b;y

A. Hedayat and E. Seiden( 2 )

Cornell University and Michigan State University

ABSTRACT

The theory of F-sq_uares and especially of mutually orthogonal F-sq_uares

aims at a generalization of the concept of latin squares and widely considered
search for mutually orthogonal latin sq_uares. An F-square is an n x n square
whose cells are filled with k ::::: n distinct elements in such a vmy that each element
appears an eq_ual number of times in each row and column of the square.

The concept of F-squares and mutually orthogonal F-sq_uares is not entirely

new. It has been considered directly or indirectly by D. J. Finney, H. T. Federer,
G. H. Freeman, and s. Addelman. Hm-vever, these authors considered the F-squares
as a by-product of their general interest in experimental designs and were not
concerned with the theory of F-squares per se. The first author of the present
paper was inspired by the examples of F-squares brought fo1~ard by the previous
authors. Having in mind the usefulness of the F-squares for research in the theory
of the designs and its application in practical problems, he defined the concept
of F-squares and mutually orthogonal F-sq_uares and obtained some results concerning
them. The purpose of this paper is to develop further the theory of F-sq_uares and
bring it to closer attention of mathematical statisticians. We plan to further
the research in this area and present more results for publication shortly.

It may be worthwhile to point out at this stage that the theorems proved
thus far aim at classifying some types of latin squares maldng use of their rela-
tion to F-squares. The concept of orthogonality of latin squares is generalized
to F-squares and some of its implications to the problem of existence of orthogonal
latin sq_uares are investigated. It is pointed out here that using the concept of
F-sq_uares one can distinguish between ~vo types of latin squares both mateless in
respect to orthogonality but different in respect to their use in a broader sense
of experimental designs.

(1) Paper No. BU-2o4 of the Biometrics Unit, Cornell University, Ithaca, N.Y.
(2) On leave from the Dept. of Statistics and Probability, Michigan State
University.
Research was supported by NIH GM-05900-11.
 by
A. Hedayat and E. Seiden( 2 )
Cornell University and Michigan State University

0 - Introduction.
. The theory of F-squares and especially of mutually orthogonal
'
F-squares aims at a generalization of the concept of latin squares and widely con-
sidered search for mutually orthogonal latin squares. An F-square is an n X n
square whose cells are filled with k ~ n disti~ct .elements in such a way that each
element appears an equal number of times in each row and column of the square.
The concept of F-squares and mutually orthogonal F-squares is not entirely new.
It has been considered directly or indirectly by Finney [3 1 4,5], Federer [2],
Freeman [6], and Addelman [1]. However, these authors considered the F-squares as a
by-product of their general interest in experimental designs and were not concerned
with the theory of F-squares per se. The first author of the present paper >'las in-
spired by the examples of F-squares brought forward by the previous authors. Hav-
ing in mind the usef~ess of the F-squares for research in the theory of the
designs and its applic~tion in practical problems, he defined the ~qncept of F-
squares and mutually orthogonal F-squares and obtained some results concerning
them [7]. The purpose of this paper is to develop further the theory of F-squares
and bring it to closer attention of mathematical statisticians. We plan to further
the research in this area and present more results for publication shortly.

It may be worthwhile to point out at this stage that the theorems proved thus
far aim at classifying some types of latin squares making use of their relation to
F-squares. The concept of orthogonality of latin squares is generalized to F-squares
and same of its implications to the problem of existence of orthogonal latin squares
are investigated. It is pointed out here that using the concept of F-squares one
can distinguish between two types of latin squares both mateless in respect to
orthogonality but different in respect to their use in a broader sense of experi-
mental designs.

(1) Paper No. BU-2o4 of the Biometrics Unit, Cornell University, Ithaca, N.Y.
(2) On leave from the Dept. of Statistics and Probability, Michigan State
University.
Research was supported by NIH GM-05900-11.

