AN ITERATED QUADRATIC EXTENSION OF GF(2)

DOUG WIEDEMANN

University

of Waterloo,

Waterloo, Ontario, Canada

(Submitted November 1986)

1.

I t is well known
2 3

x'

+x

3J

A CONSTRUCTION

(see, for example, Ex. 3.96 of

[1]) that the polynomials

+ 1 are irreducible in GF(2)[x] for J = 0, 1, 2, . . . .

(x2'3'

+ x3J

+ l)(x3J

+ 1 ) = x3J++

Since

is a square-free polynomial, it follows that the period of each root of a;2*3

x

+ 1 is precisely 3J 1 , only one and a half times the degree of the polyno-

mial .

The field
Cj ~ GF(2)[x]/(x2'3J'

+ x3'

+ 1) ~

may be obtained by iterated cubic

where x0 1 is a cube
solution to x1 = xQ,

GF(22'3J)

extensions beginning with CQ ~

root of unity.

We have

Iterating, C--+1 CAxj

+ 1) 9

GF(2)(xQ)9

C1 ~ CQCX-L), where x x is any

where x- + 1 = x-.

This paper deals with an iterated quadratic extension of GF(2),

whose gen-

erators are described by

x

j +i

+ x

j+i

= x

for

J ^ 3 where arQ + x~l

= 1.

(1)

Let

Z?0 - GF(2)(* 0 ), ^ - ff0(*i>* .-- V i S V * j + i>Note that ic2 + ^ Q + 1 = 0 has no root in GF(2) so the first extension is quadratic.

To show that

each subsequent extension is quadratic, it need only be

shown that the equation for x.,1 , which may be rewritten x.

has no root in E -, for all j > 0.

Although this

+x

follows almost

x. + I = 0,
immediately

from theorems about finite fields, for example. Theorem 6.69 of Berlekamp [2],
a more elementary proof will be given here. Let

Tr(n\x)

x2\

i =l

Also, let \E\ denote the order or number of elements of a finite field E.
Theorem 1: For j > 0, xj+1
Tru

290

+Z

\x.

+1

i E^

\Ej + 1\

= 22

+2

and

) = 57r(j'+ 2)(a;:_J1) = 1 .

[Nov.

AN ITERATED QUADRATIC EXTENSION OF GF(2)

Proof (mathematical

induction):

Note xQ GF(2)

and Tr(1)(xQ)

= T P ( 1 ) ( ^ Q 1 ) = 1.

The statement of the theorem is therefore true for j = -1 if E_

be GF(2).

In a field of characteristic 2, assume x

x 4 = x 2 s 2 + 1 = xz3

+ s 2 + 1, x8

= xz7

+ z6

= xz
+ zh

+ 1.

is defined tc
Then,

+ 1,

and, in general,
2K

2K - 1

= XZ

k
x~^

2^
i =l

?K - ?v

Hence,
^r 2 +1

= xd

(Tp(J+ 1 ) (xT 1 )) 2 e

+ *?

+ 1x*

(2)

Now assume that the statement of the theorem holds for j - 1.

der 2

Then #. has or-

so, if x. x1 were in E J , by the Fermat theorem and (2), #_,, = #..., +

# (^p (J+D (^T1))2.

But Tv^

+ 1)

= 1 by

{x~^)

hypothesis, so, by

#. . -, is not in E- itself but in a quadratic extension of E .

2

is, therefore, |/77'| = 2

order 2 so, if O denotes the nontrivial

Finally, Tr '

+ 2)

is the trace map of

Tr(j' + 2 ) 0 r T M
by the last part

The order of 27-,.

, using the second statement of the hypothesis.

Note that the other root to (1) for x.,1

(j

contradiction,

,J. + 1

is tf7?\. Also, Gal(E -^JE.)

Galois automorphism, 0(x.

has
1

) = x"

to GF(2) , so

= T P ( J > 2 ) ( X .+ 1 ) = TP (J ' + 1 ) (X.+ 1 + a U > + . ) ) = Tr(j" + 1)(x.) = 1

of the hypothesis, completing

the statement of the theorem

for j .
9n

Corollary:
Proof:

xnn

= 1, when n ^ 0 and Fn = 2

Define E_1

to be GF(2) .

+ 1 is the Fermat number.

Since |27n| = 2
9

, the nontrivial member of

2n

Gal(En/En_1)
i s g i v e n by On(y) = 2/ . S i n c e t h e c o n j u g a t e of x n o v e r t h e
2? n _ 1 i s x , x ^ = # . T h u s , ^ n n = 1.
The order

of a field element

power which equals 1.

that xn

is defined to be the smallest

In the case where Fn

field

nonnegative

is prime, the above result implies

has order Fn . In any case, the order of xn

divides Fn ,

Since the Fer-

mat numbers are known to be mutually relatively prime, for example, see Theorem
16 of [3], the order
i < n*

We say

of xnxn_1

xQ

i < n, then xx
F

n?n-i

x^9

an element of a field is primitive if its order is the same as

the number of nonzero field elements.

ft

is the product of the orders of the

Yl

If the order of xi

is, in fact, F^ for

x is a primitive element of E , because

U

A.

