AN ABSTRACT OF THE THESIS OF DENNIS IRL GOSLIN for the in (Degree) (Name) Date thesis is presented Title M. S. Mathematics (Major) August 13, 1965 THE MATHEMATICAL STRUCTURE OF RECTANGULAR ARRANGEMENTS Abstract approved Redacted for Privacy (Major Professor) The author studies the class of rectangular arrangements in terms of two binary relations on the objects of the arrangement. He shows how a univalent matrix determines a unique rectangular arrangement, and how each rectangular arrangement is associated with one, two, or four distinct matrices, according to the number of corner elements of the arrangement. In this manner it is shown that the essence of rectangular arrangements can be captured in mathematical terms. THE MATHEMATICAL STRUCTURE OF RECTANGULAR ARRANGEMENTS by DENNIS IRL GOSLIN A THESIS submitted to OREGON STATE UNIVERSITY partial fulfillment of the requirements for the in degree of MASTER OF SCIENCE June 1966 APPROVED: Redacted for Privacy Professor of Mathematics In Charge of Major Redacted for Privacy Chairman of Department of Mathematics Redacted for Privacy Dean of Graduate School Date thesis is presented Typed by Carol Baker August 13, 1965 ACKNOWLEDGEMENT The author would like to thank Dr. D.S. Carter for suggesting this problem and for his advise and encouragement. He would also like to thank Professor J. Davis for critically reading this paper. THE MATHEMATICAL STRUCTURE OF RECTANGULAR ARRANGEMENTS CHAPTER I INTRODUCTION Mathematics is the study of objects of thought and relations on the objects. In this paper we will discuss a class of "simple mathe- matical structures" corresponding to rectangular arrangements objects. We will abstract the study lar arrangements to the study of class of the of relations on sets of all such rectanguof objects, inde- pendent of the nature of the objects. In this manner we will capture, in mathematical terms, the essence of the rectangular arrangement. The following are examples of rectangular arrangements: Example 1. The arrangement of the elements in a matrix. The rows and columns are specified by the definition of a matrix (see Definition 1.8 below). Example 2. The arrangement of trees in rows and columns as in an orchard. Example 3. The arrangement of people into rows and columns as in a football stadium. Example 4. The arrangement of the Gaussian integers where m and n are integers such that 0 < m < k, 0 < n < m+ ni ,Q . 2 As these examples show, the essential nature of a rectangular arrangement is quite independent particular objects which The essential thing is a set of objects make up the arrangement. of any kind, of the arrayed in rows and columns. To explain how this abstract notion can be captured in mathematical terms, first present some basic definitions leading we must up to the idea of a simple mathematical structure. Definition 1. 1. Let n be a positive integer, set. A sequence of length n in of X (xl, x2, notes the value of the function at the integer , into the set X. Such where xi de- xn) for i, The set of all such n -ary sequences is denoted by Definition 1. 2. An n -ary relation on Let X n be a X is a function, defined on the set positive integers less than or equal to n, sequences are usually denoted by and i Xn. A 1, 2 , , 1 X be a set. -ary relation is called a unary relation, a 2 -ary relation is called a binary relation, and a 3 -ary relation is called a ternary relation. The following are examples of relations on the set of all tri- angles. Example 1. Unary relation: The set of all is an equilateral triangle. n. Xn. be a positive integer and is a subset of = x such that x 3 Example Binary relation: The set 2. and x of pairs such that (x, y) are congruent triangles. y Example 3. Ternary relation: The set of triples that the area of the area of (x, y, z) plus the area of x such is equal to y z. The reader should be familiar with the following types of binary relations: Definition 1.3. binary relation A ric if for all elements Definition if for 1. 4. all elements Definition A a, b of A binary relation 1. 5. A then Definition set is symmet- then (b, a)E A. X is reflexive a, b, c of on a set A if X, (a, b) X and is transi- (b, c) be- (a, c)E A. A 1. 6. alence relation if A on a X (a, a)E A. a X, (a, b)EA if X, binary relation tive if for all elements long to of on a set A A binary relation A on a set X is an equiv- is reflexive, symmetric and transitive. In introducing the notion of a simple mathematical structure, we shall first present the abstract definition, followed by two familiar examples from algebra. 4 Definition 1.7. Let be a set of objects and X be a finite set of relations on X is the set relations on (X,A1,A2, X together with the finite set {A A2, ,An} of The mathematical structure is denoted by X. . ,A1} simple mathematical structure on A X. {A1,A2, ,An). The relations are called the distinguished Ai relations of the structure. Example where X 1. A group is a mathematical structure is the set of elements of the group, E (X, E, M, I) is the unary re- lation consisting of the identity e: E M {e }, is a ternary relation, the group operation (multiplication): M and = I = {(x,y,z):z = xy }, is a binary relation, the group inverse: I Example structure 2. (X, U, S) = {(x,y):xy = e }. The natural numbers form a simple mathematical where X is the set of natural numbers, is a unary relation consisting of the element U={1}, 1: U 5 and is a binary relation, the successor: S S = successor {(x, y): y is the of x} . Finally, we define the notion of a matrix in a set Definition integers. Let tive integers 1. 