ON THE ALGEBRA OF ORTHOGONAL LATIN SQUARES by JACK BRUCE GOEBEL A THESIS submitted to OREGON STATE UNIVERSITY in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY June 1962 APPROVED Redacted for privacy Professor/hf Mathematic Charge of Major Redacted for privacy Chairman of Department of Mathematics Redacted for privacy Chairman of Shoo1 of Science Graduate Committee Redacted for privacy Dean of Graduate School Date thesis is presented Typed by JolAn Erttss Ju y 26, 1961. TABLE OF CONTENTS Page INTRODUCTION 1 CHAPTER I. MANN-RINGS 3 CHAPTER II. BOSE-RINGS 32 CHAPTER III. SOME EXAMPLES 40 CHAPTER IV. APPLICATIONS OF MANN-RINGS TO ORTHOGONAL LATIN SQUARES 53 APPENDIX 59 BIBLIOGRAPHY 61 ON THE ALGEBRA OF ORTHOGONAL LATIN SQUARES INTRODUCTION In the book Design and Analysis of Experiments by H. B. Mann, (7), a method is illustrated by which orthogonal Latin squares may be obtained from Galois fields. If we inspect some of his proofs, we find he does not use all of the properties of a field. If we delete from the axioms of a field many of the properties not used, the remaining system is what we call a Mann-ring. Its formal definition is given below. In this paper we demonstrate that a Mann-ring determines at least two mutually orthogonal Latin squares. This suggests a method for attacking problems arising from the design of experiments (7), and from various combinatorial problems dating back to the Kirkman schoolgirl problem (6). In recent years much progress has been made toward the construction of mutually orthogonal Latin squares (2,4,7,9). Some of these have used finite geomet- ries and some have used finite fields. In 1959 a variety of new methods appeared in papers by E. T. Parker, R. C. Bose, S. S. Shrikande, et al. Of paramount interest to Parker was the 10X10 Latin square. Our attempt to find a triple of 10X10 mutually orthogonal Latin squares leads to the first theorems of Chapter I. Since the construction of special types of Mannrings will lead to a set of mutually orthogonal Latin squares, we proceed through the remainder of Chapter I and all of Chapter II to study the algebraic structure of such systems. Some of the statements demonstrated there were suggested by (1). At the end of the paper will be found an Appendix listing some of the unsolved problems in this area, 3 CHAPTER I MANN-RINGS Definition: M Let binary operation. if all in y H.14 Indeed groupoid. A Definition: on M, for all x in S and for say be a left-S- (M, 5,') and 11 is associative then we is a left-S-semi-group. left-S-groupoid in which and is a groupoid. Let two compositions be defined M. in such a manner that ., (M, s,.) (M,S,+,$) S=M be a non-empty set of elements and M Let be a subset of a group and system is a left-S- roupoid may even be undefined. If the operator (M, s,.) S be a may be non-associative or non- y. x S C M Let Definition: let M "*" M. commutative. THEOREM: M We say that We note that say be two sets and S is an element of y x and is a left-S-groupoid. (MO.) Then the is called a Mann-ring if the system satisfies the following postulates: (1) There exists an element e.m.m for all m in M. e in S such that is 4 There exists an element dsa=dsb implies d-e is in for all S and (d-e)-m=dsm-m Definition: in for all such that S and a (d-e)sa=(d-e)eb b and a a=b d in for all in b implies M. a=b M. m in M. A Mann-ring in which (M,S*) is a Ieft-S- semi-group will be called an associative Mann-ring. Note: A Mann-ring is not a Veblen-Wedderburn system. In Chapter III we put some additional restrictions on our system and get a Bose-ring which is a Veblen-Wedderburn system with two postulates missing. Thus a Veblen- Wedderburn system is a special type of Mann-ring. THEOREM 0: A Mann-ring determines at least two orthogonal Latin squares. Proof: The addition table of the group square with the (i,j) element being m a new Latin square whose (i j) This is a Latin square because if then tinct. mj = mk so j = k (M,+) +mj . element is ci.mmi is a Latin We construct dsm +m = d-m js m k as the group elements are dis- We conclude that the row elements are distinct. We can check to see if each column had distinct elem nts by supposing dsmi + mj = dsmk + mj and therefore d.m. = drnk and so mi ink by postulate (ii). To see orthogonality we suppose that (rai+raj, thmil-mj) = (ms+mt, dims-lint). That is, suppose the (i,j) element of the superposed squares is equal to the (s,t) element. Then mi +mj =ms +mt and d.m.+mj=d-m s t By using the associativity of M and adding on the left side of both members of (a) we get: mj=-mi+(ms+mt). get d.mi .+m + Substitute this in the left of (b) and .m +mt by associativity. Whence we now add to the right side of each member and get d.mi- i=d.ms-ms again by associativity. Therefore (d-e).mi=(d-e).ms by (iii) and hence mimes by (ii ). i=s as the group elements are distinct. Hence mj=mt from (a) and therefore 3=t and thus we have orthogonality and the theorem is proved. It can be shown that as a consequence of (ii), (ii'), and (iii), S must contain at least three elements if M/f01. If M={0} and M is a group then 060=0 so S=f01=M and e=d=d-e=0 and all four postulates are clearly satisfied if we define 00=O. We will call [03 Note: 6 the trivial Mann-ring and in the rest of the paper we will mean M / [01 when we say Mann-ring. contains at least three distinct elements and if d/e+e then M contains at least four distinct elements. THEOREM: A Mann-ring (M,S,+,*) Call it 0. Then postulates (i), (ii) and (ii') require S to contain e, d and d-e. We note that e cannot be 0, for if Proof: M contains an additive identity. e=0 then d-e=d so postulate (iii) becomes for all dm=dm-m or This is a contradiction so e is not O. Furthermore e is not d for if d=e then postulate (iii) implies 0em=0 for all m, and (ii') tells us 0.a=0b implies a=b for all a and b. Hence 0=0 m=0 implies a=b contradiction. m. for all a and b. This implies M=(O, a is distinct from 0 and d. We claim that d is not O. If d=0 then by (iii), (0-e)bm=0m-m, or 0em=0 for all m. Now combine this with (ii) and we get that 0=0 implies a=b for all a and b, i.e., that M=[01, a contradiction. We now have that 0,e, and d are distinct. It may be that d=e+e as in GF(3). Otherwise 0,11,d, and d e are all distinct since d-e=e would mean d=e+e, the case we are ruling out, and d-e=d means d-d=e or e=0 and we know e is not 0 so d-e is not d. Furthermore d-e cannot Hence e or d and e would not be distinct, a fact we have already proved. Therefore if Oei-e, then we have be 0 four distinct elements in the Mann-ring. Now we attempt to construct a Mann-ring of order 10 using the ring of integers modulo 10 as the group M. We would like a Mann-ring (M S,+.) where S=M and "." obeys the following rules: M is closed under the operation ".". There exists a left identity e for ft.". The left cancellation law holds for two elements d and d-e. There is a right distributive law for d-e. We let a and b be any two elements from M and define the multiplication