You are on page 1of 13

The Annals of Human Genetics has an archive of material originally published in print format by the Annals of Eugenics (1925-1954).

This material is available in specialised libraries and archives. We believe there is a clear academic interest in making this historical material more widely available to a scholarly audience online. These articles have been made available online, by the Annals of Human Genetics, UCL and Blackwell Publishing Ltd strictly for historical and academic reasons. The work of eugenicists was often pervaded by prejudice against racial, ethnic and disabled groups. Publication of this material online is for scholarly research purposes is not an endorsement or promotion of the views expressed in any of these articles or eugenics in general. All articles are published in full, except where necessary to protect individual privacy. We welcome your comments about this archive and its online publication.

T H E COMPLETELY O R T H O G O N A L I Z E D L A T I N S Q U A R E
BY W. L. S T E V E N S
Galton Laboratory, University College, London, W.C.1
CONTENTS
1. 1.1. 1.2. 1.3. 1.4. 2. 2.1. 2.2. 2.3. 3. 3.1. 3.2. 4. 4.1. 5. 5.1. 6. 7.
8.
PACE

Introduction . . The Latin square . The Graeco-Latin square . Sets of orthogonal Latin squares . . General problem to be solved . . The completely orthogonalized prime-square . . Construction of a Latin square . . Orthogonality of any two squares of the complete set . . Example . Generalization to non-prime squares . . . Definition of a field . . Construction of a complete set of orthogonal squares . The nine-square . . Generation of the set by a circular permutation . . The existence of fields . . Generation of the set by a circular permutation . . Examples of complete sets to 125 (16, 25, 27, 32, 49, 64, 81, 125) Theorem in conclusion . Addendum . References .

82 82 82 83 83 84 84 84 85 85 85 86 87 88 88 89 90 93 93 93

1. INTRODUCTION 1.1. T h e Latin square A LATINS Q U A R E of side s is a square arrangement of s2 letters, s of each s kinds, such that in each row and in each column each letter occurs exactly once. Thus when s = 3, a suitable arrangement is

A B C

B C A

C A B

It is always possible to write down a Latin square of any side.


1.2. T h e Graeco-Latin square

It may or may not be possible to write down another Latin square with the same letters, such that when the two squares are superimposed, each letter of one square coincides exactly once with each letter of the other square. Two squares of side 3 with this property are A B C A B C

B C

C A

A B

C B

A C

B A

When the second square is written in Greek letters and superimposed on the first, the

W. L. S T E V E N S

83

composite square is (calleda Graeco-Latin square. I n general, any two such Latin squares are said to be orthogonal to each other.

1 3 Xets of orthogonal Latin squares ..


If it is possible to form a set of Latin squares, such that any two of them are mutually orthogonal, the set is known as a set of orthogonal squares. 1.31. Lemma. The number of squares in a set of orthogonal squares of side s is not greater than (s- 1). For the letters of each square may be permuted among themselves without destroying the property of orthogonality. Let this be done so that the top line of each square commences A B .. . . Then, since between any two squares B coincides with itself in the top row, it follows that in the first column B cannot occupy any position twice or the top position a t all. Hence there are not more than (s- 1) positions available for B. The same fact is obvious from consideration of the Analysis of Variance. For differences between the plots with the same letter in one Latin square, or between rows or columns, account for (s- 1) degrees of freedom, and since there are altogether s2 - 1 degrees of freedom, it follows that there are not more than s + 1 independent methods of subdivision. Of these rows and columns account for two, leaving only s - 1 different subdivisions by letters. 1-32.DeJinition. When a set of (s- 1) orthogonal squares exists, the square is said to be completely orthogonalized.
1.4. General problem to be solved

The problem, yet awaiting solution, is to determine the size of the largest set of orthogonal squares of side s. The following table summarizes the results to the time of the present paper :
S

Number in largest set Complete


I

Reference

I
I

3 4

i
7 8 9 Any prime

Complete
99
2, 9,

1%

Fisher & Yates (1934

Fisher (1937) Fisher & Yates (1938)