8. Let mX n (i, j) X be a set, m, n be positive denote the set of all ordered pairs of posi- such that i < is a function defined on the set X and X. m, j < mX n n. An mX n matrix in into the set X. Such matrices are often represented by rectangular arrays as in Figure 1. 1, xln x11 x12 x21 x22 . xZn ' . xmlxm2' xmn Figure 1.1 where notation i = 1, 2, x.. denotes the value of the function at the point (x..) ,m, (i, j). is also used to represent the matrix. For each the sequence (x il' x i2, . ' x. ) The 6 is called the ith row of the matrix; for each .th column. The matrix j = 1 , 2, .. , n, the sequence is called the J only if the function is one -to -one; i.e. , is univalent if and (x..) 13 x = implies xid (i,j) =(k,.) In the next chapter we shall define a rectangular arrangement as a simple mathematical structure consisting of a set binary relations R, C (for "row" and "column" ), certain conditions specified by the axioms of X and two satisfying rectangular arrange- ments. We then show that every univalent matrix determines a unique rectangular arrangement of its elements. In Chapter III we shall study the properties of rectangular arrangements and establish the general relationship between these structures and matrices; namely, that each rectangular arrangement that is not a single row or column corresponds to four distinct matrices (depending on which corner element is chosen for x11). Note: The term "mathematical structure" place of "simple mathematical structure" of this paper, for the sake of brevity. will be used in throughout the remainder 7 CHAPTER II THE MATHEMATICAL STRUCTURE (X, R, C) To set up the mathematical structure of a rectangular arrange- ment, we take the set consisting X ment and two binary relations is as follows: The pair R x is next to belongs to C if and only if = y, The interpretation of X. on C (x,y) or x R, the objects in the arrange- of belongs to R if and only if in the same row. Similarly, y x = or y, is next to x (x, y) in the y same column. The mathematical structure for this arrangement is then denoted by (X, R, C). Before writing down the axioms for this structure, it is convenient to introduce four new relations is interpreted as the set of all pairs are in the same row, and ER R (x, y) replaced by "column" Definition exists a sequence (zi, zi} 1)eR for Definition at most one The pair 2. 1. (z1, z2, i 2. 2. = such that and x Roo y , and Coo is similar, EC . (x,y) belongs to such that zn) x= R zl, if there y = zn and -1. 1, 2, The object y such that ER, EC. Here is a unary relation consisting of all end -of -row objects. The interpretation of with "row" Coo, y x x belongs to and (x, y)cR. ER if there exists 8 Definition exists a sequence (z., 1 z. 1)eC for Definition The pair 2. 3. (zl, z2, , zn) 1, 2, , n- i = 2. 4. exists at most one y such that x = C z1, y if there = zn and 1. The object x such that y The mathematical structure belongs to (x, y) belongs to # and x EC if there (x, y)eC. is called a rectangular (X, R, C) arrangement if and only if the following axioms are satisfied for all belonging to x, y, u and v X: The interpretation of Axioms 2. 1 and 2. 2 is obvious: they will be used without further explicit mention throughout the next chapter. Axiom 2. 1. The binary relations R, C on are both X reflexive. Axiom 2. 2. The binary relations R, C on X are both symmteric. Axiom 2.3a ensures that for every object belonging to X, there are at most two objects next to it in the same row. Axiom 2. 3b ensures the same property for columns. Axiom 2. 3a. z - x and There are at most two distinct (z, x)eR. z's such that 9 z 4 x and z's There are at most two distinct Axiom 2. 3b. such that (z, x)EC . Axiom 2. 4a ensures that for every row of the arrangement there is at least one end -of -row object. Similarly, by Axiom 2. 4b, for every column there is at least one end -of- column object. Axiom 2. 4a. There exists a zEER such that (x, z)ER00 . Axiom 2. 4b. There exists a zEE such that (x, z)EC . We can interpret Axiom 2. 5 co as the assertion that each row intersects each column in exactly one point (see Figure 2. 1). Y X Figure Axiom and (y, z)EC 2. 5. Z . 2. 1 There exists exactly one z such that arrangement as in Figure 2. 2 where (x, z)ER . oc If we have an are next to each other in the same row, x and u x and y are next to 10 each other in the same column, and y and v and u are in the same column, in the same row and y is next to are in the same row, v then u is next to in the same column. v v This is implied by Axiom 2. 6. x y u.... v Figure. Axiom 2. 6. then (u, v)ER In If and (x, u)cC, (x, y)ER, order to exclude the example X shall be concerned with finite sets Axiom 2. 7. We pause to We (u, v) »R 00 and (y, v)EC m (y, v)cC. jects, we require that the set 2. 7. 2. 2 The set X of an arrangement without ob- be non -empty. of In this paper we objects. is finite and non -empty. consider the independence of Axioms 2. 1 through shall illustrate their independence by giving an example of , 11 an arrangement that satisfies all but one of Axioms 2. 1 through 2.7. The proof that the example satisfies all but the specified axiom is left to the reader. The mathematical structure with and C = {(a, a), (b, b) For the sake of X = {a, b satisfies all but Axiom } (a, b) that and We r belong to R, and (a, b)cR a r Figure {(a, b), (b, a) 2. 1. (a, a) (a, b) and representation and (b, a) R, a b R r Figure 2. 3a a c 2. 3b and to denote c belong to C. belong to (b, b) of the to denote that 2. 