by: 1) a.b=xaP+ybr-za bt where x,y,z,p,r,s, arbitrary but fixed constants from and t are M. Fixed constants x.y.z,p,r s, and t cannot be chosen in M such that the multiplication defined by THEOREM: (I) will have properties 1), 2), 3), and 4). Proof: The proof is divided into several short steps labeled as Lemmas 1 through 11. Lemma 1: 9 9 =9. If g is an element of M then =g and 8 Clearly 25=2 Proof: and 6=24 Also 4=22 29=2. and and so the Lemma is true for 2.4,6, and 8. also clear for g=3 and since 9=32 7=33 and 8=2 3 It is it is true for 3, 7, and 9. We know that 5=5 modulo 10 for all t>0 so g9=g5=g modulo 10 for all g in M. In view of Lemma 1, the statement "let s=1", [or let p, r, or t, etc. be 11, will mean the same as "let s=1, or 5, or 9" wherever these symbols are used as exponents in this theorem. If a and b are in M then (a+b)n -an bn is an even number, for any integer n, 0<n<10. Lemma 2: If n=1,5, or 9 then (a+b)n-an-bn=0 by Lemma 1. If n=2, this expression reduces to 2ab. Otherwise Proof: (a+b)n-an.bn contains a factor of the form ab(a+b), ab(a2 +b2 )* a or b or etc. Such factors are always even for if either is even then so is ab and if both are odd then the second factor, [i.e., a+b, a2 +b2 etc.J, is always 1 even. Lemma 3: all b If there is an e in M such that eb=b for then either: (1) x=5 and e (ii) e=5 and x is even or is even or 9 or x=0 e=0. Proof: If e'b=b fore xeP=0. If x for all b then e0=0 and thereIf we have case (iii). x=0 or eP is a divisor of zero, [e.g., x=2 xp00 and then eP=5). This implies either eP=5 and therefore e=5 and x is even which is (ii) or else e is even and x=5 which is (i) or else If Lemma 4: Proof: xeP=0). e=0. If and x=5 e e is even then either t=1 and r=1 or t=1 and y=5 or r=1 and y=1. is even and e.b=b then ybr-z sb b (as Now let b=1 is even. Put and we see y=ze5+1 y=ze5+1 is odd since in ybr-ze sbt.b and got (br-bt)=b-br. Call this equation (1). This must hold for b=2.3,4,6,7,8,9. It is clearly true for b=0,1, and ze 5. Now recall 4=22. 8=23, 6=24, 9=32, 10 so put these in equation (1) for ing seven equations: b and 7=33 modulo and get the follow- 10 zes (2r.2t)=2.2r zes (22r_22t).4.22r ze5(23r-23t) 8-23r zes(24r_24t) 6.24r zes( 3r.3t)=3_3r ze (3 3 2t )=9-32r zes(3 ..33t)=7...33r. We observe that equations 2,,4 ,6 and 7 contain fac- tors in their left hand members of the form u2 v2 u3 -v3 , and u4 v4. Hence by proper multiplication of equations 1 and 5 by factors such as u+v, u2 +uv+v2 , etc., we may use 1 and 5 to eliminate by equating the right hand members. This done we get: u3+u2v+uv2+v3, 2') 3,) (2-2r)(2r+2t) (2-2r)(2+2r) (2_2r)(22r4.2r2 22t) = (2-2r)(2 +2.2r+22r) 2r22t 23t 4') (2-2r)(23r+22r (3-3r)(3r+3t) 6_24r, 32r = (3 3)(32 3 3 ) =7-3 If we factor these in the obvious way we get: 2') (2-2r)(2t 2) 0 or 11 3') 4,) (2-2r)(2 2)(2r+2t+2) = 0 6") (3-3r)(3t-3) = 0 (2_2r)(2t_2)(22r4.212t.0.2r4.22t4,2 e2t +22 ) 7') (3-3r)(3t-3)(3r+3 3) Clearly the only solutions are r=1, or t=1, Ei.e., r=1,5,9 or t=1,5,91. We check these solutions. If r=1 in 1) through 7) then either t=1 or ze5=0 or ze5=5 (since b'-b is even for all b). Now e was even so zes is not 5 and therefore ze =0 and since y=ze5+1 this implies y=1. The solutions are: r=1 and t=1 or r=1 if t=1 above. and y=1 or in 1) through 7) then r=1 is a solution as If t=1 then y=5 is also a solution, where y=ze5+1. Hence Lemma 5: If e=5 t=1 and Proof: We want eb=b and x y=5 Is a solution. is even then r=1. for all b so e.b=5.b=x5P+ybr-z5sbt=-. Since x is even we know b x5P=0. Hence we have ybr_b_z5sbt=0 and br_b y=ze5+1=z55+1=5z+1. Therefore z55(br-bt) and br -bt is always even so we have implies that r=1 (or 5 or 9). b 0 0 which 12 is even then the circle operation fails to have the required right distributive property. If Lemma 6: Proof: and e=5 By Lemma 5, x r 1. Therefore a.b=xalp (5z+1).b.zas t since y=ze +1. We want Since r=1 and e=5 (d-Oc=d.c-e.c=d.c-c for all c. this reduces to: xL(d-5)13-dP-5P1-yc ze [(d-5)s-d6-55]=0 Now y=5z4.1 for all c modulo 10. is even so let x=2k. This is true and therefore is true for c=2. So let c=2 and x and we get: 5)P-dP-5/21-(5z4-1)2-z2t[(d-5)5-d5-5s1 0 modulo 10. Hence k[(d-5)P-dP 5P]-(5z+1)-z2t-1[(d-5)9-d5-55] = 0 However since 5 E 0 modulo 5 we have: modulo 5. kte-d] ,1_z2t-l[d5.d5] 0 modulo 5, or -1 E 0 modulo 5 and this is clearly an impossibility so we have no right distributive law for case (ii) of Lemma 3. We now prove a similar result for case ( ) of Lemma 3. is even, the circle operation If x=5 and does not have the right distributive property. Lemma 7: Proof: By Lemma 4 either: t=1 and r=1 or t=1 and y=5 or 13 (c) Suppose (a). r=1 and y1. Then the right distributive property is: 5[(d-e)P-dP-eP3-zcE(d-e)s-d for all c. sl=yc modulo 10 is even and hence Now (d-e)P-dP-eP c(-z[(d- )s-da-081 yi 5 0 modulo 10. We know y is odd is even and so the expression in brackets is odd by Lemma 2. We then take c to be odd and get an because e odd number congruent to 0 malujo 10, an impossibility. Suppose (b) is true. Then the distributive law would imply: 5[(d-e)P-0-eP1-acE(d-05-d5- 5c modulo 10. Since the expression in brackets is even, let c-5 and get 0=5 modulo 10 which is impossible. Suppose (c). Then a.b=5aP+b-za5b implies -zct[(d- 5[(d-e)P-e-e] = 0 055 modulo 10. and (d-e).c=dic-e.c $] E c modulo 10, since by Lemma 2. Now let c=5 and get This is again an impossibility so the Lemma is proved. Lemma 8: Suppose e=0 as in (iv) of Lemma 3. Then a.b does not have the right distributive property. eb=b implies that 0b=b and hence that yb b. This is to hold for all b so we try b=1 and get y=1. Proof: 14 Hence br=b. So if e=0 ).c (d This means r=1 (or 5 or 9). then a.b=xab +b-zas bt . We want doc-e.c. Therefore x[(d )1)-e-eP]c-zc [(d-e)s-ds si =0 modulo 10. Since 0=0 this says x[dP-ej-c-zet[ds-ds] = 0 modulo 10. Hence -c=0 modulo 10 for all c. This is impossible so case (iv) of Lemma 3 has no right distributive law. If Lemma 9: x=0, as in case (iii) of Lemma 3, then either Proof: If x=0 then and r=1 and zes a.blaybr- a = or . bt. We want an identity such that e.b=b and hence ybr-ze bt=b for all b. If we let b=1 we see that Y=ze5+1. Our right distributive property requires that -zet((d-e)s-ds-esj=yer modulo 10 for all c. If we let c=5 we get 0 on the left in virtue of Lemma 2. Hence 5y=0 modulo 10 and this implies that y is even. Therefore ze5+1 is even. Furthermore putting y zes+1 in e'b=b reduces e.b to zes(br-bt)+br-b which is the very same equation we solved in Lemma 4. The volutions 15 were: or and t=1 and ze6=0 r=1 and ze =5 or t=1 and y=5. or even so (d) and (b) are not possible and therefore either: In the present case we have y or r=1 and t=1 r1 and zes=5. The second case of Lemma 9 is impossible be- Lemma 10: cause it lacks the right distributive property. Proof: Let e =5 and so z and therefore both z and see that z=1,3,507, or 9 e is odd. Also e5 is odd are odd numbers. Hence we and e=1,3070 or 9. We know is not 5 by Lemma 6. If z=1 then e5=5 