It was suggested by Euler that no Graeco-Latin square exists of side 4n+ 2. No proof has yet been found, but Fisher & Yates completed the demonstration in the case of 6. An examination of the above table suggests, and the present paper proves, the following theorem: If s=pn, where p is any prime, and n any number, there exists a set of s - 1 orthogonal squares of side s. The converse of the theorem was stated by Wernicke in 1910, but his attempted proof is now known to have been fallacious.
6-2

84

ORTHOGONALIZED L A T I N SQUARE

2. THECOMPLETELY ORTHOGONALIZED PRIME-SQUARE


The rules for forming a complete set of orthogonal Latin squares of side p =any prime have long been known. We shall, however, give a method of construction and a proof which prepare the way for the more general case. Consider the set of p numbers
0, 1, 2,

... (p-1).

Let these numbers be subjected t o the ordinary processes of addition and multiplication followed by reduction t o modulus p . For example, with p = 7, we have
3 x 6 = 18 = (2 x 7)

+ 4,

hence

3~6=4.

It is clear that many of the results of algebra are applicable to operations in this set of numbers. For example the equation
ax+b=c,
a#0,

where all the letters represent numbers in the set, has a unique solution for x. With p = 7
2~+3=6

is uniquely solved by

x=

5.

2-1. Construction of a Latin square

Consider the square of numbers


A = constant # 0 x , y = 0 , 1 , 2, ... ( p - 1 ) ,

where (Ax + y ) is the number to be placed in the xth row and yth column. The square so defined is Latin. For the numbers in row x and columns y and Y are the same if, and only if, Ax+y=Ax+ Y , which implies
y= Y .

Hence the same number does not occur twice in a row. Similarly the numbers in column y and rows x and X are the same if, and only if, Ax y =AX which implies

+y ,

A(x - X ) = 0,

x = X , since A # 0.
Hence the same number does not occur twice in a column.

2-2. Orthogonality of any two squares of the complete set


Consider the squares

{Ax + y}, { p y}, A, p = constants # 0, A # p.

W. L. S T E V E N S

85

Then if the number j of the first square coincides in position x y with the number k of the second square, we have Ax y =j,

+ p x +y =k ,

(A-p) x = j - k .

Since X # p, this equation uniquely determines x and thence y. Consequently each letter of one square coincides exactly once with each letter of the other square. Since X may take any of ( p- 1) possible values it follows that a complete set of orthogonal squares may be defined.

2-3.Example
Taking p = 5 , we find the first two squares to be
A= 1

A= 2
3 4 0 1 2 4 0 1 2 3 0 2 4 1 3 1 3 0 2 4 2 4 1 3 0 3 0 2 4 1 4 1 3 0 2

0 1 2 3 4

1 2 3 4 0

2 3 4 0 1

3. GENERALIZATION NON-PRIME SQUARES TO 3- DeJinitwn of a Jield 1.


I n generalizing the theory, we shall make the proof dependent on the existence of elements with certain properties, and later discuss the conditions under which such elements exist. Let us therefore suppose that there exist a set of s letters
uo, u1, u2, ..-,ui, ..., us-l having the following properties : 3.11. There is a law of addition of the letters, such that

ui ui u j + ui a letter in the set, + = =


u (uj i

+ + U k ) = (ui + u,. tuj)


thk

If ui, are in the set there is a unique letter ui

such that

ui u, = u j . +
These conditions imply the uniqueness of subtraction and the existence of a letter zero (denoted by 0 ) which, when added to any letter, leaves it unaltered. 3.12. There is a law of multiplication, such that uiuj=ujui=a letter in the set,
ui(upk)=(uiq u,,

u (Ui f u,) = u ui ui . i i f u,

86

ORTHOGONALIZED LATIN SQUARE

These conditions imply that the 0 previously discovered, when multiplied by any letter of the set, produces 0. Also suppose that if uiuj letters in the set and ui 0, then there is a unique letter are # uk such that UiUk uj = . These conditions imply the uniqueness of division and the existence of a letter unity (denoted by 1) which, when multiplied by any letter in the set, leaves it unaltered. A set of letters having the above properties is known as afield, and for the full discussion the reader is referred to Carmichael's book. 3.13. It is clear that fields exist. For example the set of numbers less than a prime is a field when treated in the manner of the preceding section. It is also apparent that many of the operations and results of ordinary algebra are applicable when the symbols are letters in a field. I n what follows we shall assume the applicability of algebraic processes without supplying the proofs, which are usually simple.