3a as in Figure r does not belong to (b, a) Figure 2.4 to denote that understood that as in Figure shall use (b, a) = brevity in presenting the remaining examples, we introduce a figure notation to simplify the arrangement. R }, and as in It is C. b 2. 3b b Figure 2.4 The arrangement in Figure 2. 5 satisfies all but Axiom 2. 2. } 12 r r r a Figure 2. 5 It is interesting to note that Axiom 2. is replaced by Axiom 2. 6 Axiom then If If 2. 6 *. and (v, u)ER that (y, x)eR. that (u,x)EC. as follows: (x, y)ER, x = u (x, u)EC, If can be omitted if Axiom (x, u)EC, and y taking and the arrangement as in Figure y = u and 2. 6b y \ Figure A Axiom x c /r "circular arrangement" as in Figure 2. 4a , it follows u \ y 2. 4b. 2. 3b. 2. 7a w / u c 2. 6b satisfies all but and the "circular arrangement" as in Figure 2. 7b satisfies all but Axiom 2. cZ\c Figure 2, 6a = satisfies all but Axiom r u y satisfies all but Axiom z r r (y, v)EC it follows from this axiom y = An arrangement as in Figure 2. 6a x and (u, v)ER (v, y)EC. taking (x, y)cR, 2. 6* 2 3a 13 r/yr x \ r c/y\c / \// \// Figure z ` satisfy all but Axiom r 2. 7b 2. 8, will r r f e c C c C r d Figure 2. 6. c arrangements, as in Figure b C Axiom c 2. 5. a 2. 9 z Figure 2. 7a Two separate rectangular Figure x h 2. 8 represents an arrangement that satisfies all but (Take x = a23, u = a32, y r r a CI11 a l2 a cI C c r r a22, y r C = a21 a33 r CI a22 r a23 2. 9 as represented in Figure 2. 7. r c r a21 r a 12 I in Axiom 2. 6) 3--f--a32 An "infinite arrangement" a 11 a33 CI Figure satisfies all but Axiom = CI r a22 Figure 2. 10 2. 10 14 The proof that the following mathematical structures are rec- tangular arrangements is left to the reader. Example ure a, b, Consider the row of objects 2. 1. as in Fig- c 2. 11. r a r b Figure 2. c 1 1. The rectangular arrangement is given by the set X = {a, b, c together with the binary relations (b, b), (c , c ), (b, a), (a, b), consisting of the pairs 2. 1 it follows that the pairs (a, c) R oc and 2. 4 and (c, a). 2. 3 of , all pairs b), (a, a), and the relation C From Definition of the set R and also By Definition 2. 2, = (a, c }. it follows that that EC (c and (c, c). (a, a), (b, b) ER Similarly, by Definition consisting of the pairs R (b, c) consists }, = {a, b, c } . Coo= 00 C and by Definition 15 Example Consider a univalent 2. 2. matrix mX n (x..) as 1J given by Definition 1. 8 and illustrated in Figure A 1. 1. rectangular arrangement is given by the set -<i <m, X= {xi. :1 together with the binary relations R and C, xij), (xij, xij), (xin, xin): {(xij, xi(J+ 1)), (xia+ = R 1<j <n} 1 < i_< m, 1 <j <n }, and C = {(xi j, x( i +l )j), (x( i+1 )j, x iJ' iJ xij), i m x : j 1<i<m, 1<j<n}. - This mathematical structure is called the rectangular arrangement associated with the univalent matrix (x..). (X, R, C) From Definition R 0o 2. =x 13 it follows that 1 i/ ) : . 1 -< i -< m, 1 < j < n, < j < n, 1 -< < n . By Definition 2. 2, ER= From Definition {x x , it 2. 3 . in 1 : -< i <- m }. it follows that Coo= {(xij, J ): J 1 < i < m, 1 1 < Q < m } . 16 By Definition 2.4, E = {x , j The main x.: 1 < j < }. J result of the next chapter is that every rectangular arrangement is associated in this way with some univalent matrix. This result shows that our definition captures the essence of rec- tangular arrangements alone, by excluding arrangements of all other kinds. 17 CHAPTER III SOME PROPERTIES OF RECTANGULAR ARRANGEMENTS (X, R, C) In this chapter we shall derive trary rectangular arrangement We several properties of an arbi- (X, R, C). shall first show that each rectangular arrangement has a dual, obtained by interchanging R and which is again a rec- C, tangular arrangement. Definition 3. Let 1. Then the structure (X, R, C) obtained by interchanging (X, C, R) is called the dual of (X, R, C). Theorem The dual, 3. 1. ment (X, R, C) when R and changed with Proof: C EC. be a rectangular arrangement. of a (X, C, R), Coo ER It follows are interchanged, is interchanged with from Definitions Roo 2. 1 4 is interchanged with Moreover, it is easy to check that Axioms if these interchanges are made in the axioms. 2. 2 -2. Roo that when and Coo, 2. is inter- EC. 1 -2. 6 2. 3b and R ER with remain true Thus, Axiom are essentially unchanged, Axioms 2.3a and C, Moreover, are interchanged to form the dual, and and rectangular arrange- is itself a rectangular arrangement. C R 2. 1 and are interchanged, 18 as are Axioms 2.4a and 2. "dummy variables" and x 4b; Axiom 2. (X, R, C) is a is unchanged provided the are interchanged, and Axiom y unchanged if the "dummy variables" It follows that if 5 y and u 2. 6 is are interchanged. rectangular arrangement, so also is (X, C, R). An important fact, closely related to Theorem 3. 1, is that each proposition that can be proved in the theory remains true if R, and Rco and ER of rectangular structures are interchanged with C, Coo respectively. The new proposition obtained in this way will EC, be called the dual of the original proposition. To see that the dual of a proposition says the same thing about the dual structure dual of P as says about P P itself is true, notice that the is true if and only if P (X, C, R) (X, R, C). Several lemmas and theorems of this chapter are arranged in dual pairs, marked "a" and "b" such as Theorem 3.3a and b. , case only the "a" statement need be proved, since the "b" In each statement is its dual. Theorem 3. 2. The relations R and C are equivalence relations. Proof: To prove that R co is an equivalence relation, we must show that it is reflexive, symmetric and transitive (see Definitions 1.3 to 1. 