-1-
 -2-

I. F-Squares
Definition 1.1. Let A= [a .. ] be an nxnmatrix and let z:::· = (c1 ,c 2, ••• ,c) be
~ J..J • m
the ordered set of distinct elements of A. I.n addition, suppose that for each

k=l, 2, ••• ,m, ck appears p:tecisely Ak times (Ak :2: 1) in each rm-1 and in each
column of A. Then, A will be called a frequency square or, more concisely, an

F-square on I: of order n and frequency vector (t.,1 ,x 2 , • •• ,>..m) •

We now introduce some notation. A matrix A will be said to be an

F(n;A.1 ,t., 2 , ••• ,A.m) square if A is an F-square of order nand frequency vector

(A.1 ,A. 2 , ••• ,Am). This notation may be contracted by.the use of pmvers to denote
successive equal values of A.'s. Thus F(n;A.m) represents F(n;A.,A. 1 ••• ,A.) while

F(n;A.f,x 3,x4,A. 6 , ••• ,A.m) represents F(n;A. 1 ,A.1 ,A 3,x4 ,A4 ,x 6, ••• ,A.m). In particular,
in an F(n;A.m) square, m is determined uniquely by nand A.; hence we will represent
such a square simply by F(n;A.).

Examples:

1) Let 2:: = (1,2,3} then

1 2 3 3 2 1
2 3 1 1 3 2
3 1 2 2 1 3
3 1 2 2 1 3
2 31 1 32
1 2 3 3 2 1

is an F(6;2) square on I:.

2) Let I: = (1,2 1 3,4} then

1 2 3 4
2 1 4 3
34 1 2
4 32 1
 -3-

is an F(4;1) square onE, while

1 2 3 4 1
1 1 2 3 4
4 1 1 2 3
34 1 1 2
2 3 4 1 1

Note that F(n;l) square is simply a latin square of order n and exists for

all n.

Lemma 1.1. An F(n;A. 1 ,}.. 2 , ••• ,A.m) square~ E = (a1 ,a2 , ... ,am} exists if~ only
m
if n = E A..
i=l ~

Proof. The necessity of this condition is obvious by the definition of an

F-square. Sufficiency can be proved as follmTs: Construct an F(n;l) square on

an ordered set 0 = (b1 ,b 2 , ••• ,bn)• Partition 0 into m disjoint subsets

s 1 ,s 2 , ••• ,sm such. that Si contains A.i elements. Define a many-one map cr from

0 onto E as follows:

cr(x) = a.
~
if and only if xEs.,
~
•

If we now apply cr to the elements of F(n;l) square then we obtain an

Example •
Let us construct an F(6;1,2,3) square onE = (A 1 B1 C}. To do this, con-

struct an F(6;1) square on, sayO = (1,2, ••• ,6} such as

 -4-

1231:-56
236145
362514
4 5 1 2 6 3
514632
645321 •

Then let 0 = s1 U s 2 U s 3 ivhere s1 = (1}, s2 = (2, 3}, and s3 = (4 ,5,6}. And

let a(s1 ) ~ A, o(s 2 ) ~ B, and a(s 3 ) ~ c. Now applying cr to the above F(6;1)
square we obtain

ABBCCC
BBCACC
BCBCAC
CCAB CB
CACCBB
CCCBBA •

Before proceeding further, it should be noted that this idea is not

entirely 'iTithout practical importance. For example, if we let the elements
of them-set Z to be treatments then an F(n;A1 ,A2 , ••• ,~m) square onE is an
experimental design having the properties that:
a) Treatment effects 1 row effects (read treatment effects are orthogonal

to rmT effects),

b) Treatment effects 1 column effects, and if~- =A, i=l,2, ••• ,m, then we
:l

also have

c) Treatment arrangement is balanced within rows and columns

d) Rmv effects 1 column effects.

 -5-

II. On the orthogonality of F-squares

Now we rigorously introduce the concept of the orthogonality of F-squares
and then of latin squares as a special case of F-squares.