and e=5 but this contradicts Lemma 6. e If z=3 then 3 and so 1,8=5 and so e5 but this contradicts Lemma 6. Similarly if z=7 or 9 then ze55 implies e=5, a contradiction of Lemma 6. So z=5 is the only possible case. Let e5=10347, or 9. Then and e=1 307. or 9. a.b=yb-5 sbt and y=ze5+1=6 a.b=6b-5 sbt. Now suppose Then z=5 Then so (d-e)ic=dc-e*c for all c. esi = 6c modulo 10 and letting c=2 16 gives 0=2 modulo 10, a contradiction. The first case of Lemma 9 does not have a left cancellation law for any two "consecutive" elements a and Lemma 11: a -e. Proof: or 9, In Lemma 9 we sdw y=ze5+1 and is even so z=1,3,7, e=1,3,7, or 9,[e is not 5 by Lemma 6 and is not 5 by Lemma 10]. So a.b=yb-zasbk z If we want a.b=a.c to imply b=c then (y-za5)(b.c)=0 implies b=c, so we need (y-za5,10)=1. This means that y-za5=1,3,7, or 9. Since y is even we need a to be odd. Hence we cannot cancel a on the left unless a is odd. But e was odd so if a is odd then a-e is even and cannot be cancelled. This completes the proof of the theorem. Since definition I of the circle operation fails we could try some new definition, such as a.b=p(a)b or a.b=ap(b) etc., where p(a) is some polynomial in a with coefficients in M. This suggssts the following theorem. There do not exist polynomials p and q with coefficients in M such that a.b=p(a)q(b) has properties THEOREM: (1), (2), (3), and (4). 17 Proof: We want the following to hold: (d-e).c=d.c-c; d.a=d.b implies a=b; implies a=b. If we take then or c=1 (d and )-a= d-e .b in (1) such that (d-e).1=del-1 p(d-e)q(1)=p(d)q(1)-1. So p(d-e)q(1)-p(d)q(1)=-1 p(d)q(1)-p(d-e)q(1)].. Thence q(1)[p(d)-p(d-e))=1. since p(d-e) is not equal to p(d) This also implies that p(d)-p(d-e) is odd, for an even number could not be congruent to 1 modulo 10. parity. Thus p(d) and Now (2) requires p(d-e) p(d)q(a)=p(d)q(b) a=b, and also a=b. These are equivalent to: p(d-e)q(a)=p(d-e)q(b) p(d)[q(a)-q(b)]=0 implies a=b. q(a)=q(b) implies a=b and is to imply that p(d-e)[q(a)-q(b)]=0 Thus each of these conditions implies that p(d)[q(a)-q(b)]=0 p(d-e)[q(a)-q(b))=0 and p(d-e) implies implies q(a)=q(b) q(a)=q(b) a=b. and we know that cannot be divisors of zero or else the above implications would be impossible. This is because we can always obtain a factor of 2 or of 5 in by choosing a=2 since and p(d-e) p(d) to imply since this is certainly a consequence of Since p(d) are of opposite and b=4 or a=5 and q(a)-q(b) b=0, etc. Now are not zero-divisors, they both must be mdd numbers, a contradiction since they were of 18 opposite parity. THEOREM: If M is a Mann-ring such that: M=S; is associative; both distributive laws hold; H.11 then is a ring with left identity. M Proof: We need to see that (e+e)(a+b) for any a and M b is abelian. Consider in M and e the left identity. (e+e) a-i-b)=t4 a+b)+e( a +b)a+b+a+b and (e+e)(a+b)=(e+e)a+(e+e)b=a+a+b+b so b+a=a+b. be an additive group and S be a set of left operators on M with the following properties: There exists an e in S such that ea=a for all Definition: a (ii') in Let (Mo+) M; There exists an element d in S such that da=db implies a=b for all a and b in M; d-e is in $ and (d-e)a=(d-e)b implies a b, all a and b in M; (d-e)a=da-a for all a in Ms We call the system 6=(M,S0-,) a left-Mann-module. If is a left module with the set S from a ring with left cancellation and a left identity then is a Mann-module. 19 Definition: Let(M,S + e) Mann-rings. (M.,S' and We call a mapping 0 of M into if: (m1+m2)0=m1oft20 for all ml' m2 in M. m0 is in S. is in S' if (sm)0=(80)(m0) for all s in S, and morphism of u into s) be two M' a homo- 3, m in M. We note that a homomorphism is defined in terms of m the addition and multiplication of the left-S-groupoid only and is independent of the four additional postulates defining a Mann-ring. Therefore we would expect that a homomorphic image of a Mann-ring might or might not be a Mann-ring as the homomorphism might not preserve the left cancellation properties of postulates (ii) and (ii'). If THEOREM: a Mann ring d is a homomorphism of a Mann-ring g into then the image of M is a group in M'. Proof: 0 is a row" homomorphism and this is a known theorem for groups. Definition: Let tr(M,S,+,.) be a Mann-ring. We say that h=(N,C,+,*) is a Mann-sub-ring if h is a Mannring and (N,+) is a subgroup of (M,+) and C is a sub- set of S. We note that if ring into a Mann-ring is a homomorphism of a Mann- then the kernel K d of 0 20 is a normal subgroup of (M,+) since o is a group homo- morphism. In standard ring theory it is natural to define a subset of a ring called an ideal. However Mann-rings not only lack both distributive laws but multiplication may be undefined for some elements. Since an ideal is a subring closed under multiplication by ring elements, the definition of ideal is not applicable in our case. Further difficulties arise from the fact that a Mann-ring must contain a left identity and thus the subring must contain a left identity. We shall see that in many cases of Mann-rings there is only one non-zero idem- potent and hence only one left identity. Another approach to the theory of ideals is by way of the kernels of homomorphisms. We can show that if there is a homomorphism of a Mann-ring, its kernel is indeed a normal subgroup of (M,+). Actually this is all that can be expected since the homomorphism is defined in terms of the left-S-groupoid and not in terms of the postulates for a Mann-ring. Thus a generalization of the concept of ideal appears not to be fruitful. However one is tempted to try to prove theorems con- Just as in the case of the ideal concept, difficulties arise because of the lack of total cancellation and distributive laws. cerning the homomorphic images of Mann-rings. 21 Since a homomorphism does not depend upon the Mann-ring postulates, there is no reason for them to be preserved. Indeed, one type of Mann-ring would be that for which S=M and we have a groupoid. In that case we could add inverses and get a quasi-group and we know there exist quasigroups possessing homomorphism upon systems which are not quasi-groups -- and similarly for loops (see reference (1), page 92, and page 1101 section 1). It is still an open question as to what loops G have the property that every image of under a multiplicative homomorphism is also a loop. The same problem for Mann-rings appears even more formidable. G Definitions Let a=(M,S,+,-) be a Mann-ring. If xx=x in 3, we call x an idempotent of S. If the Mann-ring is associative then x=xx=xex, an element of xSx since e is in S. So we define x as regular in S" if x is in S and x=xsx for some s in So i.e., x is in xSx. All idempotent elements are regular. If x=x(six), we call s1 the left relative inverse of x. If s =s (xs ) then x ansi are relative inverses of each other. THEOREM: If is an associative Mann-ring and if x is 22 with right inverse then potent if and only if xe=e. an element of C x is an idem- If xx=x then xe=x(xx-1)=(xx)x 1=xx-i=e. So xe=e. Conversely if xe=e then (xe)x=ex, so xx=x(ex)=(xe)x=ex=x and therefore x is idempotent. Proof: Corollary: In an associative Mann-ring, an idempotent with right inverse has the property that m, and hence is a left identity. xm=m x for all Definition: Let i:=(M,S,+.