3-2. Construction of a complete set of orthogonal squares


Assuming that a field of s letters exists, we may write a square of letters as

where (uhu,+u,)is the letter in row uxand column u . The letters may have no obvious , , order, but we may designate the s rows and columns by means of the letters 0, 1, u ... us-1 in any way we please, naturally keeping the designation constant throughout the remainder of the discussion.

3.21. A square so defined is Latin.


The letters in row uxand columns u, and uy are the same if, and only if, which implies

u u ul/UXU, up, x, = u = uy. ,

Hence the same letter does not occur twice in a column. Similarly the letters in column u2/ rows uxand ux are the same if, and only if, and
UhU,

+u,= uhux+ u,,

Uh (u, - ux) 0, =

which implies

uz= ux since u # 0. h

Hence the same letter does not occur twice in a row.

3.22. Any two squares of the set are orthogonal.


Consider the two Latin squares
~"X%+U1/)) {U,UX+U,>

ux, = constants # 0, u # u p , up h

W. L. STEVENS

87

Then the letter uj of the f i s t square coincides, in row ux column uy,with the letter u k of the second square if, and only if, uxux4- ul/ j, =u
upuz uu=U k ,

(ux- up)ux ui U k . = -

This equation has ab unique solution (3.12) for ux and thence u y ,since u - up# 0. h It is clear that a complete set is generated, since uAmay be any of the ( 8 - 1) letters of the set, excluding zero. 3-23. Moreover, it may be observed that since uxux, where ux is constant, are the s letters of the field in some order, it follows that any square of the set is derived from any other square of the set by a permutation of the rows. This is a characteristic of all completely orthogonalized squares so far discovered.
4. THE NINE-SQUARE

We shall now exhibit a field of 9 symbols, and thus produce a completely orthogonalized square of side 32 = 9. The symbols will be the 9 ordered pairs of numbers
00
01 02 10

11

12

20

21

22

4.01 Addition (subtraction) is performed by adding (subtracting) corresponding members of the pairs separately, afterwards reducing to modulus 3. Thus
21 + 22 = 10, 21 - 22 = 02,
00 is the zero of the field.

4.02. Multiplication and division are performed by means of the multiplication table :
00 00 01 00 00 00 00 00 00 00 00 00

01
0 0

02
00

I0
0 0 I0

I1
00 I1

I2
00 I2 21

20
00

21 00 21 I2 I1 02 20 22 I0 01

22
00

02 10 I1 12 20 21 22

01 02 I0 I1 I2 20 21

22

02 01 20 22 21 I0 I2 I1
01

20 02 I2 22 01 I1 21

22 I2 20 01 21 02 I0

22 01 I0 I1
20

02

20 I0 01 21 I1 02 22 I2

22 I1 21 I0 02 I2 01 20

is the unity of the field.

It may be verified that this table satisfies the conditions (3.11, 3.12). For example it
is symmetrical about the leading diagonal, and hence multiplication is commutative. Any row contains all the symbols, and hence ax = b has a unique solution for x. Also the sum' or difference of two rows is identical with a third row, showing that multiplication is distributive. The associative property is illustrated by
10 x (11 x 22) = 10 x 10= 02, (10 x 11) x 22= 12 x 22= 02.