5). 19 For reflexivity, let a be any object in be the sequence of length two such that z (a, a)ER; hence by Definition 2. 1, (a, a)ER For symmetry, Definition znn sequence Then i wl there is (z., z. .. b, = , wn n -1. a sequence i = w. = ,n- 1. zn -. 1, and by Axiom 2. 2, Hence (b, a)ER and , (w.,w.1+ 1 there is (v., vi+ 1)ER a sequence for all i (v1, v2, = , 1, 2, , ul = b, um= c, (u., u. such that n- 1, 1)ER .) sider the sequence w2, (w , for w. ui-n+ Then Hence w = a, (a, c)ER j = 1, wm +n -1 and , 00 We have shown assertion is that C 00 = R that and c 00 R 1 By Definition v1 a, vn= b, = ,un) 2, , i= 1,2,,n -1 for i = n,n +1,. (w.,w. m- 1. Con- such that { i 1)ER for for i a, ,n. i = 1, 2, and a sequence (u1, u2, wm +n -1) v. z 1= Consider the (b, c)ERco. such that vn) By . is reflexive. R For transitivity, suppose (a, b)ERco and 2. R such that zn) , l' z2) is reflexive. 00 pair belonging to 1, 2, such that R (z By Axiom 2. 1, a. = and , z2, (z for all ,wn) a = be any (a, b) 1)ER (w1,w2, 1, 2, = 1 and b = 2. let z2 = and let X, )ER ,m +n -1 for i =1,2, . ,m +n -2. is transitive, 00 is an equivalence relation. is an equivalence relation. The dual 20 It is II) clear from the interpretation of Roo and C 00 (see Chapter that the equivalence classes [1] of these two equivalence relations are just the rows and columns rectangular arrangement. This of the is the motivation for Definition 3. The equivalence classes of 2. and R C 00 are called rows and columns, respectively. If an object is not at the end of a row, then there must be two objects next to it in the same row. This, and the dual statement for columns, is the assertion Theorem of Theorem 3.3a -b. If x does not belong to exist exactly two objects z such that b. 3. 3a. exactly two objects Proof: z :x and that xCER. If z such that x z and (z, x)eR, by Definition 2. 2 we (x, z)cR. then there exist (x,z)eC. there are not at least two distinct z's such that reach the contradiction Thus, there are at least two such 2.3a there are at most two such z's EC, does not belong to x If # x and z then there ER, z's and by Axiom z's. Hence there are exactly two and the theorem follows. Out next theorem asserts that two objects which are in the same row, and also in the same column, are necessarily equal. 21 Theorem x = 3. 4. If belongs to both (x, y) and R then C y. By Theorem 3.2 Proof: statement (x, z)cR , co (y, z)EC (x, x)cR and 00 is true for co (y, y)EC z = . 00 Hence the and for x z y, = but by Axiom 2. 5 there is exactly one such z. Thus Lemma 3.1a asserts that the columns containing two different objects of a given row are disjoint. Figure (See x 3. 1. = z = y. ) z w Y Figure 3. Lemma 3. la. (x, y)cRoo, and w and both (z, x) be objects such that x, y, w, z and (w, y) x Co Then belong to y, z are distinct. b. and both Let 1 Let (z, x) and x, y, w, z (w, y) be objects such that belong to Roo. Then x L y, (x, y)EC z and w , are distinct. Proof: to Coo. Suppose that z = w. It follows from Theorem 3. Then 2 that (w, x) (x, y)EC and . 00 (w, y) belong By Theorem 3. 4 22 x = contradicting the hypothesis that x y, y1 y. Thus and z w must be distinct. Theorem 3. asserts that an object next to and in the same 5a column as an end -of -row object is also an end -of -row object (see Figure 3. 2). Corollary 3. la asserts the same property for any object in the same column as an end -of -row object. X x1 Y Y 1 Figure 3.2 Theorem 3. 5a. If xEER and (x, y)EC, then yEER. b. If xCEC and (x, y)ER, then ycEC. x y, then obviously yEER. Proof: y If = does not belong to z's such that objects z1 and z2. x y, suppose Then by Theorem 3. 3a there are two ER. distinct If y and z (See (z, y)ER. Figure 3.3. u1 x u2 zl y z2 Figure 3.3 ) We shall call these 23 By Axiom 2. 5 there exists a ul from Axiom It follows (u If u1 Theorem 3. = 2 then x, (z l,y) such that (x, zl) Since such that u2 # x, z1 =z2, xEER and (x, y)EC00 b. If xEEC and (x, 1 In (u2, z2)EC. that u1 2 = u2. (zl, z2)ER00, contradicting the fact that If (z.,z. y, Similarly, x. and z 1= are distinct. Hence yEER. 1)EC for i = (z z2, , then y)ER then directly from Theorem It follows on the length of a sequence and 3.4 that (zl, z2)ECoo. Since by Theorem 3. Corollary 3. la. Proof: 2 and by C, u1 (x, u2)ER, it follows from Definition 2. xEER, (ul,zl)EC. and belong to Thus y. z1 it follows from Theorem 3.4 that and z2 (x, y) It follows from Theorem EC00. Thus by Theorem 3. 2, z (ul, x)ER that and contradicting the assumption that there exists a u2 2. 6 (zl, ul)EC00 and 3. 5a zn) such that yEER. yEEC. by induction x= y= zn z 1,2,,n -1. order to establish that each rectangular arrangement is asso- ciated with some matrix, we introduce a special type of sequence. Definition 3.3. w. e = w. if and only if sequence A i = j for (w1, w2, i, j = 1, 2, , wn) such that ,n is called a uni- J valent sequence. The next theorem shows that we may always take the sequence 24 (z 1, C z2, , in Definition 2. zn) (x, y)ER If a unique univalent sequence = and u (ui, ui+ Proof: one sequence for (u., u.1+11)ER for (x, y)ER . (u1, u2, i ' , least element, which such sequence m , sequences is a nonempty set a for and x co , such that un) x u1, , such that un) 1, 2, = then there exists a y, f , x = ul, n -1. by Definition 2. such that x = there is at least 1 u1, y = m-1. The set of lengths of = 1,2,,n -1. = i um) 1, 2, = , (u1, u2, 1)EC Since i (x,y)EC If unique univalent sequence = 2. 