°) be a Mann-ring. By the MacNeish kernel S* of Es we mean the maximal subset of M such that every element of (ii), (ii'), and (iii) and also except e S* S* satisfies has left cancella- tion, right distributive law, and is closed under subtraction. We form S* by starting with all suitable elements in S and then extending our definition of multiplication from SXM into M to a multiplication of S*XM into M. So S* depends upon S, upon the original multiplication and upon (M0-). be a Mann-ring and let i*..(m,s*,+,.) be the "MacNeishft-ring formed by replacing Now let S go.(m,s0.,..) by the MacNeish kernel S*. 23 Every element in THEOREM: has a unique right in- S* verse. Proof: We have left cancellation for all in s S* so M= sM. Corollary 1: In an associative MacNeish-ring, the only idempotent is the left identity Proof: So xx=x If by the two previous theorems. xe=e then e. is the unique right inverse of e of all idempotents and xm=m so x and therefore x is a left identity. (x-e)m=xm.m=0 But then the distributive law tells us that for all So we choose a different from m. that (x-e)a=0 a=b, unless and x-e=0. So x-e=0 The left identity Corollary 2: and e and get (x-e) =(x-e)b hence (x-e)b=0 b hence x=e. is unique in an asso- ciative MacNeish -ring. Proof: fm=m If for all th n m so fe by ff=f corollary 1. Let 6 M S*,+..) be an idempotent such that for m in So define M we see xM=(xm:m be a MatNeish-ring and let x and e x are in m=xm-1-(m-xm)=xm+(e-x)m, is in M1 (e-x)M*f(e-x)m:m and define is in M. S*. x Then 24 Then we note that b is in xMn(e-x)Me0 So (e-x)M. is in b if xbreb the left Pierce decomposition of decomposes a MacNeish -ring if if xbe0 and MexM 0 (e-x)M and So an idempotent M. and x xM are in e-x x S*. We now proceed to define and discuss hyper-Mannrings and hypo-Mann-rings. ring and let to be in y.m we have be in M for M y.m choices for n m all could let y.me0 is not and let 0 rm such that for all in Then define if the group M ym=m for all or we could let m, 0m=0, =sufyi S. for each given M. m is of order m or we y.mem if or any other multiplication was an element of Now define be a Mann- 3e(M,Sice,y) but not in For example we could let n. y y Let M. We then see that (M,51,+,') * Is a Mann-ring where the dot operation is the old dot multiplication on into SXM into M and is the new dot on laM M. So any Mann-ring is a subring (i.e., "is isomorphic to a subring") of "larger" Mann-rings formed by increasing the size of S by any "suitable" definition of left mul- tiplication for the elements of original n. S. We are not M that were not in the These "larger" Mann-rings are still of order adding elements to ing elements into S M, rather we are shift- that were not originally in S. Since the postulates of a Mann r ng still hold for 25 any such extension of S, we see a "maximal" Mann-ring of the same order as would be (M,M,+,*). Call any such ring a hyper-Mann-ring containing 3 and label it by b7(M,M,+,.). There will be many such rings depending law* upon the different multiplications or defines. Similarly we see that S really need contain only the elements e,d, and d-e to qualify as a Mann-ring and it may even be that d-e=e. So define 3=(M,S,4-,*), where S=[,,d,d -el, as the hypo-Mann-ring contained in The above discussion demonstrates the following theorem. is embeddable in a hyperMann-ring T. We note that a Mann-ring will be embeddable In as many hyper-Mann-rings as there are binary operations definable from (M-S)XM into M. By embeddable of course we mean the multiplication agrees with the original mulTHEOREM: Every Mann-ring tiplication on 5 SXM. Now choose some fixed multiplication from MXM into M and thus a fixed 3 containing 3 (isomorph cally). we have the folThen considering the hypo-Mann-ring lowing theorem. THEOREM: Every hypo-Mann-ring of order 3 and in a fixed M is of order 3-0.(M,S0-,e) n where has 2n-3-1 $ is extensions 26 Proof: SC SiC S2C.C S =M is any element not in a where S. where S3+1=S5U [al The number of distinct S 's then depends upon the ways of choosing the distinct This can be done in a. ( n-3) (n-3 (n-3 ) n-3 2 1 different ways. Q.E.D. We see that the Mann-rings formed by using these subsets of M as the S of the left-S-groupoid can all be regarded as subrings of each other as of M. Corollary: n-2 3C EiZ Hence, If S is a subset = C is of order M 2 then there will be extensions for each fixed multiplication. is isomorWe have noted that every Mann-ring phic to a subring of a hyper-Mann-ring 5 and contains a hypo-Mann-ring Z. Suppose S is of order t, where t=2 or 3. Then there is a chain of sub-Mann-rings Ek every order k, t<k<n, such that for each fixed multiplication, 2 1 00-eitc at+1C For a given C ft fx(M,S,+ .) is or order + IC where C 141 is of order n and we can determine the number of distinct binary operations ft 0 " from MXM into M such that " 0 " agrees with the original multiplication on SXM. S a 27 For in M-S in yl we want For a chosen M. of the elements of n There are 0 in yi n-elements by possible mappings of M-S yip m (M-S)Xm nn2-an S and if there are into Now m M. n n(n -a) If is of order S (2n no a So associ- with .1)(nn(n-al+ .) S Mann is of order 3 then the family of Mann-rings S (2n -3 -1)(nn(n-a)) Definition: such that Mann-rings between distinct multiplications. we have a total of a, If Mann-rings. Two Mann-rings are equivalent, (we write if they have the same group groupoid was any multiplication. "." 2n-2-1 was of order 2. S 0 ated with each "parent" Mann-ring of order contains n-a so there are is the original Recall that there were rings. M -S. and each can be possible multiplications restricted to of order in y2 elements so we get a total of n (nn-a)n Similarly for M. m y10 m beEll one we can let elements n -a mapped onto one of m defined for each yl 0 m (M,S,) if and only if (M.+) and the same left-S- in the sense that s.m=s0m for all ), $ (M,S,+,°)=(M,S,+,G) in S and all m in M. We will now investigate specific methods of getting 28 new Mann-rings from old ones. Suppose given a Mann-ring = (M,S,+,.). If M is of order n we can use the original multiplication tt.11, to define n multiplications say "(a)ft . For each element from SXM in M we define the operation (a) a into M by: d(a)m=d.m+a and (d-e)(a)m=d(a)m-m, if is different from and s(a)m=s.m d-e. So the new ring 3(a) is the old system E with only the multiplication for d and d-e altered. We verify that Zi(a) is a Mann ring: s d and d(a)a=d(a)b means da+a=d.b+a so d.a=d.b and so (d-e)(a)a=(d-e)(a)b is the same as d.a+a a=d.b+a-b. So d.a-a-d.b-b or (d-e)a=(d-e)b so a Therefore a=b. postulates (ii) and (iA') are verified and (iii) is valid by the definition of (a). To see that these rings are all distinct, we suppose that d(a)m = d(rk)m for a,PeM and some m in We note also that if a=0 we get (a)=". and therefore we get 3 back again. We can get n more distinct Mann-rings if e+e . In that case we define ral to be a mapping f SXM into Then dm+a=d.m+p so a=P . 