= 22"+1 -

fL

1 = \En\

We h a v e n o t b e e n a b l e t o d e t e r m i n e i f x x

1.
.

w n- 1

1988]

291

AN ITERATED QUADRATIC EXTENSION OF GF(2)

2.

BASIS SETS

There are several natural ways to construct a basis of En as a vector space

over GF(2).

is a degree 2 n + 1 extension of GF(2).

# = GF(2)(xn)

lection of elements of the form xn

shown by induction on n.

where a9 b E _1.

xQ9

Another basis is the col-

where each 6^ {0, l}. This can be

= 1 and x1

Clearly, x

ratic extension of En_19

xQ ,

0 < i < 2 n+ 1 , because

One such is of course the set of powers x,

span E .

Since Z?n is a quad-

every member of Z?n is uniquely expressible as axn + b ,

Assuming a and 2? can be expressed as sums of the ^ " l ^

it follows easily that En

is spanned by

the xnn

,,#

;r00. It immediately

follows that these elements form a basis because the number of them is the same
as the dimension of the space spanned.
Another basis consists of elements of the f orm xnn - x^

where e . e {1}.

This is shown by a similar argument which

uses the fact that each element of

En equals axn

+ cx^1

Theorem 2:

+ b = axn

= (a + c)xn

x6n.--x6Qo
x 2n l

iii)

6^(0,1}

0 < i < 2n

b9

c e

En_1.

ii)

{ e {-1,1}

x^-'-xl*

+1

It has already been shown that i) and ii) each form a basis.

ments iii) are the conjugates of xn

are linearly
XQ = x

for some a,

The following are bases of En:

i)

Proof:

+ cxn_1

independent.

The ele-

over GF(2), and it will be shown that they

This will be done by induction.

Certainly, xQ

and

+ 1 are linearly independent over GF(2) . Assume that the conjugates of

22?1

x n_ 1

in En_1

linearly independent.

are

each conjugate

of xn

to its

The transformation On(y)

reciprocal.

If a combination

= y

takes

of the conjugates

vanishes, then grouping by reciprocal pairs gives

r W f + s.*-21) = o,

o)

i =0

where ai,

g^ e GF(2).

Applying On to both sides interchanges ou and 3^.

Add-

ing this to the original equation gives

<<** + MOrf +x'n2i) = t\^i

0= t

By the inductive hypothesis, ai

2n- 1

Wn-i-

=0

i=0

2
Z a--^
"~n - 1

+ 3^ E 0.

Thus, the sum (3) can be rewritten:

i=0

this time the hypothesis implies a292

E @- E 0.

Thus, iii) forms a basis.

[N ov.

AN ITERATED QUADRATIC EXTENSION OF GF(2)

In some sense the most interesting is the basis i) because the set for E
is contained in the set for En.

Therefore, the union of all bases given by i)

is a basis for the infinite field which

interesting property

is the union of all the En.

of the basis i) is

Another

that every boolean polynomial in n

variables corresponds to an

element of En.

These boolean polynomials can be

multiplied as elements of En

in a straightforward if tedious manner. To multi-

ply two such elements, collect all terms containing xn

(ax + b)(cx

+ d) = (aox^

to one side.

Then using

+ be + ad)x + (ac + bd)

the product is computable in terms of a few products in E

,.

Using this for-

mula, it can be seen, though the proof is omitted, that the "degree" of
product of the two elements does not exceed

the sum of their degrees.

the

By the

degree of a field element, we mean the degree of the associated boolean polynomial.
Each basis element of i) can be identified with the 0-1 vector, or bit vector, (6 n , ..., 6 0 ) which, in turn, can be identified with the integer
6n2n
Let bi

+ -.. + 6 0 2 .

be the basis element associated with the integer i. We now prove a fact

regarding the expansion of a product of two basis elements as the sum of basis
elements.
Theorem 3-

For any

i, j , and k the expansion of b^bj

if the expansion of b^b^

Lemma:

contains bk

if and only

contains bj .

For all i and j , b^b-

contains the basis element bQ = 1 if and only if

i = J.
Proof of the Lemma:

Once again, we use induction on n.

Obviously, the Lemma

holds whenever the two basis elements are in E_ . Assume it holds whenever the
two basis elements are in E
statement of the Lemma is
the product is in x E
r

bA

= x d,

x En_1

where

. Now, in En, if both bi

true.

and bn

c, d e E

n,

cannot occur in the expansion. If bL

then b-b-

Yl x

and only if o = ds

= xx

<s d

and does not contain bQ

are in En_1,

the

is a factor of one but not the other,

n -1

Yl

If xn

and bj

ilYl

^od + cd.