~~.S~· Given an F-square F1 (n; ~ 1 ,"" 2 , ... ,)-k) on a k-set .E = {a1 ,a2, ••• ,ak}
and an F-square F2 (n;u1 ,u2 , ••• ,ut) on at-set 0 = {b1 ,b2, ••• ,bt}• Then we say F2
is an orthogonal~ for F1 (and write F2 1 F1 ) if upon superposition of F2 on
F1 , a. appears )..u. times with b .•
~ ~ J J

We now give a series of examples to elucidate the content of this definition.

The sets .E andO will for convenience be taken as·the sets of integers 1,2, ••• ,k
and 1 1 2, ••• ,t respectively.

F1 (6;1) F2 (6,;1 1 2 1 1 3)

123456 123425
216543 4 5 2 3 1 2
341265 232541
1)
4 6 5 1 3 2
l 214253
532614 541232
654321 325124

F1 (6;2) F2 (6;2)

1 1 2 2 3 3 1 2 3 1 2 3
1 2 1 3 2 3 2 31 1 2 3
2 1 1 3 3 2 2 3 1 3 1 2
2)
2 3 3 1 1 2
1 3 1 2 2 3 1
3 3 2 2 1 1 1 2 3 2 3 1
3 2 3 1 2 1 3 1 2 3 1 2
 -6-

F1 (5;1,2 2 ) 1<.,2(5;1 2 ,3) ··t:

. ~ ··-·- ·•

1 2 2 3 3 1 2 3 3 3
31 2 2 3 331 2 3
3)
3 3 1 2 2 1 2 3 3 3 1
2 3 31 2 3 1 2 3 3
2 2 3 31 3331 2

F(4 ;1) F(4;2)

1 2 3 4 1 1 2 2
4 1 2 3 1 2 2 1
4)
34 1 2
1 2 2 1 1
2 34 1 2 1 1 2

F(3;1) F(3;1)

1 2 3 1 2 3
5) 2 3 1 1 3 1 2
31 2 2 3 1

~ 2J2.
~
Let S.~ be an n.-set,
~
i=l 1 2,.,.,t. Let F.~ be an F-square of

order non the set S. with frequency vector

~
i.
~
= (~. 1 ,~. 2 , ••• ,~. n ).
~ ~ ~~i
Then, we

say {F1 ,F2 , ••• ,Ft} is a set of t mutually (pair-wise orthogonal F-squares if

F.
~
1 F.,J iFj, i 1 j=l 1 2, ••• ,t. In particular, if n.=n, i=l,2 1 ••• ,t, and every
~

F. is of type F(n;l) 1 i.e. a latin square of order n, then v1e denote such a set
~

as an o(n,t) set.

Example,
The follmqing three F-squares are mutually orthogonal.
 -7-

2
F1 (5;2 2 1 1) F2 (5;1 1 3) F3 (5;1 3 ,2)

1 2 3 1 2 1 2 3 3 3 1 2 3 4 4
2 1 2 3 1 3 3 1 2 3 34 4 1 2
1 2 1 2 3 1 2 3 3 3 1 1 4 1 2 3 4
3 1 2 1 2 3 1 2 3 3 2 34 4 1
2 3 1 2 1 3 3 3 1 2 1~4123 •

___
,....,..,
Definition 2.3.
~,.,...,....
Let E be an n-set. Let F be an F-square of order n on the set

L with frequency vector~= (~ 1 ,A 2 , ••• ,At)• Then, we say F is of degree! with

respect to the decomposition n = u1 +u2+••• +us if there exist r-1 F-squares

F1 ,F2 , ••• ,Fr-l on an s-set 0 with frequency vector u = (u1 ,u2, ••• ,us) such that
{F,F1 , ••• ,Fr_ 1} is a set of r mutually orthogonal F-squares and r is the largest

such integer. In particular, if A·=l; u.=l, i=l,2, ••• ,t=n 1 j=l 1 2,, •• ,s=n 1 i.e.
J. J
F,F1 ,F 2 , ••• ,Fr-l are latin squares we say that F is of type E(n,r). F is said
to be orthogonally mateless with respect to the decomposition n = u1+u2+••• +us
if its degree is one. F is said to be orthogonally~ if its degree is at

least tv10 with respect to every decomposition of n.