88

ORTHOGONALIZED LATIN SQUARE

The nine symbols therefore constitute a field, and eight orthogonal Latin squares may be formed by taking any column of the multiplication table as the leading column of a square, and adding in turn the symbols 01, 02, 10, ..., 22 to generate the remaining columns. 4.1. Generation of the set by a circular permutation We may permute the rows and columns of the multiplication table so that it reads like a diagonal square
00
01

I1 00 I1

20
00

21
00

02
0 0

22
00 22 I0 I2 01 I1 20 21 02

I0 __
00 I0

I2
00

00 01 I1 20 21 02 22 I0 I2

00 00 00 00 00 00 00 00 00

00

01 I1 20 21 02 22 I0 I2

20 21 02 22 I0 I2
01

20 21 02 22 I0 I2
01

21 02 22 I0 12
01 I1

02 22 I0 I2
01 I1

I2
01 I1

I1

20

20 21

I2 01 I1 20 21 02 22

20 21 02 22 I0

Here every column is obtained from the preceding column by the identical circular permutation on the letters other than 00, equivalent to moving every letter, other than 00, up one place. We shall identify the 9 pairs of numbers with the first nine letters of the alphabet
00 01 02
C

10

11

12

20

21

22

d
b c f h i a c a d i g b

e
d e h e f i f d g
c a e

f
g h b h i c

g
i g a

and write the Latin square generated by the first column of the above table
a b e g h c

a b b c f d

di e i : h
f d a b c a e c b

i g d e f d

h f e

c g z

a h g

b i h

The remaining seven orthogonal squares are now formed in turn, by keeping the top row fixed, displacing the second row to the bottom, and moving every other row up one. It may be noted that the solution is different from that given by Yates (in Fisher & Yates, 1938) when written with identical top lines ; i.e. it cannot be transformed into Yatess solution by any permutation of the letters, rows or columns throughout. It is, however, equivalent to Fishers 9-square in The Design of Experiments (1937).

5 . THE EXISTENCE

OF FIELDS

It has been shown (Carmichael, ch. IX) that a field of s letters exists if, and only if, s =pn, where p is prime and n any integer. This implies that complete orthogonalization is possible for a square of side pn.

W. L. S T E V E N S

89

The proof of the existence of the field pn is not given here, but a method for generating such a field will be indicated. Let B = bnxn+ bn-,Xn--l + ... + bo be a polynomial in which the coefficients are positive integers. It may or may not be possible to express B as the product of two factors which are multiplied together in the usual way, followed by reduction of the coefficients to modulus p , prime. For example, with p = 3, we may write x2+ 2 = ( x + 2) ( x + l), whereas x 2 + 1 cannot be so expressed. It can be shown that at least one polynomial of degree n without factors exists for any n and any prime p . The p n letters in 1,he field may now be formally identified with the pn polynomials
an-,x"-l

+ un-,xn-2 + ... + a,,

where a,-,, a,-,, ..., a, each take any of the values 0, 1, ..., ( p - 1). The law of addition of the letters is then identical with the addition of the corresponding polynomials, followed by reduction of the coefficients to modulus p . Multiplication of the letters is performed by multiplication of the corresponding polynomials followed by reduction to moduli B and p , where B is a polynomial without factors. For example the multiplication table for the 32 field of section 4.02 might have been formed by taking B = x 2 + 1. Then to find the product of 21 and 12 we write
( 2 ~ 1) ( x + 2 ) = 2 x 2 + 5 x + 2 + = 2 (x2+l)+5x
= 2x.

Hence 21 x 12=20. The fields formed by using different polynomials for B are the same, that is to say, one field may be formed from another by a permutation of the letters. With the laws of' addition and multiplication here defined, it can be shown that the letters have the pro.perties (3.11, 3-12)necessary for them to form a field. I n fact, with the exception of the uniqueness of division, all the required properties are obvious. Fields defined in the abov'e way are known as Galois fields.
5-1. Generation of the set by a circular permutation

It is clear that multiplication of the set of pn- 1 letters, excluding zero, by any one of the letters, is equivalent to a permutation of these letters. It can, however, be shown that
there exists a letter in the field, such that all the letters excluding zero can be expressed as powers of this letter. Consequently, since multiplication by this letter is equivalent to a permutation, :it must be equivalent to a circular permutation of the pn- 1 letters excluding zero.