29 M b dCalm = (d-e)-m + a and (d-e)[alm . Calm -m and all other elements map as before under the original multiplication. It should be noted that the multiplication for d-e is completely determined by the addition table and the multiplication for d because of postu- late (iii). Thus it is not really necessary that the multiplication of d-e be specified. $o if our original group (M,+) had a multiplicative left identity e of additive order 2, (i.e., e+e=0), then we do have distinct Mann-rings generated by this device of redefining multiplication for d and d-e. This demon2n strates that for every pn GF(pn) there exist at least Mann-rings and for every GF(2n) least there exist at 2n+1. We end this chapter on Mann-rings with some models of order four to illustrate the independence of our pos- tulates. In all four models we choose the fours group for the additive group. So let (M,+) be the fours group (i,e,d,d-eI, with addition defined by x+x=0 for all x. Let S=1.(e,d d-el. We will use d+e in place of d-e since +e=-e. 30 Example (1): We define multiplication by the fol- lowing table: Here we see that (ii), (ii'), and (iii) are valid but (i) fails so (i) is independent of the other postulates. Example(ii): We let (MO.) be the fours group again (where x+x=0) and Si.4e,d,d-el. We define mul- tiplication by the following table: Here we see (i), (ii'), and (iii) are satisfied but (ii) is not. Therefore (ii) is an independent postulate. Example (lit): For the same define multiplication by the table: M and S as before, 31 Here (i), (ii), and (iii) hold and (ii') fails so (ii') is an independent postulate. Example (iii): Use the same define multiplication by: ON+ and S and Clearly (/). (ii). and (ii') hold but (d+e)d=d while dd+d=0 and d is not 0 so (iii) fails and so postulate (iii) is independent of the rest of the system. 32 CHAPTER II BOSE-RINGS In the previous chapter we discussed how to generate new Mann-rings. However each new Mann-ring used the d and d-e of the original ring. We saw in Theorem 0 that such a Mann-ring will generate at least two mutually orthogonal Latin squares. In order that we may assert that our system will indeed generatz more than two such squares we want more elements to satisfy the postulates. We begin by requir- ing that S=M-[01, and all elements to satisfy postulates (ii) and (iiT) as per the following definition. Definition: A Bose-ring is a Mann-ring ij=(M,S.+,.) with S=M-10/ and such that: ma=mb implies a=b for all a.b in M and all m not (a+b)m=amtbm O. for all a,b,m in M. So a Bose-ring has a left cancellation and a right distributive law. By an associative Bose-ring we man a Bose- ring in which the multiplication is associative. THEOREM: If a Bose-ring Z=(M,M-0,4-,.) is of order then it determines a set of n-1 pairwise orthogonal n 33 Latin squares. Each of the n-1 elements of S is a suitable "d" for the proof of Theorem 0 in Chapter I. So we get Proof: squares in that manner and these include the original square. Their mutual orthogonality follows from the proof of Theorem 0 in Chapter I. n-1 Example: The Veblen-Wedd rburn systems are examples of Bose-rings. THEOREM: In an associative Bose-ring, Proof: If a Ox=(x-x)x=xx-xx=0. is not 0 neither equal to Then x0=a. Now suppose then choose O. Ox=x0=0. b and c not equal and x0=x(0b)=(x0)b=ab and similar- ly, x0=x(0c)=(x0)c=ac. So ab=ac and if a is not 0 then b=c, a contradiction to the assumption on b and c. Thus it must be that a=0, i.e., x0=0. Q.E.D. Note: The above theorem proves that an associative Bose- ring is a y-ring as defined and discussed in the following page. The question of when x0=0 is partly answered in the following proposition. Definition: Let a be a Mann-ring with S=M. We call op-ring if x0=0 for all x in B. We note that if x0 is not 0 for some x then B 34 we can let x0=p where following phenomenon. if pz=(x0)z=x(0z)=x0=p. that pz=p is not p and we observe the 0 is associative: B p So there exists a non-zero for all in z B. such This characteristic of a non-T-ring leads to the following theorem. An associative Mann-ring B is a T -ring if and only if zx=z for all x in B implies z=0. THEOREM: Proof: B. Then fore in Suppose z0=z. z=0. B is a trring and let B B But is a T-ring so Conversely suppose that implies (x0)b=x(0b)=x0 Consider z=0. so (x0)b=x0 therefore by hypothesis x0 z0=0 for some b in x in and there- for all zx=z, for all all zx=z, x in B and x B. x0=0. Q.E.D. This suggests areas where not to look for T-rings. For example, consider the set of transformations of a ring into itself. Let Tb be that transformation taking every element of the ring into the element for all Tx b. Then in this set of transformations. would not form a T-ring. T T =T b x b Thus they Example 3 of Chapter III shows that they can be formed into a Mann-ring. THEOREM: In an associative Bose-ring, the right 35 cancellation law holds. Proof: By a previous Theorem, left cancellation, if xa=x0 a=0. x0=0, for all x. By is not 0 and xa=0 we have so a=0. Thus if x is not 0, xa=0 implies Now suppose c is not 0 and that ac=bc. We x must see that a=b. We note ac-bc=0 and so (a-b)c=0. a-b is not But we chose c not Now if 0 0 then c=0 so a-b=0 by left cancellation. and therefore a=b and we have right cancellation for all non-zero elements of M. Q.E.D. Corollary 1: (proved later as an independent theorem) In a Bose-ring in which the associative law holds, the only idempotent is e or 0. Proof: If xx=x is not so that x-e is not 0. then (x-e)x=0. If x is non-zero then x-e=0 Suppose x e since we have right cancellation from the Theorem. But x-e is not 0 and therefore e is the only non-zero idempotent. Corollary 2: In an associative Bose-ring, unique left identity. Proof: Use corollary 1. e is the 36 Later in this paper we will remove the restriction of associativity from both corollaries 1 and 2. Now a previous theorem tells us that right cancellation holds. Thus xa=ya implies x=y if a is nonzero. Hence Ma=M and there exists a11 in M such Note: that a e. So right cancellation implies the existence a1 of left inverses. Also the left cancellation implies -1 rightinversesasuch that orem will show that if x such that xi =xr Imx aa -1 is not xx x e. Our next thethen 0 x=e. In an associative Bose-ring, the left inverse equals the right inverse for any non-zero element. THEOREM: If Proof: that is not x lx=e 0 let the left inverse be and the right inverse be r 1 such that such xr=e. We note that: rx=(er) =((lx)r)x 1(xr)x=1(e)x=lx. r=1 Thus rx=lx and so by the associative law and the right cancellation. Q.E.D. THEOREM: e In an associative Bose-ring the left identity is also a right identity. Proof: For x in M and x not 0 let xe=a. Now 37 choose an element and write (xe)b=ab or xb=ab. Then by right cancellation we see