~~· A set of r F-squares {F1 ,F 2 , ••• ,Fr} is said to be a mutually

(pair-wise) orthogonally rich set if F. is orthogonally rich and F.

J. J.
l F.,
J

Now we give a series of examples to clarify the above definition.

Examples •

1. The degree of
123456
236145
F =3 6 2 5 1 4
415263
541632
654321
 -8-

is at least 2 with respect to the decomposition 6 = 1+1+1+1+2. An orthogonal

mate with respect to this decomposition is

1 2 5 5 3 4
513245

435521
251453 •

Indeed the degree of the above F-square is at least 2 with respect to any
decomposition of 6 except 6 = 1+1+••• +1.
2. The following F-square

1 2 3 4
F=4123
3 4 1 2
2 3 4 1

is orthogonally mateless with respect to the decomposition 4 = 1+1+1+1. How-

ever, its degree is at least 3 with respect to the decomposition 4 = 2 + 2. A
pair of mutually orthogonal mates for F with respect to this decomposition is:

1 1 2 2 1 2 1 2
1 2 2 l 1 2 1 2
Fl = and
2 2 1 1 2 1 2 1
2 1 1 2 2 1 2 1 •
3. The following F-square is of degree 3 with respect to the decomposition

4 = 1+1+1+1.
1 2 3 4
F=2143
3 4 1 2
4 3 2 1
 -9-

A pair of mutually orthogonal mates for F with respect to this decomposition is:

1 2 3 4 1 2 3 4
3 4 1 2 2 l
Fl ;::: and F2 = 4 3
4 3 2 l 2 l 4 3
2 l 4 3 3 4 l 2

(By the following theorem, each of the F-squares in the last example above is

orthogonally rich.)

The following theorem is of fundamental importance in the development of

the theory of orthogonality of F-squares.

Theorem 2.1. An F-square R ££ ~ ~ ~ is orthogonally rich if and only if

it has~ orthog0nal mate with respect to the decomposition n = 1+1+••• +1.

Proof. To prove the sufficiency, let A be an orthogonal mate for R with respect

to the decomposition n = 1+1+ ••• +1. Then, by lemma 1.1 in section one, A can

be used to generate any other F-square of order n with frequency vector

~ = (~ 1 ,A 2 , ••• ,As) withAl+ A2 + ••• +As = n. Since A 1 R, therefore those

F-squares derived from A will be necessarily orthogonal to B. The necessity

part follows directly from definition 2.3.

Examples •
1 2 5 5 3 4
5 1 3 c:.'"' 4 5
1 5 2
l. B= 3 5 4
5 4 2 3 l 5
435521
251453
is orthogonally rich. An F(6;l) orthogonal mate forB is:
 -10-

123456
236145
A=362514
4 1 5 2 6 3
541632
654321 •

From theorem 2.1 we note that all other orthogonal mates for B corresponding to

different decomposition of n, i.e. F(6;6), F(6;5,1) 1 etc., can be obtained from

A.
1 2 3 4 5
3 1 2 54
2. B =2 4 5 1 3
4 5 1 3 2
5 34 2 1
is not orthogonally rich since it is mateless with respect to the decomposition

5 = 1+1+••• +1, Mann [9].

1 2 3 4 5
5 1 2 34
B=34512
2 3 4 5 1
4 51 2 3
is orthogonally rich. An F(5;1) orthogonal mate for B is

1 2 3 4 5
4 5 1 2 3
A=51234
3 4 5 1 2
2 3 4 5 1 •
Note that {A,B} is a mutually orthogonally rich set.
 -11-

~ ~· A sub-F-square of order t and frequency vector (A. 1 ,J.2 , ••• ,A.k)

.. \ .
denoted by SFS(t;A. 1 ,A, 2 , ••• ,)..k) is an F-dqhare of order t and frequency vector
(A. 1 ,A. 2 , ••• ,A,k) embedded in a larger i~square. If an F-square has only the

trivial SFS(l;l), then we say F contains no SFS.