3 then there exists x # y, and oo (u1, u2, (u.,ui +1)ER for and un b. y or in Definition Roo, to be univalent. , Theorem 3. 6a. y for 1 um and of all such positive integers. Therefore it has shall denote by n. We shall show that every we (u1, u2, , un) minimal length is a univalent se- of quence. Suppose that u. = u. i the sequence and vi= such that u.e (v1, v2, i+ j for x= V1, y for some , .Q = vn -j quc71.ce ) where j, i, j = ,n. 1, 2, and (vj, vj+ (ù1,u2, with this property; hence, u. i for vv = ui2 ,n-j+i is i+ 1, i+ 2, vn- J +i This contradicts the fact that = i < Then J 1)ER , for i = 1, 2, ,i, shorter sequence = . un) is the u. for J a .Q j. 1,2, , n- j +i shortest se- -J.. 25 The proof of uniqueness depends on the following: Proposition such that n There is no univalent sequence (u1, u2, 3. 1. (u., u. > 3, for FR 1) i 1, 2, = ... n- 1 and (ul, un) , un) . . . , E R. Proof: Suppose such a sequence (ul, u2, ... , un) exists. By Axiom 2.4a there is a such that (u1, z EER ... term of the sequence (u1, u2, no , By definition 2. 2, z) E.R09. un) belongs to ER; hence, is z not a term of this sequence. As proved above, there is a univalent sequence (vi, v2, for i 1, 2, = ... 1, 2, ... (ul, u2, i = ... , ... , vr) such that vi r -1. Since un), there is a k, , , n and yk+ = and ui -1, ui+ un = u1 if i +l tion is proved. = n. 1 ul, yr u1, and yr = = 1, 2, ... , n.. Then there are where we take uo = un if i = and ... , wm) for i (w., w. 1)E.R 1, 2, = exists a greatest integer k, - If k valent, n = = n, un and ... , m -1. = wn n = 1 < k = wm, and since (w1, w2, > n. = + 1 < . < = wm . . . n and of = Then there u, for all wm) is uni- m and the two sequences are equal. If k the least integer such that k wl, y Since the sequence n, such that w. < let 3. 6a, be a univalent sequence such that x (ul, u2, ... , un) is the shortest such sequence, m i < k. 1, Since this contradicts Axiom 2.4a, the proposi- Returning to the proof of uniqueness in Theorem (wl, w2, ui for some = such that w f vk and (vk,w)ER; namely, w yk+ 1, is not a term of z = and (vi, vi +1)ß.R z = r, such that vk < k < u.j for all j 11 1 three distinct values of w v1 = < n, let w. for some 3 .Q be 26 j = k + 1, k + 2, (uk, uk+ ... , , ... Consider the sequence m. , 'wj-1,...'wk+2'wk+1). u By the definition of k, we have sequence is of length > 3. ukl uk +1 hence, the wk+l # uk; Furthermore, by the definition of f, the sequence is univalent. By Proposition 3. 1 no such sequence exists. This contradiction rules out the case k <n, completing the proof. shall next show that a row with at least two objects has We exactly two distinct end -of -row objects. Theorem 3. 7a. If there exist distinct x, y such that (x, Witco, then there exist exactly two distinct z'svER such that (x, z)r_ R00. b. If there exist distinct x, y such that (x, y)e C00, then there exist exactly two distinct z's SE such that (x, z) S C00. Proof: Since x / y and (x, y)F. R00, it follows from Theorem 3. 6a that there exists at least one univalent sequence (u1, u2, such that x = u. for some j = 1, 2, J i = 1, 2, ... , m-1. Since the set X ... , m and (u., u. 3. is finite (Axiom ) 2. 7), ... , um) R for the set of lengths of all such sequences is a finite nonempty set of integers. Therefore it has a From Definition 3. 3, u1 # un. greatest element, which We we shall denote by n. shall show that u and un belong to ER. Suppose u1 does not belong to ER. By Theorem 3. 3a there 27 exist exactly two z's such that z#ul and (u1, z)ER. Let these z's be denoted by zl and z2. Since u2 $ u1 and (u1, u2)ER, it follows from that u 2= zi for i= or i= 2. We may assume without loss of generality that z2 = u2. We will show that (z1, u u2, , , , un) is a uniAxiom 2. 3a 1 valent sequence, contradicting the fact that (u1, u2, ... , un) is the longest such sequence. Suppose that z = the univalent sequence (zl, u, u2, for i= 3. 1 1, 2; and since no such sequence t ui for uk, and z z1 = ... i uk- 1), , = 1, 2, . , . , k -1. Now k >3 since zl uk we have (z1, uk -1)ER. t ui By Proposition exists. Thus no such k exists and is a univalent sequence. Consider (z1 , u1, ... , un) This contradiction establishes that ujIER. Using a similar argument, it follows that unf ER. To prove that u1 and un are the only two such objects, we need the following proposition: Proposition 3. 2. such that (ui, ui )`R for i +1 = , , If z is any object such that (z, to ER. some ... un) 1, 2, ... n -1 Let (ul, u2, j = 1,2, be a univalent sequence and both ul and un belong ul)ER, then z = u, for ... ,n. Proof: Follows by an argument similar to that used in the first part of the proof of Proposition 3, 1, Returning to the proof of the theorem, suppose ER contains a third object sition 3. 2 z z such that =u. for some i not belong to ER. u1 t = 2, 3, z # ... un and (u1, z)ERco. , n-1. By Definition By Propo2. 2 ui does This contradicts the assumption that zEER, and completes the proof. If an object is both an end -of -row and an end -of- column, we 28 shall call it a "corner object" We show in . Theorem 3. 8 that every rectangular arrangement that is not the special case of a single row or column has four distinct corner objects. Before we show the existence of four such objects, we shall show that there is at least one corner object for each pair of objects, where one of the pair is an end -of -row and the other an end -of- column. Lemma 3. 