x=a and thus xe=x. We already knew that 0e=0 so xe=x for all x in M. not b 0 Q.E.D. In a Bose-ring in which THEOREM: for all x0=0 the x, right cancellation law holds. If Proof: But c (x-y)0=0 is not zero let so xc=ye so that (x-y)c=0. (x-y)0=(x-y)c and therefore c=0 by left cancellation, if (x-y) is not zero, and contradiction so x-y=0 right cancellation law. and x=y. c=0 is a This demonstrates the Q.E.D. Corollary: exists a c if is in M and therefore there such that c-1c=e and so c has a left Mc=M c inverse. Definition: By a subring of a Bose-ring mean a Bose-ring *n=(N,E,+,) such that group of Note: NI,S,+,) (N 4.) we is a sub- (M,1). is the only idempotent in a Bose-ring so the same left identity e must be used for all subrings of a Bosering. e 38 In a Bose-ring, the left identity is unique. THEOREM: Proof: Choose two elements different from b. If ties then Also and and e2 (e1 e2 )a=eI a -e2 a=0 (el -e2)b=elb-e2b=0 That is, e a is are two left identi- b such that since they are identities. and so the left sides are equal. (el -e2)a=(el -e2)b. But we have left cancella- tion for every non-zero element of the Bose-ring so el e2 non-zero would imply a=b, a contradiction. 01=02 a Hence and the left identity is unique. Q.E.D. THEOREM: In an associative Bose -ring, there is only one non-zero idempotent. Proof: Choose any element in M. Suppose x is a non-zero idempotent. So xx=x. Let xa=b. Then x(xa) xb or xa=xb and since x is not 0, we have left cancellation and so a=b. That is, xa=a. But a was arbitrary so x is a left identity. Yet e was the unique left identity so x=e. Q.E.D. We can remove the restriction of associativity if we change to a different type of proof. We do that in the following theorem. 39 In a Bose-ring with right cancellation, the only THEOREM: non-zero idempotent is e. If Proof: is not x and 0 xx=x then xx=ex x=e. so Q.E.D. If THEOREM: is a homomorphism of a Bose-ring o into a Bose-ring Eil=(Ms1M"-(01,4-..) 3=(M,M-(010-1-) the kernel of Proof: If exists an x is zero. o is in M and x is not such that xx- =e. x-1 then = (xo)(x -1 o) = eo. Also So 0 (xx then there )cy (em)o = (eo)(mo) = inc so 00 Is the left identity in Now if x is not 0 and x0=0' we have )0 (x0)(x-10)=0"(x 0)=0'0 since 0/m1=0, in a Bose-ring P. Therefore mo=0' for 11 m. This is impossible so xo is not 0' unless x=0. Then eo=(xx Oa = (m-m)0 = mo-mo = 0'. So the kernel of o is (01 Q.E.D. If we considered ideals as kernels of homomorphisms then the above shows that a Bose-ring has no proper ideals Note: and no proper homomorphic images. 40 CHAPTER III SOME EXAMPLES If Example 1. and S=M M then is a field with the usual operations is a Bose-ring. a field, such that d-e is not Example 2. (M,S,+,) M is a Mann-ring. example 3. GF(3) Rg=-g all e [e,d,d-el for any element and d-e is not 0 then Let M be the set of all transformations of into GF(3). M contains 27 elements. Let be defined by (Ti+Ti)g=Tig+Tjg. We define R for all -T=RT for on GF(3). T, (M,+) Now let g in GF(3). Then letting is a group. S be the symmetric group 33 is contained in M and (M,S,+,*) is a Mann-ring with multiplication defined by the usual operator notation. It can be verified that we can take d to be S /012) '210 and by d_e . 1012 ) or we can take At7)121 `201r ' ='1021 d-e = (Tg). In this example each element T1 is a ' r1(1 41 transformation of GF(3) 10,1,21 and there are 27 ways of mapping (0,1,21 into a triple (a,b,c) and these are the ternary representation of the integers For example (1,0,1) 10 and (2,1,2) is equivalent to 23 to the base 3. Therefore we will arrange and label the 27 transformations such that Ti base 3. Then 1 T15 T19'T21 012% and 012" T21 the only choices for andare 11 and M For example T1+Ti=T(141)mod 27. S = [T5TTl T 'T19 are the respective d-e's. (T-u' S = (T5 T26 I T j = 012% 1abc )t S and T 1 Hence abc.j in base 3 and =83 although we could be any subset containing T51T 1,T15 or else T5,T21,T19. 1-1 d, 1 = 53. TI ,T15,T19 It seems natural to take let abc=i since 17 base 3= 122. Thus (T.1) T17 where (°a) At least we need both and therefore in 53 d d-e and in order that (II, to be S#4. ') be a Mann-ring. The similar example of all transformations on the field GF(22) does not work as there are no 1-1 42 transformations is not in d-e. For all and d in d -e 4' so that example fails. S4 To continue with example 3 we attempt to form a homomorphism of into itself. 3 the additive group Since T +TfT(.3.4.nmod 27 is the abelian group of integers modulo 27 generated by Tl and so it is cyclic. (M,+) Hence the only subgroups are the cyclic subgroups c)( generated by T9 and the subgroup 3 =IT0' T9' T18 I 3(9=1T0,T3,T6,T9,T12,T150T18 'T21* T241 generated by T3. These are normal of course and are the only candidates for the kernels of proper homomorphisms of M. We find that we can not get a homomorphism of 3. If we take t.v3 as the kernel then T a=Tj a if ii mod 9. Therefore MI -4 (0*3' .A.4( Scr = (T/ T/ T1 T/ T'T/1 for d, namely Tli ... # T8 T1 and 3 We see also our two choices and T21 map onto T and Ti respectively. We must define multiplication for the left -Sgroupoid (MI,S*,+.8). Let S'=S. In order that we insure the homomorphic properties of a we define multiplication by (TiTi)a = (Tiv)(Tia) for T in S. Since each element in Ss has a pre-image in 5, this 43 determines the multiplication if we agree that for S a=T' Tsi and For example T in ij then T'T' = (Ts Tj )a. TIT.; (T19Tj)a and T' T'i 2 (T1 TJ )a' etc. The multiplication table for S' would then be: / T" IT / t t 1 / T'4 I I T I T3 ' T1 T T,6 T° 1 / 8 $ that this multiplication gives da=db implies for any choice of d. We also see that T; is our We observe a=b new identity. However no choice of d can satisfy pos- tulate (iii). Thus while (i), (ii), and (ii') are satisfied, postulate (iii) fails. as a The only other possibility was to use kernel and to have M'=Ma = (T'TIT'l. This is an addi- tive group with S=S' so Se=M°. However M' has no 44 left identity e and under our usual multiplication (sm)a is not (s )(ma) so a is not a homomorphism. Any left vector space (left module) is a Example 4. Mann-module. A Mann-ring need not be finite. Consider a Example 5. of the set of analytic functions of a complex variable. Let M be the set of simple (schlicht) functions on a region D. Then f(z) is analytic, onevalued, and does not take on one value more than once in D. So D of the z-plane is represented as D' of w=f(z)-plane in such a way that there exists a 1-1 corsubset M respondence between them. We know a simple function of a simple function is simple (Titchmarsh, Theory of Functions, 2nd edition, p. 198). Thus fog=f(g) is defined and M is closed under 4.o. We may take S to be any set consisting of at least e and one other element d such that d-e is not 0 or e. We know d -1 is simple. Then if d is in S there exists d- in M such that d -1 and Example 6. d-1 d = e. We therefore take S to include (d Any Veblen-Wedderbu n system is a Mann-ring and indeed is a Bose-ring. 45 Example 7. ments d, The rings of integers having consecutive ele- d-1 relatively prime to their characteristics are Mann-rings. For example 115 with d=8, and d-1=7, etc. The Mann-rings of order 3. If (M,S,+,-) has three elements then M=(0,e,d) only possible + and tables are: Example 8. and the where we get the multiplication table by using postulate (iii) and the addition table. We see it is not necessary to specify left multiplication for zero, i.e., we can table is that of GF(3) and the multiplication table is a subset of that of GF(3). So if we define 0.m=0 then this is of necessity GF(3). If take S=iedl. The we define to be not + at all then this is not 0 or if we do not define 0.m GF(3). The Mann-rings of order 4. We have noted that given one Mann-ring we can devise ExaTple 9. 