90

ORTHOGONALIZED LATIN SQUARE

The rows of the complete set can therefore be arranged so that the squares of the set are in a sequence, and any square is obtained from the preceding one by keeping the top row fixed, taking the second row to the bottom, and moving every other row up one. This was done in the example of the 9-square, and the method wiU be used below in order to economize space.

6. EXAMPLES COMPLETE SETS OF

TO

125

We give a completely orthogonalized square of side s, for all s=pn< 125, n > 1. The case of s = 9 has been given in the text, and 4 and 8 will be found in Fisher & Yates (1938). The present writer also possesses solutions for s = 25 which, like the Yates 9-square, are not equivalent to the solution derived from a field, and such solutions are doubtless possible for all pn. Only the first square of each set is given, since the others may be derived by the method of section 5.1. Above 27, only the first column of the first square is given. To derive the first square express the numbers 0, 1 , ..., (s - 1) with a radix p and write them down in order, to form the top row. To find any other row add the leading letter from the first column to every letter of the top row, by the method of section 4.01. For example, the top row of the first 49 square runs
00 01 02 03 04 05 06 10
11

12

13

... ...

The third row commences with 24 and therefore runs


24 25 26 20 21 22 23 34 35 36 30

The remaining squares of the set are then formed as before.


(16=Z4)
a b b a c d d c
e f
r

i
m o p

j
n p o n

k
o m r

l
p n n

f e

x
n o

h
p

j
a e
p

i
b f
o

l
e g
n

k
d h
r

r n

n r

n n

o p

p o

i k
1

j l
k c
o

k i
j b
n

l j
i a
m

e a

f b

g e

h d

h
1

g k

f j

e i

d
p

g h

h g

e f

f e

c d
1

d c

a b

b a

k
g

j
f

i
e

d
n

c
m p

b
p r

a
o n

b
o n r n p o p

a
r

d
n

c
n

j
g b

i
h a

l
e d

k
f c

g d

h c

e b

f a

j c
h
n r

i d
g
n

l a
f
p

k b
o

c f

d e
k a

a h
j d

b g

k
L i
o

l
o

i
n

j
m

l g

k h

a g

b h

c e

d f

l b r

i c p

j
p

k
m

g l
n

W. L. S T E V E N S
(25 = ij2)
a b x s b c y h s c d z i h d e v k i e a w f k

91

s c

h i k i k f e a b m n o p 1 1 m n o p

s h d

1 m n o p m n o p 1 h i k f g r s t u p

q
r

r s
o

s t
p

t u
l

u q
rn

q y s

w x v w
d x s
h C

y x

z y
b v q

v z
c w r

v w s b a

w x y x y z t u q c d e b c d
k d y
l

z v r a e
h b w

o p 1

m a

t
n

u
o

q
p

r
l

s
rn g

h c u
n

i k f d e a q r s
o p l
o

g b t
m
p

z t
o

v u
p

w x q r
l
m

h , i
z

s q b u

k f w x u q

e a y z t u
i

c x h

f e z k
h t g w a

g a v

y r

e x

a y

b z

c v

d w

m n o

g h { e o

j
s s

m n

w
p

x
l

y
b f w x u v a
o

m n o

e a i k z v v w s t y z d e
m n

c d g h x y y z q r w x b c
p

r s t c d e q r s 1: f g h
c p 1

u a t j
d e w c h q x z
o

m n

w h z v p { t u

v g

x y i k w x q r
h

z y s
j

m n

m n o

e u y

c o a b q r z v

r
k
z

c d s t w x

v e q r
o

i k u q h j x y b c

t u q r q r s t k f g h r s t u

s u i q

a b c r , b c d : y z v l e a b i k f g r s t u y z v w v w x y p 1 m n iux y 2

e c 2 c

k f g f s h c d e j k f
o p l

3 k o b g h

p 1 h t

m n o m n o p p

u q
n

r s
p

s t
l

t u
m

k
l

m n

m n

r
e g 3 x k

s
a h k y f

t u y z
g
o

q r v w
i

s x

w d a u b

x y z v e a b c b c d e q r s t c d e o

b c d j k f f s h z v w g h i

1 e

p 1 m n

rnn o p

b
o

c
p

m n

d 1

(27 = 33)
a b s c a u ? ~ ; z i g h 1 j k v w x 2 ? y b c t d e f e f d v w x S t U u s c a b
o