Examples.
1. The underlined cells in the follo\·ling F-square form an SFS (2;1) •
..
1 1 2 2
1 2 2 1
2 2 1 1
2 1 1 2 •

2. The following F-square has no SFS of any order.

1 2 3 4 5
2 3 4 5 1
3 4 5 1 2
4 5 1 2 3
5 1 2 3 4 •

Proposition 2.1. If L is an F(n;l) on an n-set L then: a) L cannot have an

SFS(t;l) if n is odd and t ~ n~l , b) L cannot have an SFS(t;l) if n is even

n
and t ~ 2 +1.

Proof. a) Suppose that L has an SFS(t;l). \'le may assume without loss of

generality that this SFS(t;l) is formed by the first t rows and columns of L.

Then the rectangle formed by the first t rows and the remaining n-t columns

could not include any element of the SFS(t;l). On the other hand, t ~ 2n+l
. 1'
~p ~es n- t < 2
n-1 . Hence in each row some of the elements would have to appear
more than once which contradicts the property of L. The proof of b) is analogous.
 -12-

It is obvious by theorem 2.1 and the series of examples 11hich were given

before in this section that, if an F-square F(n;A 1 ,A 2 , ••• 'At) is _or~_hogonally

; ,_ i ,;

mateless with respect to the decomposition n = 1+1+ ••• +1, then~~ not
necessarily mateless with respect to a coarser decomposition of n. However, if

F is orthogonally mateless with respect to any then it is o1~hogonally mateless

with respect to the decomposition n = 1+1+ ••• +1.

Mann [9] proved that if a latin square L of order n = 4t+2 has a sub-latin
square of order 2t+l then L is orthogonally mateless. In the language of the

theory ofF-squares Mann's result states that any F(n;l), n = 4t+2, containing

an SFS(2t+l;l) is orthogonally mateless with respect to the decomposition

n = 1+1+ ••• +1. The following theorem provides us with a much stronger result,

viz. it says that such an F-square is orthogonally mateless idth respect to a

coarser decomposition of n than n = 1+1+••• +1. Therefore, Mann's [9] result

turns out to be a special case of this theorem.

Theorem 2.2. Let L be ~ F(n;l) square ·2£ ~ n-set E, n = 4t+2, t ~ positive

integer. Then L ~ orthogonally mateless ~ respect to the decomposition

n = x+y if L contains ~ SFS(2t+l;l) and x is ~ ~ integer.

Proof. Let m = 2t+l. Denote the given SFS(m;l) by 1 1 • With no loss of generality
we can assume that 1 1 occupies the square formed by the first m rows and columns
of 1. Partition L as follows:

L = •
 -13-

Note that 1 2 ,· 1 3, and 1 4 are also SFS(m,;l). This is so because 1 is an F(n;l)

square. Note also that, the squares 1 1 and 14 (and similarly 1 2 and 1 3 ) contain

the same elements of !:. Now if 1 has an orthogonal mate vlith respect to the

decomposition n = x+y, then this will mean that there is an 1 1 = F(n;x,y) on

say 0 = {A,B} such that 1' l 1. This implies that upon the superposition of

1 1 on 1, A will appear x times on each ro"'"l, column and element of !:. Assume

that r A's appear on 1 1 • This implies that r A's will also appear on 14 • Since

the contents of 11 and 14 are the same, this means that, under the orthogonality

assumption of 1 and 1', x(2t+l) A's should appear together on 1 1 and 14 or

2r = x(2t+l). Therefore, r = x(2t+l)/2. But if x is odd, x(2t+l)/2 has no

integer solution, hence a contradiction.

As an immediate consequence to this theorem, we have:

Corollary 2.1. Any latin square of order 4t+2 contai~ing a sublatin square of

order 2t+l is orthogonally mateless.