46 more by defining our multiplication for dem in terms of one of the elements a. For example d(a)m = d.m + a. We saw this would give rise to as many as 2n rings if el* = 0. We know GF(22) is a Mann-ring of order 4 and characteristic 2 so these devices give us 8 Mann-rings of order 4. We prove these are the only ones and then display these 8. First we make some observations, The group (M.+) group. . is either the cyclic group of order 4 or the fours The two addition tables are displayed. The cyclic group. We note that ab=c always. Now the additive group of GF(22) 2. The fours group. is the fours group and hence the 8 Mann-rings already mentioned have table 2 as their group. We will show that no Mann-ring of order 4 has the cyclic group 1 as its group. Suppose first we consider the case where e+e is not equal to d. S contains at least 3 distinct elements 47 e, d, d-e. If e+e is not d and e+e is not then we have either: Case (a): e+e=0 or Case (b): e+e=d-e. In Case (a) where e+ =00 e corresponds to b in table 1 and again we have two choices, namely letting d correspond to a or to c in table 1. Call these Cases a 1 and a2. We first let d correspond to a and and check out caseal . A somewhat tedious but routine calculation shows that it is impossible to define a multiplication from SXM into M satisfying (i)-(iii). The calculation is simplified by noting that postulates (iii), (i) and the multiplication for d will determine the table of multiplication. An example is as follows: If we let d*O=d-e, ded=d0 and d.e=e then (iii) implies that (d-e)s0=d-e, (d-e)sd=0, and (d-e)e=0, and the last two equalities are a contradic- tion of (ii'). All other choices lead to similar contradictions so case al fails. Next we try case a2 which is symmet- rical and the same simple calculations show that it fails. Next we try Case b where e+e=d-e so d=e+e+e. Since we are letting e+e=d-e, we cannot have e+e=0 and therefore e cannot equal b in table 1. Thus there are two cases: Case bl where we let e correspond to a, 48 d to c, where e is Now in case d-e to and b is a, in table 1; and Case d-e is b2 in table 1. c, d b we get the addition table displayed: 1 and b Again we see that postulate (iii) and this table determines the multiplication of (d-e)..m if d.m is known. There are only 24 choices for dem and all lead to a contradiction of postulate (ii'). Similarly Case b2 fails to produce a multiplication table that yields a Mann-ring. The only choice left would be to let e+extd. e=d -e. Then M contains elements (0,e,d,x1 and e+e=d so d.m=m+m, all m. Now d is non-zero so is not b in table 1. Therefore we get: Case (a) e is a in table 1, or So Case (b) e is c in table 1. In both cases d must correspond to b in table 1 so 49 We see then that dm=m+m, all m. Hence dee=d and dx=d, a contradiction of (ii) so no Mann-ring can exist whose additive group is the cyclic group of order 4. Thus any Mann ring of order 4 must have (Ms+) equivalent to the fours group (table 2). We note that e+e=d and x+x=d. so that e+e is not d and so S contains 3 distinct elements and M=[0,e,d,d- 1 which we can write as e+e=0 10 e d d+e1 since -e=+e. Now our postulates cannot distinguish between d and d e where e+e=0 so we get the same + table if we let e be a, or e be b, or e be c. In fact these correspondences are all just permutations of the rows and columns of our additive group since x+x=0 and 0+m=m so we can write down the only possible addition table: And as before we use (d+e m=dm+m, i.e. postulate (i ), 50 and this addition table to determine the multiplication table. There are then only 41 cases for d.m and 8 of these we know give Mann-rings and a simple check show the rest fail. We write down the 8 possible cases now. We get them as follows: Start with GF(22). We know that its multiplication gives a Mann-ring. We define four new mul- tiplications as follows: d(e)m=dm+e and d(0)m=dm d(d)m=dm+d d(d+e)m=dm+d+e. This gives 4 Mann-rings whose multiplication tables are: (e) e ( e d IMIN 0 111111 d 1111 d+e d+e 1111 d d+e d+e d 0 0 d+e The opera ion associated th the element y is indicated by (Y). Now since e+e 0, we can get 4 more Mann-rings. We can use the fact that e+e=0, e e, and hence to verify that the following definitions actually yield Mann-rings. Let "." be the multiplication of d-e=d+e GF(4). Define th operation [y] as follows: associated with the element y d[O]m = (d+e).m + 0; d[ejm = (d+e).m. + a; dfdlm = (d+e).m + d: d(d+e]m = (d+e).m + d+e. These four multiplications give 4 Mann-rings with the mul- tiplication tables as follows: 53 CHAPTER Iv APPLICATIONS OF MANN RINGS TO ORTHOGONAL LATIN SQUARES Orthogonal Latin squares were first considered in a paper by Euler in 1782 (5). Early papers on the subject include those of MacNeish and Bose. ni n If pl'p2'p3 *0. P W is the prime power decomposition of an integer function n(v) by v der v. n(v) n n n(v) =min(p110,24', it is known [MacNeish-Mann] at least and we define the arithmetic -1 that there exists a set of mutually orthogonal Latin squares of or- be the maximum number of mutually orthogonal Latin squares of order v. Then the MannLet N(v) MacNeish theorem is N(v)n(v). MacNeish conjectured that N(v)=n(v). Since n(v)=1 in this case, this implied the correctness of Euler% conjecture about the non-existence of two orthogonal Latin squares of order v=4t+2. MacNeish's conjecture was disproved by Parker. Bose and Shrikhande (3) obtained counter examples to Euler's conjecture. We saw in Chapter I, (Theorem 0), that Mann-rings lead to mutually orthogonal Latin squares. In this 54 chapter we investigate further the relation between Mannrings and mutually orthogonal Latin squares. We know the addition table of (M,+) in a Mann-ring [=(M,$,+,.) forms a -Latin square whose i,j th element is mi+mj. also know that another Latin square can be formed whose .th element is dmi+mj and that these two squares are J..) orthogonal if d satisfies postulates (ii), (ii'), and (iii). Now if there are two such des in S, say d and d2 and then they determine two Latin squares, say M M2. To make the discussion easier we introduce some notation. Let the Mann-ring postulates (i), (ii), (ii'), and (iii) be referred to as postulate set I. Let Postulate set II be the followings di-dj is in S. (di -dj)a=(di-dj)b implies a=b. (di-dj) =d THEOREM 4.0. If Ml and m-dj m. M2 are the Latin squares de- fined by d1 and d2 respectively and if d1 and d2 sa - tisfy postulate sets I and II then M2 M1 is orthogonal to 55 Proof: Suppose the (I j) element equals the (s,t) ele- ment in the superposition of (d 1m .1' d2 i+mi) = (di s+mt, d 1m1 - dims I .e., with Mi s +mt ). Then and m d2 m1 + mi = d2m5 So dimi -d2, i or (di -d2) dims -d2m = (di -d2)ms by 3) of postulate set II and therefore by 2) mi=ms and so mi Q.E.D. If di THEOREM 4.1. and they determine Latin squares Proof: In the Mi ( ms = m and M ,t) element is Now suppose that dims + m so satisfy postulate set di M d1m5 mt. dimp + mt. Then dims= imp and therefore s p and each column element is unique. Similarly, row elements are unique. If THEOREM 4.2. and II then Proof: 4 d and d satisfy postulate sets I Is orthogonal to M4. The proof is the same as for Theorem 4.0. 