2. zeER (EC such that Proof: and zEEC. zeER If (y, Theorem 3. Roo oo Proof: 5 ycEC, zeER there is a 8. Coo, 0o We see If r,EC, E R z zeER, such that (x, z)EC co and by Corollary 3. lb and the lemma follows. there exists then there exists a (z, y)cR and Corollary 3. la By Therefore, belong to (z, x)eCoo By Axiom 2. z)eR. and cm a E C pair (x, y) which does not contains exactly four objects. that the hypothesis does not allow the arrange- ment to be a single row or column. As a guide in the proof we may visualize the arrangement as in Figure 3.4. 29 yl z 1 . z3 . Y x1 z x . . . 2 w x2 Y2 z4 Figure 3. By Axiom 2. (w, y)EC . co Since (x, y) from Theorem 3. 2 and w x1 and distinct (x2, x) y1 and y2 (y2, y) belong to the following C00. a such that w does not belong to that there are distinct (x1, x) there exists 5 4 # x, x2 belong to and w belonging to or C R00, # y. By Theorem belonging to R00. (x, w)ER ER such that It follows from Lemma 3. 2 it follows 3. 7a such that By Theorem 3. 7b EC and there are (y1, y) and that there exist z's: (i) z iEER SEC such that (z (ii) z2eER nEC such that (z2, y2)ER yl)ER00 and (z x1)EC00 . and (z2, x1)EC00 . 30 (iii) z3EER'mEC such that (z3, y1)ERo and (z3,x2)e Coo (iv) z4EER,-EC such that (z4, y2)ERco and (z4, x2)ECco We see that z Since (y1,y2)ER0. for f z 2; zk # zß for that Now suppose from k,1 1, k (z1, u1)ECco (u, u2)ECco. and and ER EC 5. y1 = contra- y2, Similarly, it follows that contains a fifth point there exists a 5 and a (u, ul)ERco, (See Figure 3. y2. 3.4 2 1,2,3,4. By Axiom 2. z2, z3, z4. z = then by Theorem 3. = z z by Theorem (y1,y2)ECco, dicting the distinctness of y1 if u2 distinct such that u1 such that u, u2) (z ERoo and ) u2 z 1 u u1 Figure 3. By Corollary 3. lb, ul If = u1 z1 u1 = la, and or ul = 2, and u2 u1 belong to u2 then z1, Theorem 3. 3. z2, (z 5 belong to EC. Thus by Theorem and by Theorem 3.7a u2) (u2, u)ER. and (z and by Corollary ER u) u2 = belong to z1 3. 7b, or Roo u2= z3. and by It follows from Theorem 3. 4 that u2= u. 31 This contradicts the supposition that u so that # z1, u1 = u1 z2. = zi for u Similarly, u2 contradicting the assumption that z4, = i = 1,2,3,4. Thus and by Axiom 2. z3 zl, is distinct from u 5 Thus the theorem follows. z2, z3, z4. Theorem 3.9 implies that a rectangular arrangement with one object has one corner object, and a rectangular arrangement having one row or one column and at least two objects has exactly two corner objects. Theorem 3. fied; i, e. If the conditions 9. every pair , for Theorem 3. y)ERUC, (x, are not satis-. 8 then ERS EC contains one or two objects. Proof: Definition set If the 2. 2 X contains exactly one object and by Definition zeER ERA EC. belongs to If the set contains at least two objects, let X two distinct objects. Then (x, y)ER (x, y) (x, y)EC00 but not both. Suppose that Cog does not belong to Let w does not belong to Roo, does not belong to Rog Co By Theorem 3. 2 uC 00 R rThC 00 . 00 (x, y)eR00 Thus it follows from Theorem 3. (x, y)EC00, (w, x) and does not (x, y) X. 2 or (x, y)ER and be any object belonging to Thus both be any x, y It follows from Theorem . 00 3.4 that belong to then by Thus one object zeEC. 2. 4 z, that (w, y) If (x, w) (w,y) belong to contradicting the supposition that 32 Thus (x, w)ER . Hence, there is o one row. It follows from Theorem 3. 7a that there exist exactly two (x, y) does not belong to u's belonging to ER. If follows from Theorem 3. Thus uEER EC. Coo. there exists a that 4 u = y. such that y Thus by Definition 2. 4 ucEC. there are ER, Since only two objects belong to it (y, u)cC, exactly two corner objects. Similarly, if and (x,y)EC00 does not belong to (x, y) Roo, there is one column with two corner objects. Figure 3. 6 illustrates Theorem z w x y Figure 3. Theorem 3. C , and 1Oa. (z, w)ER , b. R , and (z, w)EC Proof: , If then ul, By Axiom 2. 5 there exists (y, w) belong to 1 and z = un a (x, z) and belong to (y, w) (z,w)EC. By Definition 2. = and (x, z) (z,w)ER. (x, y)EC, If x such t h a t 6 (x, y)ER, then 3. 1Oa. v2 there exists (ui' ui+ l)EC such that a sequence (u1, u2, for i = 1, 2, (v2, u2)ERco and un) , n -1. 33 and by Axiom 2. (v2, y)ECco, By _ontinuing in this way, (v2, v3, i ä that ,n-l. 2, 3, u v = If a row (vn, un)eR and (v2, y)eC. sequence (vi, vi+ 1)cC for It follows from Definition 2.3 and Theorem 3. and for n Thus u. we obtain a Hence the statement (y, v )eC . n 00 true for such such that vn) , and (v2, u7)eR 6 u = v = n u = w. and 00 By Axiom 2. Hence w. (z, u)cR 5 (y, u)eC 00 2 is there is exactly one (z,w)eR. has at least two distinct objects, and one object in the row is a corner object, then there exists another object in the same row that is a corner object. Lemma 3. 3a. and xeERn EC. xeERn EC. and (x, z)eC00, (x, z)eRoo and Therefore, only one Let x y, (x, y)eR such that z be objects such that Then there exists a unique z 00 # x, x # y, (x, y)EC such that z z # x, zeERn EC. zeER. zeERn EC. z x, y By Theorem 3. 7a Proof: be objects such that Then there exists a unique b. and x, y zeERn EC. and (x, z)eR00, Let This is the assertion of Lemma 3.3a. there exists a z such that It follows from Corollary 3. lb that Furthermore, by Theorem satisfying the lemma. 3. 7a, z zeEC. there is x, 34 Lemma 3. 4a asserts that any row with at least two objects can be arranged in a univalent sequence in exactly two ways. Lemma 3. 