A very natural way of vlriting F(n;l) squares are the cyclic ones. For this

reason this family of F-squares has received a considerable amount of attention.

For example, it is known that if L is an F(n;l) square based on a cyclic permuta-

tion group of order n, then a) 1 is orthogonally rich if n is odd, b) 1 is

orthogonally mateless with respect to any decomposition of n as long as "1" is

in the decomposition (for instance see Hedayat and Federer [8]). While cyclic

F(n;l) squares are orthogonally mateless with respect to many decompositions

of n if ·n· is· even, they are at least of degree 2 with respect to the deconiposi-

tion n = 2+2+ ••• +2 as the following theorem shows.

 -14-

Theorem 2.3. If Lis an F(n;l) square and if.L is based on a cyclic permutation

group of order n (even) then L is at least of degree 2 with respect to the de-

composition n = 2+2+ ••• +2.

Proof. By construction. There is no loss of generality if we let G be the cyclic

.
per.mutatLon group generated by ~ 21 ••• n ;) o If we now consider the entries of
• • .n-
the cells on the main diagonal and the diagonals parallel to the main one we see

that the entries of each diagonal together with its complement are occupied by
the same elements. Considering each diagonal followed by its complement as an

entity we shall construct presently a square L' orthogonal to L with respect to

the decomposition n = 2+2+ ••• +2. Filling in its n diagonals parallel to the main

diagonal as follows: Take au ordered tuple of n/2 distinct elements say

0 = (a1 ,a2 , ••• ,an/ 2 ). Fill in then spaces of the main diagonal of L' repeating

each element of 0 twice in the prescribed order. Permute cyclically the elements

of 0 and fill in the diagonal starting with the second position in the first row

as before. Continue the process until all the n diagonals are completed.

The following example elucidates the content of the above procedure.

123456
612345
1 ,_561234
4 5 6 1 2 3
345612
234561

_.fl 2 3 4 5 6' An orthogonal mate for L with

is an F(6;1) square generated by\6 1 2 3 4 5) •
respect to the decomposition 6 = 2+2+2 on the set 0 = (A,B,C} is
 -15-

ABCABC
CAB CAB
L' =CA B CA B
BCABCA
BCABCA
ABCABC

Corollary 2.2. Let n be an even integer. If L is an F(n;l) square and if L is

based on a cyclic permutation group G of order n, then L is at least of degree

2 ~<Tith respect to any decomposition of n as long as every component of the de-

composition is even.

Proposition 2.2. Let L be an F(n;l) square based on a cyclic permutation group

G of order n on an n-set E. Then L contains an SFS(t;l) if and only if t divides

n.

Proof. Clearly the first column of L forms a group isomorphic toG provided that

we make each element of G correspond to the element of L into which it maps the

unity of L. Obviously the subgroup of L corresponding to a subgroup of G of

order t will fonn an SFS(t;l) within the rO'ws formed by the subgroup of G. The

same will apply to the cosets of the subgroup of L.

Remarks. If n = 4t+2, t a positive integer, then by proposition 2.2 any F(n;l)

square based on a cyclic permutation group of order n has an SFS(2t+l;l). Such

an F-square by theorem 2.2 is orthogonally mateless with respect to the de-

composition n = x+(n-x) as long as x is odd, however, it is of degree 2 for those

e decompositions of n considered in Corollar~r 2.2.

 -16-

The result of theorem 3.2 gives a temptation to conclude that a similar

result might hold for the family of F(n;l) squares, n = 4t+3, t a positive integer.

Namely, if L is an F(n;l) square, n = 4t+3, then L is orthogonally mateless with

respect to the decomposition n = (2t+l) + (2t+2) if L has an SFS(2t+l;l). We

discovered that this is not the case as the following example sho\'TS:

123 4567
2316745
L =) I
l 2 5 4 7 6
4 6 5 7 3 2 1
574,3612
6 4 1 I2 1 5 3
75611234

which is an F(7;1) square with an SFS(3;1). Note that Lis not based on a cyclic

group (see proposition 2.2). Indeed this square has an orthogonal mate with

respect to the finest decomposition of 7 viz., 7 =1+1+•• ~+1 1 and the following
F(;l) square is an example.

