {V} U S U N, à = a for all a e S, à # a for all a e N, and a ; a = 1, a ; b = 0' for all a, b G S U N with a ^ b . These algebras are all representable by [Mai, 5.19]. In the case N = 0, these algebras are called £„({1, 2, 3}) in [Mai, 2.4], where « = |At03|. Now we use splitting to obtain nonrepresentable ones. RA's from representable Example 4. Let 21 be the following proper relation algebra: Set U = U' U U" , where U' CIU" = 0 and \U'\ = \U"\ = 5. Let R', S', R", S" be the relations shown in Figure 2, and let 21 be the proper relation algebra with atoms {Idy,, R!, S', Id^» , R", S", U' x U", U" x U'} . Now by Lemma 5, II' x U" is splittable in 21. Let 03 be the RA obtained from 21 by splitting U' x U" into at least six atoms. Then 03 is not representable, because the only representations of Id^-, R', S' and Id^" , R", S" are the ones shown in Figure 2, and clearly there cannot be six disjoint relations below U' x ¡j" with domain U'. It is also true that splitting U' x If" into two parts already yields a nonrepresentable RA (with ten atoms). When showing this in Example 4a below, we modify Example 4 so that we obtain a symmetric integral nonrepresentable 03 G RA with five atoms. Example 4a. Let everything be as in Example 4. Let R = R' + R" , S = S'+S", and D = (U'xU")+(U"xU1). Let 2l' be the proper relation algebra with atoms Idy, R, S, and D. It can be checkedthat D ; Ù ■0' = Ù ; D ■0' = R + S, and y <y;(R + S) for all y e{R,S,D}, so D is splittable in St'. Let 03' G RA be obtained from St' by splitting D into two symmetric parts, say into D1 and D" . Then 03' has five atoms, Id^ , R, S, D', D" . Assume 03' is representable. It can be shown that Si' has only one representation (up to isomorphisms of the ten-element set U ). Therefore we may assume that the representation of 03' is the same as that of Si' and D = tí + D" . Let u e U be arbitrary. Let D'(u) = {v e U" : (u, v) e D'} and D"(u) = {v e U" : (u, v) e tí'}. Then License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use H. ANDREKA, R. D. MADDUX, AND I. NÉMETI 1092 tí(u) n tí'(u) = 0 and \tí(u) U tí'(u)\ = 5 . Now, since tí < D = tí ; R = tí ; S and 1' • (R + S) = 0, we have \tí(u)\ > 3, and similarly |2?"(k)| > 3, since D" < D" ;R = D" ;S. This contradicts \tí(u)UD"(u)\ = 5 . Thus 03' is a nonrepresentable symmetric integral RA with five atoms. The essence of Example 4 was that we had an atom U' x U" such that the structure of the algebra forced U' to be small in any representation. We then split U' x U" into too many parts. Now we use the same idea to obtain a series of nonrepresentable RA's. Example 5. Let p = In + 1 be any prime number. We will construct a nonrepresentable 03n e RA. We start with a proper relation algebra 21^ on p = {0, 1, ... , p - 1} . In the following, + and - are understood modulo p . For i e p , define Rj = {(«, m) : |« - m\ = i (modp), «, m e p}. This gives us a proper relation algebra 21'^ with atoms {R0, Rx, ... , Rn} . In Si' we have R0 = ldp = V, and if i < j < n, then R¡ = RI, R¡ • Rj = 0, 1 = p x p = R0 u ■• ■U Rn , and R¡ ; Rj = R¡+j U R,_j . Let 2ln be the proper relation algebra obtained from Ql'n by putting two disjoint copies of 21'^ together as we did in Example 4; set U = if U U" where U' n U" = 0 and \U'\ = \U"\ =p. Let R\,...,R'n and R"x,..., R"n be the corresponding relations on U' and U", respectively. Then At 2lw = {Idy. ,R\,...,R'n, Idy» ,R"X,..., R"n, U' x U", U" x [/'} . Let 03„ be the algebra obtained from 2ln by splitting U' x U" into two atoms. By a slight generalization of the argument in Example 4a, we see that 03n is not representable. First, it is easy to show that 21'^ has just one representation on a p -element set, and hence that 2ln has just one representation on a lp -element set, namely the one used to describe 2ln . But U' x u" = D + tí in 2ln , so for any u e U' the images D(u) and D'(u) are disjoint subsets of U", each of which contains at least « + 1 elements (since D = D ;R¡ and tí = tí ; R¡ for / = 0, ... , « - 1 ), contradicting \U"\ = p . Exampleó. The algebras <£„({2, 3}) (see [Mai, 2.4]) are splittable. €„({2, 3}) has « atoms, namely 1' and «-1 atoms below 0', such that a = a, a; a— a, and a ; b = 0' whenever a and b are distinct atoms below 0'. It is not known whether every £„({2, 3}) is representable, but if a given <£„({2, 3}) is embeddable in iHec7, then there cannot be three elements u, v, w G U such that (u,v), (v, w), and (u,w) are all in the relation correlated with any diversity atom a, since a- a; a = 0. Therefore, by Ramsey's Theorem, for every « there is some integer r(n) such that if <£„({2, 3}) is embeddable in ÍHet/, then \U\ < r(n). Thus r(n) is also a bound on the cardinality of the underlying set used to represent any relation algebra obtained from £„({2, 3}) by splitting. If 21 is obtained from <£„({!, 3}) by splitting one of its atoms into more than r(n) parts, then any representation of 21 must also contain License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 1093 SPLITTING IN RELATION ALGEBRAS more than r(n) elements. This contradiction shows that such an 21 is not representable. Algebras constructed in this way can be used to prove Monk's theorem, that RA is not finitely axiomatizable [Mo]. For more details on how this can be done, see [Ma2, Theorems 12 and 13]. (In [Ma2], algebras obtained from £„({2, 3}) by splitting are further modified so that they are generated by a single atom.) Monk's original proof used Lyndon algebras. The algebras obtained from £„({2, 3}) by splitting are similar to the nonrepresentable cylindric algebras used by Monk to show the representable CAQ's are not finitely axiomatizable whenever 3 < a < to (see [HMT, 3.2.76 and 4.1]). References [HMT] L. Henkin, J. D. Monk, and A. Tarski, Cylindric algebras, Part II, North-Holland, Amsterdam, 1985. [J] B. Jónsson, Varieties of relation algebras, Algebra Universalis 15 (1982), 273-298. [L] R. C. Lyndon, Relation algebras and projective geometries, Michigan Math. J. 8 (1961), 21-28. [Mai] R. D. Maddux, Some varieties containing relation algebras. Trans. Amer. Math. Soc. 272 (1982), 501-526. [Ma2] _, Nonfinite axiomatizability results for cylindric and relation algebras, J. Symbolic Logic 54(1989), 951-974. [Mo] J. D. Monk, On representable relation algebras, Michigan Math. J. 11 (1964), 207-210. (H. Andréka and I. Németi) Mathematical (R. D. Maddux) Department Institute, of Mathematics, Budapest Pf. 127, H-1364 Hungary Iowa State University, License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use Ames, Iowa 50011