4a. there exist distinct x,y If such that then there exist exactly two distinct univalent sequences such that x for some u. = j = ,n, 1, 2, ul both (x, y)ER (u1, u2, and , , un) un J belong to ER, and then if (z, x)ER u. (u., 1+1 for )ER for some z = u. i 1, 2, = , moreover, n -1; J. J b. x,y such that there exist distinct If then there exist exactly two distinct univalent sequences such that x for some u. = ,n, j = 1, 2, both (x, y)EC (u1, u2, , .,un) u1 and n -1; moreover, un J belong to and E (z,x)EC00 then if m z2. z = It follows by sequence i = 1, 2, i = 1, 2, . = z1, , there are exactly two distinct un = z2 and Proposition 3,2 that x ,wn) ,n. Then w1 n -1, 1, 2, z's = (u., 1 u. = such that 1)ER for for some u. w. = un -i+ i j. 1 wn = z1, the two univalent sequences are different. and ,u1 (u1, u2, 2, = 1, , -1. Consider the for (w.,w. 1)ER and the sequence is univalent. Since z2, zl be denoted by there exists a univalent sequence (wl,w2, , = J Let the (x, z)ER00. By Theorem 3. 6a u i for some u. By Theorem 3. 7a such that such that for 1)EC 3 Proof: z'sEER (u., ui+ for w ul 35 Now suppose that (v1, v2, such that vm) , there exists a third univalent sequence x = for some v. 1, 2, j = » m, , J both . i = 1, 2, z 2= and vm and v1 , belong to vm, z 1= or v. and n. for u. = 1 1 i = vm 2. 2 and that v2 n = = = u2 If v that z v 1= z u2, ( u1 vl = = it u 1 and from Axiom 2. 3a ,k where k is the smaller 1, 2, Since the sequences are univalent, z2 for (vi, v. 1)cR m -1. It follows from Theorem 3. 7a follows from Definition that and ER ,un) and ( v1 of v2, m , v) This implies that the two sequences are the m. same, contradicting the assumption that they are different. Thus z2 = m = and v1 n and z1 w. i = = Using a similar argument, it follows that vm. for v. i = 1, 2, ' . ,n, contradicting the assump- tion that they are different. Hence there exist exactly two such sequences. If a (xi.), rectangular arrangement is associated with a univalent matrix we shall show that x11 has to be one of the corner objects. J Once this corner object has been picked, this matrix is uniquely determined. This is the assertion Lemma 3. 5. If a of the following rectangular arrangement ated witha univalent matrix (xi.), then lemma. (X, R, C) x11cERfm EC. If a is associ- rectan- J gular arrangement is associated with two univalent matrices (x..),(w..) 13 1J with x11 - w11 then (x..) J = (w..). J 36 It is easy to verify by Example 2. Proof: Suppose the matrices 2 are (x..) and (w..) xl that 1EER r Ec. dimension of J m1X n1 and n1 It follows from Definition 2. n2. induction that loss (x In of , x respectively. Let n be the lesser m2X n2, and w = lj x generality that 1(n -1) the fact that Similarly, and ) n2 (x In , > w n1. 1(n +1) 2. 1 3a, < i < m, by induction on we have xij belong to ) 2. 2). = wij contradicting n1 and m1 then 1)' R, Thus by assume without We there exists a m be the lesser of that the matrices (x..) and (w..) 13 13 i, If and Axiom 2. 3a ,n. j = 1, 2, (see Example x1nEER letting for lj 2 of = n2 = n. it follows m2, are both mX n. For each fixed using Definition 2. j for all j, 1 < j < n. 2 and Axiom Thus (xij) = (wij) and the lemma follows. In Chapter II (Example 2. 2) we saw that a rectangular arrange- ment is associated with every univalent matrix. The question we now consider is: For every rectangular arrangement a mX n ment univalent matrix (X, R, C) (x..) (X, R, C) is there such that the rectangular arrange- is associated with this matrix (x..) 13 ? The answer to this question is yes; in fact for every rectangular arrangement there is one, two or four distinct matrices, depending on the number of corner objects. This will be proved in the next theorem. Theorem 3. 11. Let (X, R, C) be a rectangular arrangement 37 with n corner objects. Then is associated with exactly (X, R, C) univalent matrices. n From Theorem 3. Proof: and 3. 8 9 to consider the following four cases. Case gular arrangement with one object, it is obviously sufficient 1 will consider the rectan- Case 2a will consider the rec- tangular arrangement having one row such that the set contains at least two objects, Case will consider the rectangular arrange- 2b ment having one column such that the set contains at least two objects and Case will consider the rectangular arrangement having at least 3 two rows and two columns. Case the set X 1. Let be a rectangular arrangement such that (X, R, C) contains exactly one object. This rectangular arrange- ment is associated with exactly one matrix; namely, the one -by -one matrix (x11), Case 2a. where Let is the single member of x11 (X, R, C) only one row, and such that X X. be a rectangular arrangement having contains at least two objects. By Lemma 3. 4a this row is the range of either of two distinct univalent sequences. Let one of these be denoted 1X n matrix (x) with xlj=ui for (u1, u2, j = 1, 2, ,un). Define the , n. Using J Example 1 X n 2. 2, matrix the rectangular arrangement associated with this .) J is the set 38 X= {xl.J : and the two binary relations R = and R 1 < j < n }, C, {(xlj,xl(j +1))' (xl(j+ 1)' x1j), (xlj,xlj), (xln'xln): < j < 1 n}, and C = {(x , lj x lj ): 1 < j < Thus we see that with this 1X n n } . is the rectangular arrangement associated (X,R,C) matrix (x1 ). J Similarly, define the 1X n matrix (x1).) with x1 = j for j = 1, 2, ' ' ' ,n. It follows that arrangement associated with this un- j + 1 (X,R,C) is also the rectangular 1X n matrix (x1 ). j The two matrices are different since their one -one elements are different. By Lemma 3. 5 there exist at most two such matrices. Thus each row is associated with two distinct univalent matrices. Case 2b. Let (X,R,C) only one column and such that be a rectangular arrangement having X contains at least two objects. Using an argument similar to that used in Case 2a, it follows that this rectangular arrangement is associated with two distinct matrices coinciding with the two univalent sequences of Lemma 3. 