56 Notation: ing e Let S* be the minimum subset of such that every element of S* except contain- S e satis- fies postulate sets I and II. Then we have the following corollary. Corollary 4.2: If is of order a then the Mann-ring 3 . (m,s +,.) determines a set of s pairwise orthogo- S* nal Latin squares. Proof: Theorems 4.0, 4.1, and 4.2. Now it may be possible to adjoin elements of M-S to the set S in such a manner that the order of S* is increased and is still a Mann-ring. For example, if d is in S but d-e is not in S we can join d-e to S with the stipulation that the left multiplication for d-e be determined by postulate (iii). If d-e then failed to satisfy (ii') we would reject it as a new element of S. If d.m is to satisfy postulate (ii), there exist only n2 choices for d.m for fixed d and m so it can be determined if a suitable multiplication is definable from dXM into M such that postulate sets I and II are satisfied. In view of the foregoing discussion, it seems natural to make the following (M,S,+,.) 57 definition. Definition: The MacNeish kernel maximal subset of M containing is the S* of S* such that every either satisfies postulate sets I and II or can be made to satisfy sets I and II by a suitable element of S* extension of the multiplication of SXM onto If M. is of order k then the Mann-ring (M,S*,+,0 determines a set of k pairwise orthogonal THEOREM 4.3. S* Latin squares. Proof: Theorems 4.0, 4.1, 4.2, Corollary 4.2, and the above definition. Q.E.D. THEOREM 4.4. Let S* be of order a. Then there is no set R in M such that there exists a suitable extension of multiplication from S to R such that = (M,R,+,.) determines a set of a+1 pairwise orthogonal Latin squares. Proof: is defined to be maximal with respect to postulate sets I and II and thus with respect to the number of pairwise orthogonal Latin squares determined by 3. S* No proper subset of S* could determine a pairwise orthogonal Latin squares. THEOREM 4.5. set of a 58 Proof: Every element of satisfies postulate sets I and not containing deS* is S* and II. If a set C S* such that (M,k,+,.) determines a pairwise orthogonal Latin squares then the set i U fdl is in S* and will determine a+l pairwise orthogonal Latin squares, a contradiction of Theorem 4.4 above. Q.E.D. If THEOREM 4.6. S* = M-(61 is then te = a Bose-ring. Proof: Routine verification. We note that S* is contained in S* so the MacNeish kernel of 3 will depend upon the original set S in 3 . (moso.,.) and upon the multiplication from S*XM into M. So S* really is dependent upon all of We note that a consequence of theorems proved in the sections on Mann and Bose-rings is that if the multiplication is associative then e is unique in S* and is the only idempotent element in S. 59 APPENDIX Some obvious questions may have arisen in the mind of the reader, some of which should not be asked and some of which are not new to this paper. It has already been pointed out that the concept of "ideal" really does not apply here as the "ring" structure is too weak to support it. usual arithmetic of Mann-rings allows bizarre behavior. Furthermore the unx.0=0 and other For this reason the name of "ring" should not lure one into too serious consideration of ideals, quotient-Mann-rings, or kernels of homomorphisms. It would seem more fruitful to investigate unions inter- sections, central elements, isomorphisms. etc. In this paper we mentioned a direct sum decomposi- tion which may suggest further exploitation. Some questions are In a Bose-ring, is e the only non-zero idempotent? In a finite Bose-ring, is not 0 x.0=0? We know in some infinite Bose-rings and x.0 zero in some Mann-rings so we expect perhaps not 0 in some finite Bose-rings. is is nonx*0 is The answer has significance in view of theorems concerning in Chapter II. x.0 T-rings It is postulated in the definition of Veblen-Wedderburn systems and therefore probably i 60 an independent axiom. Can we define some sort of "strong homomorphism" that will preserve postulates (ii), (110), and (iii)? Then the image of such a homomorphism would of course be a Mann-ring. Can the restriction of associativity be removed from the theorems concerning idempotents in Bose-rings? Which of these theorems carry over from Bose-rings Mann-rings? What can be said about the set of idempotents in a Mann-ring? We notice that in the proof of Theorem 0 in Chapter I we used the associative law of (M,+) to establish orthogonality of the two Latin squares. Since we did not need this associative property of (M.+) to show that 3 actually determined a Latin square, it would appear as if the group M might be replaced by a loop and still have Theorem 0 hold true. This would mean a new proof of orthogonality would have to be devised and a new algebraic structure different from the present Mann-ring would emerge. 61 BIBLIOGRAPHY I. Bruck, R. H. A survey of binary systems. BerlinGottingen-Heidelberg, Springer-Verlag, 1958. 185 p. Bose, R. C. On the application of the properties of Galois fields to the problem of construction of hyper-Graeco Latin squares. Sankhya 3:323-338. 1938. Bose, R. C. and S. S. Shrikhande. On the falsity of Euler's conjecture concerning 10X10 orthogonal Latin squares. Proceedings of the National Academy of Sciences 45:734-737. 1959. Bose, R. C., S. S. Shrikhande and E. T. Parker. Further results on the construction of mutually orthogonal Latin squares and the falsity of Euler's conjecture. Canadian Journal of Mathematics 12:189-203. 1960. Euler, Leonhard. Recherches sur une nouvelle espece de quarres magique. In: Leonhardi Euleri Opera amnia,. ser. 1, vol. 7. Leipzig, B. G. Teubner, 1923. p. 291-392. Gardner, Martin. Mathematical games. American, November 19480 p. 181-188. Scientific Mann, H. B. Analysis and design of experiments. New York, Dover, 1949. 195 p. MacNeish, Harris F. Euler squares. matics 23.221-227. 1922. Annals of Mathe- Parker E. T. Orthogonal Latin squares. Proceedings of the National Academy of Sciences 45:359-862. 1959.