g h y V t f

h a z W x d
p

i g ? X v w e
q

m n r

z ? u s

U X

u v

k 1 j n o a b c d e f g h i a b z g h c a r p q 1 j u s t x v d e f g h h i g b c
y z

o p q m q r

r p

g c b k w i a

h d f o ? a e

h i e f d e
m n

s t j
p
r

t u v w x y z u s w x v z 3 k 1 ? n o p q
q r j k l r p q 1 j k n

3 y
r

y z b c f d v a
o

? y z c a b

u s t f d e

o x v w i g h
k o l m

n o m n

m n o p q r j q r p k l j n

r p q 1 j x v w 3 y h i g b c u s t x v g h i a b c a b f d

k o m n z u s t a e f d w 3 y z c d e f e i g h

f
q

d
r

e
p

u s t i g h
k l j

x c
n

w b
m

j k l m n o p q r m n o p q r j k l n o m q r p k 1 j

p q r 2 k l r n n o 1 j k o rnn r p q s t u v w x y z ?

w
n

w x x v
o

? z y t

s u

t
1

u s
j g x t a v k y l

i o z 1 y u a d e

a
p

b
q

m n r

? j z s

y k ? t b c e f f d x v

t
o

f d e 1 j k w x v
p
q

s t u v W X x v w ? y z d e f g h a g h i a b c h i g b c a

m n r

e t
p
o

f d u s

h i g w x v
k l ~q

b
z

c
?