4b. 39 Let Case 3. be a rectangular arrangement having (X, R, C) at least two rows and two columns. By Theorems 3. 8 exist four distinct elements belonging to ERr.EC. Let one of these w, y, (ul, u2, be denoted by z. Let - and un) , and 3. (v1, v2, , - there 9 vm) be defined as follows: 1) we ERr.EC 2) and and (y, z)eC (ul, u2, , 3. 6a such that is the univalent sequence given by Theorem un) ul, z = (vl, v2, w , = un vl, For fixed i, j, 1 z object given by Axiom y - m, - < j < 1 n, 1 < j < such that (xi., xij )ER Theorem 3. l0a that for 1 < i < (x(i+ 1)j' xij) m, 1 1<. < n, J and 1 belong to n. C By for < 00 x.. - ,m -1. be the unique < i < (xi(j+ 1)' m, v.)cR (x.., 13 1 00 Thus it follows from Theorem 3. 10b, 1 and for (x.., )e:C 13 xkj 00 k < m. and (xij, xi(j+ 1)) < j < let (x.., u.)eC 1J n, ,n -L 1J J <i <m, j = 1, 2, J l)eC for i= 1, 2,vm and (v.,v. 1 = < i < 2. 5 By Theorem 3. 2, for (u., u.+ l)eR and is the univalent sequence given by Theorem vm) = 3. 6b such that 1 such that y # z, . J 4) z, w (w, z)eRco is the object given by Lemma 3. 3b y ycER SEC 3) is the object given by Lemma 3.3a such that w 1 xij) belong to (xij, x(i+ 1)j) < j < n. R and By Axiom . 40 2. 1, (x.., x..)cR 13 13 Corollary 3. la, il and x. by Corollary 3. lb, for (x.., x..)cC 1J 13 and x1 Now suppose that in x.. 1J . . = xm x . . . 1 u. for By -< i -< m 1 for EC us . < j < n. 1 rs . (See Figure 3. 7. 1 < j < and n. ) n s x. lj v . r v ER belong to j J v. - - m, < i < belong to x. and j 1 x rs m Figure 3.7 By Axiom 2. 2, (v., v )ER 1 r sequence 00 . (x.., x )ER. It follows from Theorem 3. 1J rs By Theorem 3. 4, (v1, v2, it follows that u. , = v) us, v. 1 = v is univalent. = x rs if and only if i = (r, s) = (i, j). r that since the Using a similar argument, which is true if and only if J x..1J r ; hence, 2 j = s. Thus 41 (xil' xrs)ER for (r, s) 4 (i, 1), (r, s) (i, 2), then by Definition 2. 2 xil does not belong to ER, contradicting the fact If that xil EE R . then by Definition the fact that (xi If xrs)ER 2. 2 x. x. EER' If in (r, s) # (i,n -1), (r, s) i (i,n), for (xi. , xrs)ER for 1 contradicting ER, does not belong to m, < i < 1 < j < n, J (r, s) # (i, j), (r, s) (i, j+ 1), (r, s) # (i, j -1), Axiom is 2. 3a contradicted. Thus it follows that R = {(xij, (xi(j+ 1), x..), (xij, xij), (x in xin ): x. (J+ 1)), 1 m, < i < 1 < j < n }. Using a similar argument, it follows that C = {(x..iJ , x 1)j 1 ) ( x(i+ ): x.), (x , x x ), (x.., iJ iJ mj mj 1)j' ij associated with at least one and i z4 1, 2, 3, 4. = < i < - m,' < j < n }. Thus we see that a rectangular arrangement z3 1 mX n univalent matrix. be the four corner objects. Take Then the rectangular arrangement ated with these four mX n is (X, R, C) xll Let = zi (X, R, C) z1, z2, for is associ- univalent matrices. The four matrices are distinct since their one -one elements are different. By Lemma 3. 5 there exist at most four such matrices. Thus the theorem follows. 42 With this theorem we have completed the association of rectan- gular arrangements with univalent matrices. 43 CHAPTER IV EXTENSIONS OF THE CONCEPT OF RECTANGULAR ARRANGEMENTS Two dimensional rectangular arrangements can be generalized further by allowing the set property of Axiom 2. 7. X to be infinite; i. e. , deleting the finite In this case rectangular arrangements can still be associated with univalent matrices except that the matrices may be infinite. If there is only one corner object, then the set has exactly one element or the set is infinite. If the set is infinite, then either the rectangular arrangement is a row as in Figure column, or an arrangement as in Figure 4. all a13 a12 Figure all a12 a21 a22 4. Figure 4. 1 2 2. 4. 1, or a 44 If there are two corner objects and the set is finite, we have a single row or single column. If the set is infinite, then we have an arrange- ment as in Figure 4. 3a or Figure 4. 3b. all a12 all a12 aln a21 a22 a21 a22 a2n aml amt Figure If Figure 4. 3a 4. 3b there are more than two corner objects, then there exist four cor- ner objects and Axiom Figures 2. 7 4. 4 and 4. 5 follows. typify infinite arrangements that are not included in this generalization. al -1 a10 all alt a2-1 a20 a21 a22 am -1 amo aml amt Figure 4. 4 45 a-1-1 a-10 a-11 a-12 a0-1 0-1 a00 a01 a02 al-1 a10 all a12 Figure We have mension. 4. 5 considered only rectangular arrangements It seems that with of two di- appropriate relations it would be pos- sible to capture in mathematical terms rectangular arrangements of di- mension n. As an example it seems that a three dimensional rectangu- lar arrangement as in Figure relations, R and C ,/t112 I //'1212 212 a211 I // ra121 a221 a312 a311 could be captured with three binary as given and a third binary relation "next to in depth". alll 4. 6 r122 I 1222 I /13 ja232 a131 a231 a33 2 a3 22 a3 a321 Figure a331 4. 6 2 D for 46 We have seen that it is lations the essence of a possible to capture with two binary re- rectangular arrangement. Can arrangement be captured with one binary relation? that a more abstract kind of a rectangular It would seem rectangular arrangement could be pre- sented with a single binary relation N, for "next to" . In this abstraction the distinction between rows and columns would be lost. It seems that this rectangular arrangement would be associated with one, four or eight matrices, depending on the number of corner ob- jects. Further work is being done.