1234567
4173256
5612734
L' = 2 5 4 1 6 7 3
3467125
7356412
6725341

Note however, that L' is based on a cyclic permutation group generated by

G4l23456f\
1 7 3 2 5 6) •
 1 ..
-.L ( -

Definition 2.G. A directrix (t1·ansversa1) oi' an l~(n;A. 1 ,A.._, ••• ,A. ... ) square ::>n
... ~

at-set 1.: = (a1 ,a 2, ••• ,at} is a collection of n cells such t!:at the entries o:C

these cells .:!ontain A. 1 times a 1 , and every rm·r and column of r is represented

in this collection.

Fftample. i'he underlined cells in the followir,g F(5;2 1 1,2) square form a directrL'{.

! 5 l 4 5
5:-t5ll
l l ~ 5 .2
5 5 l .! 4
4 l 2 ~ l

~iJote that not every F-square has a directrix. For instan-:e, the follmdng

F(4;l) square has no directrix.

l 2
4 l ,;:: ---:::
-:; 4 .,
-~ l '~
2 3 4 1

·rhe follovring results can easily be verified.

Proposition 2.3. Let L be an F(n;l) square, n = 4t+3, with ?~ SFS(2t+l;l).

Then L has a direl!trix if the SFS(2t+l;l) does.

Proposition 2.4. There does not exist a pair of orthogonal l(n;l) squares,

n = 4t+3 1 having a pair of orthogonal SFS(2t+l;l).

 -18-

Concluding rema1·1cs. He lvould like to empl1asize tl.:,.nt concept;3 aldn to those

expressed in part ,, are fundamental with broad application. For example, to

determine whether or not a given F(n;l) square, viz., a latin square of order

n has an orthogonal mate, one can first, as a necessary condition, c:necl"' 'dhether

L has an orthogonal mete with respect to a coarser decomposition than n = l+l+ ••• +l.
Note that in general it is much easier to search for an orthoc;onal mate for L of

order n with respect to a coarses decomposition of n than i-Titb respect to the

finest one, viz., n = l+l+ ••• +lo

Acknowledgement

The first '1Uthor gratefully acknowledges the many helpful con:ments

and encouragement he received from Professor w. T. Federer prior to writing this

paper.

The second author is indebted to Professor ltl. T. Federer for providing

the facilitie~ for carrying out this joint research.

.... ' .

-19-

References

[l] Addelman, s. (1967). Equal and proportional frequency squares. J, Amer.

Statist. Assoc. 62:226-240.

[2] Federer, ~q. T. Lecture notes for Statist, 513-514, Cornell University,
Ithaca, New York.

[3] Finney, D. J. (1945). Same orthogonal properties of the 4 X 4 and 6 X 6

latin squares. Ann. Eugen. 12:213-219.
[41 Finney, D. J, (1946). Orthogonal partitions of the 5 X 5 latin squares.
Ann. Eugen. 13:1-3.

[5] Finney, D. J. (1946). Orthogonal partitions of the 6 X 6 latin squares.

Ann. Eugen. 13:184-196.

[6] Freeman, G. H. (1966). Some non-orthogonal partition of 4 X 4, 5 X 5,

and 6 X 6 latin squares. Ann. Math. Statist. 37:661-681.

[7] Hedayat, A. (1969). On the theory of the existence, non-existence, and

the construction of mutually orthogonal F-squares and latin squares.
Ph.D. thesis, presented to the Graduate School of Cornell University
for partial fulfillment of author's Ph.D. degree in statistics.

[8] Hedayat, A. and Federer, w. T. (1969). An application of group theory to

the existence and non-existence of orthogonal latin squares.
Biometrika, 56(3): to appear.

[9] Mann, H. B. (1944). On orthogonal latin squares, Amer. Math. Soc. Bull.
50:249-257.