a y

q r j r n n r

m n o

w x v z q r p k f d e a g k 1 j n o m q r p d e f g h a a , b c

j k y t u s j n. o m h c a b
1

i w s c z x o m n r p q 1 j t u s w x v z ? m n o p q r j k

b y x e

m q r p k c a e f d h 2 ? s t u v v w ? u f d h y i b ? y t u s w

z
a i q b 3 e y

?
b g r c y f z

y
c h
p

t
d c
k

u
r

s
p

k 1 j g h i f d e
n o

n o

m q

e a

h x b v

m i g v w c a w x

a e z u d h ? s

f b l j f d s t i g t u

49 = 72
00000 00
01 I1

64= z6
000000
00000I 1 0 0 0 0 0 I 10000 II 0 0 1 0

81= 34
0000
0001 I000 II00 1 1 1 0 1111 2111 021 I I021 2 0 1 2 I 1 2 0
I121 2112 I211 2121 0212

000

0000I 00010 0 I00 0 01000 1 0 0 1 0


01001 11010 0 1 1 1 0 I 0 0 0 1 111 00

24 63
2 0

0002 2 0 0 0 2 0 2 0 2 2 2 0

00 I 200

IIIIOO
1 1 1 1 1 0 111111 0 1 1 1 1 1 101111 0 0 II 1 1

2222
I222 0122 2012 I201 2 2 1 0
2212 I221 2122 I212 0 2 1 1
I 1 0 2 0 0 1 1 I 1 0 0 10 11 21 I 0

220 222 I22


012
401 1 0 4 1 4 1

004 3 0 0 330

333 4 3 3
0 3 4 1 4 0 410 44I

26
I5

61
05

II01I
0 1 1 1 1 1 1 0 0 1 III01 0 0 I 0 1 001I 0 0 I00 1 I0000 I01 00 I001 I

55 36
21
3 0

4' 1
141

32

54
25
04

10101 I 0 0 0 1 1 1 1 1 1 0 0 0 IIOIOI 0 I010 1


001 I 0 1 I0 I0 01 II 0 I 01 0 I0 1 01 I01I 0 0 I0 I 11 0 1 1 1 1 0 1 0 1I I 1 0 I01II 0 1 0 1 1 1 1

31 4
131

'4 4 414 241 424


242 124
412

313

44
I2

35
I0

2 2I 0 0202 2020 2 0 2 2 I 2 2 0

43 1 I43
214 02 I
2 2 0

341
034 33 0 430

I20

I01II
11111

I3

0 1I I 0 01 I I 0 I0100

46
34
0 6

I122 0112 201 I 0 0 2 1 I 2 0 0


II 2 0 0 1I 0 0 1I 0 I0 01 2100
2210

221 I 0 2 2 1 I022 0102 2010


20 21 0220 0022 2002 I 0 2 0

II2 0 I 1 20I

443
044 304 1 0 3 113

420

II001 0 0 I 1 1
1 1 0 1 1 0 1 1 0 0 01010 11100

66
53 I4
50 51

OIIOII
1 1 0 0 1 1 010110 0 1 1 0 0 1 1 0 0 0 1 1 010010 001001 1 0 0 0 1 0 II0010 III 0 01 0 1I I00 0 1I I 0 0 0001I I I0001I 010001 101000

442 344 I34 413


04I 204 020 022 400 440

21 I 421 142
014 31 0 030 0 3 0 I 0 0 I10 I11

62

16
02 2 2

222I
0222 2 2 0 2

4 1

I202
0120

I120 I112 0 1 1 1 I01I 20 11


0210 002I I002 0100
0010

56
40

444 244
024 3 2 0 2 0 3

45
23
52
0 3

0012 2001 0200 0020

311 0 I 3 23 0 320

223
322 232
123 212 I21

332
23 3

33 65 42
60

IIOIOO III 1 0 0
1 1 0 1 1 1

OIIIIO

8
i T 3
+
E :
.d

64
31

4 3

OOIIII I 0 II 01 01001I
1 1 0 0 0 1 010100 001010 0 0 0 0 1 1 100010 II0001 0 11 0 0 0 001I 0 0 0001 I 0 0000I I I 0 0I 0 0 010000

312

23 I

323 432 343 434 243 324


132 0 3 1 I1 0

423
042 404 2 0 4

8 u

224
022 402 3 0 4

31 0 33 I
0 3 3 13 0 210

3 34 I33
213 321
032 4 3 0 040

22I
422

342 234
03 2

001000 000I 0 0 0000I 0

I02
010

W. L. S T E V E N S
We may summarize the results in the following theorem :

93

7 THEOREM . IN

CONCLUSION

There exist p n - 1 mutually orthogonal Latin squares of side pn, where p is any prime and n any integer. Moreover, it is possible to write the Latin squares in a sequence, so that any square is obtained from the preceding square by keeping the top row fixed, and taking the remaining rows through the same circular permutation. To this theorem we add an important note: 7-1.Completely orthogonalized squares exist which are not equivalent to the squares generated by the rnethod of the proof.

8. ADDENDUM The substance of the present paper was communicated to the Section A* meeting of the British Association in August 1938, and reported by R. A. Fisher in Nature, 3 September. The author has sinlce received from Raj Chandra Bose of Calcutta (1938) a proof of his paper to appear in SankhycZ, in which the problem of the completely orthogonalizedsquareis solved in exactly the same way. A comparison of the two papers is interesting in showing how two authors, working independently on the same problem, produce not only identical solutions, but even papers which show a striking similarity in form.
REFERENCES RAJCHANDRA BOSE1 938). On the applicationof the properties of Galoia fields to the problem of construction ( of hyper-Graeco Latin squares. Sanlchyci, 3, pt. 4, 323-38. R. D. CARMICHAEL (1937). Introduction t the Theory of Groups of Finite Order. Roston, U.S.A. and London: o Ginn and Co. R. A. FISHER (1937). The Design of Experiments, 2nd ed., 3 35. Edinburgh: Oliver and Boyd. -(1938). The mathematics of experimentation. Nature, Lond., 142, 442. R. A. FISHER F. Y ~ T E(1934). The 6 x 6 Latin squares. Proc. G a d . Phil. SOC. 492-507. & S 30, (1938). Statistical Tables. Edinburgh: Oliver and Boyd. P. WERNICKE (1910). Das Problem der 36 Offiziere. Jber. dtsch. MatVer. 19, 264-7.
~~

You might also like