= x a and
n

The first term is in

By hypothesis, the second term contains b0

meaning i. = j .

if

This establishes the statement of the Lemma

for En in all cases, m

Proof of Theorem 3: Consider the coefficient of ^
it is the coefficient of bk

1988]

in b^bj.

in (bibj)bk.

By the Lemma,

Since

293

AN ITERATED QUADRATIC EXTENSION OF GF(2)

(bibj)bk

(bib^bj

it is also the coefficient of hi

Corollary 1:

in b^b^*

Let i 0 j be the mod 2 sum of i and j as bit vectors.

The coef-

ficient of b ~ infc.-&iis one.

Proof:

Let iDj9

i U j be the bitwise AND, bitwise OR of i and j , respectively.

It will be shown that the coefficient of bQ in

with the Lemma proves the Corollary.

and by the Lemma, this contains a b0

is one which, together

fy^jb^b-

Now, by rearranging terms,

in its expansion, m

The following corollary is an immediate consequence of the Lemma.

Corollary 2:

For any a e E

a2

contains bQ in its expansion if and only if a

is the sum of an odd number of basis elements.

3.
The minimal

polynomials over

Starting with p Q(y)

V (y)

V (y

MINIMAL POLYNOMIALS

= y
1

y' )-

+ y + 1,
It i

of the xn

GF(2)

are quite

let p (y)

= y p>Q(y

+ y' )

easy to compute.
and, in general,

clear that p (x ) = 0 for all n because

p fe+1 (^ +1 ) =^TX (x !<> = Since p

has degree

2 n + 1 , it is the minimal

result gives a method for computing the p

polynomial of xn.

The following

which is probably better suited to

calculation.
Theorem k: Let sequences of polynomials an(y)
a 0 = 1 + y29
Then an + bn
Proof: Let x

bQ = y and an

+1

= a2n + b2,

bn

is the minimal polynomial of

and bn(y)
+1

= anbn,

be defined as follows:
f o r n = 1, 2 , 3 , . . . .

xn.

= 1 and observe that, for n > 0, y = xn+1

is a root of a Q +

xnbQ

and, therefore, a root of

(a0

+ a;fc0)(a0 + x^b

0)

= a +

xn_1b1.

If ?i ) 1, dw = i n+1 is a root of
(ax + ^ n . A ) K + <-A>

= a

^n-2^2*

After repeating this n + 1 times, we see that y = xn

It follows from the definition that an has degree
294

+1
n+1

is a root of an

+ 1+bn

+ 1.

and that Z?n has degree

[Nov.

AN ITERATED QUADRATIC EXTENSION OF GF(2)

2n + 1 - 1. Thus, an + 2?n has degree

2n + 1 with x n as a root, so it must be the

minimal polynomial of x . m
k.

EXPERIMENT

The numbers FQ, F , F2, F 3 , F^ are prime so, by the Corollary to Theorem 1,
xn

has order Fn for n < 4.

In addition, using the complete factorizations [4,

5] of F n for 5 < n < 8, it has been checked on a computer that xn

1 for any

proper divisor & of Fn for n < 8. It would be desirable to know whether xn always has order Fn .

If this is true, then y

= oon_1 . . . xQ is primitive. It

would be useful to have a good way to compute the minimal polynomials of the
y

5.

A FIELD USED BY CONWAY

J. H. Conway has given an iterated quadratic extension [6, 7] of GF(2) that

comes from the theory of Nim-like games.

In our terminology,

this extension

would be defined by

c2n + cn = cn_ ...

c0 for n > 1 and c\ + oQ = 1.

It is well known that any two finite fields of the same order are isomorphic.
However, we do not yet know of an explicit isomorphism between GF(2) (xn) and

GF(2)(an).
ACKNOWLEDGMENT
The author would like to thank Norman Herzberg and Neal Zierler for their
assistance.
REFERENCES
1.

R. Lidl & H. Niederreiter. Introduction

to Finite Fields
cations.
Cambridge:; Cambridge University Press, 1986.

2.

E. R. Berlekamp.

3.

G. H. Hardy & E. M. Wright. The Theory

sity Press, 1971. Fourth Edition.

4.

R. P. Brent & J. M. Pollard. "Factorization of the Eighth Fermat Number."

Math. Comp. 36 (1981):627-630.

5.

J. C. Hallyburton, Jr., & J. Brillhart. "Two New Factors of Fermat Numbers." Math. Comp. 29 (1975):109-112; see also Corrigenda, Math. Comp. 30
(1976):198.

6.

J. H. Conway.

7.

Algebraic

On Numbers

Coding

Theory.

and Games.

J. H. Conway & N. J. A. Sloane.

and Their

Appli-

New York: McGraw-Hill, 1968.

of Numbers.

Oxford: Oxford Univer-

New York: Academic Press, 1976.

"Lexicographic

Codes from Game Theory." IEEE Transactions

Codes:

on Information

Error-Correcting

Theory

32 (May

1986).

1988]

oqr