See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/243088499

Free Groups from Fields

Article in Journal of the London Mathematical Society · April 1997

DOI: 10.1112/S002461079600436X

CITATIONS READS
8 198

2 authors, including:

Stephen D Cohen
University of Glasgow
170 PUBLICATIONS 2,526 CITATIONS

SEE PROFILE

All content following this page was uploaded by Stephen D Cohen on 23 July 2014.

The user has requested enhancement of the downloaded file.

 FREE GROUPS FROM FIELDS

STEPHEN D. COHEN  A. M. W. GLASS

This paper is dedicated to Brian Hartley, in memoriam

1. Introduction
In the nineteenth century, field theory brilliantly resolved a number of questions
that had taxed mathematicians for centuries ; for example, ‘ The circle cannot be
squared ’ by straight edge and compass, and solving polynomial equations by radicals
is not always possible. These successes have continued to be held up as superb
examples of the power of mathematical thought, and are demonstrated at an
undergraduate level. The purpose of this article is to provide another such natural
example which leads to a concrete realisation of the free group on 2 generators.
Let p be any odd prime number. Consider the permutations t : x * x­1 and
"
e ¯ e( p) : x * xp of the set of real numbers. In 1986, Samuel White [7] proved that if
a , …, an, k , …, kn are arbitrary integers (positive or negative) with a , …, an, k , …,
" " # "
kn− non-zero, then the map ta" ek" … tan ekn is not the identity map ; that is, there is a
" " "
real number ζ such that
((…(ζ­a )q"­a )q#­…­an− )qn−"­an)qn (1)
" # "
is not ζ, where qj ¯ pkj (for j ¯ 1, …, n). In other words, the subgroup Gp of the
symmetric group on 2 (the set of real numbers) generated by t and e( p) is free ; there
"
is no interrelation between t and e( p). Of course, t−m is the map x * x®m and e−m
" "
is the map x *p ox ¯ x"/p (the root in 2 being indeed unique).
m m

As a result of some of our recent work, we can now provide a more transparent
proof of White’s Theorem that requires only background well below that of the full
Galois correspondence. We hope that White’s Theorem is sufficiently intriguing and
the proof we give sufficiently elementary that it will be of interest to a large audience ;
we have made our explanation sufficiently leisurely to try to ensure this. It could also
serve as an introduction to the more elaborate applications of the basic method (see
[2, 3, 4]).

Caution. Although the proof uses ‘ elementary ’ ideas and is simpler than White’s
original one, it is not trivial and involves proving three facts simultaneously by
induction (each step requiring a key idea following initial preparations).

2. Pure extensions
Let q ¯ pm be a power of an odd prime p. Also, let E be a subfield of 2 and, for
a ` E, let α ¯qoa ¯ a"/q be the real qth root of a. ‘ Adjoin ’ α to E to form the pure
extension field F ¯ E(a"/q) ¯ ²3q− " ε αi : ε ` E ´. If, in fact, a is not the pth power of an
i=! i i

Received 3 November 1994 ; revised 2 October 1995.

1991 Mathematics Subject Classification 20E05.
J. London Math. Soc. (2) 55 (1997) 309–319
310  .   . . . 
element in E (that is, a"/p a E ), then the extension F of E is a vector space over E with
basis ²1, α, …, αq−"´ and so has degree q. In this case the set of fields intermediate
between E and F form a ‘ tower ’
E X F X F X … X Fm ¯ F,
" #
where Fi ¯ E(a" ) for i ¯ 1, …, m ; each is a pure extension of the previous one of
/pi
degree p.
Now, clearly, if b ` E is closely related to a by b ¯ atcq, where c ` E and p i t, then
E(b"/q) ¯ E(at/q) ¯ E(a"/q). Conversely, if a and b are elements of E such that
E(a"/q) ¯ E(b"/q), where a"/p a E, then abt ¯ cq for some positive integer t (indivisible
by p) and c ` E.
The above facts are all plausible but not exactly trivial to establish. Some can be
derived from results in textbooks (such as the recent one by S. Roman [6]) but we
present some non-standard parts as Lemmas A and B in an Appendix. Of course,
such facts are not restricted to real fields : one can easily consider a pure extension
F ¯ E(a"/q) of a subfield E of the complex numbers #, where a ` E and α ¯ a"/q has the
property that αq ¯ a, though, in general, α is not uniquely defined. In such a case, if
E contains a qth root of unity ω (1 1), we can define an automorphism τ of F (that
is, a mapping from F onto itself that is both an additive and a multiplicative
isomorphism) such that τ(α) ¯ ωα by τ(3q− " ε αi) ¯ 3q−" ε (ωα)i. Indeed τ is an
i=" i i=! i
E-automorphism of F : it fixes each member of E.

3. A transcendental chain
Suppose that p is a fixed odd prime and that we are given a non-empty word w
of Gp. When xw ¯ x­a everything is easy ; so we may suppose that the bijection w
is described by (1), where now each qj is a power (positive or negative) of p. We want
to show that w does not act like the identity on 2. To do this, let η be any real
transcendental number. We shall prove that ηw 1 η. Indeed, take the orbit S(η) of η
under Gp, that is, the subset S(η) ¯ ²ηw* : w* ` Gp´ of 2. We shall prove that ζw 1 ζ
for every ζ ` S(η).
Consider the field 1(η) of rational functions in η with rational coefficients. Let L
be its algebraic closure – so L comprises all roots of polynomials with coefficients in
1(η) : it is a subfield of #. Of course, L contains, as a subfield, the field K of all
algebraic numbers – these comprising all roots of polynomials with rational
coefficients. Then set L ¯ Lf2, the field of real numbers algebraic over 1(η), and
!
K ¯ Kf2, the field of all real algebraic numbers. (Of course, we can recover L and
!
K by adjoining i ¯ o(®1) to L and K , respectively ; see Lemma C of the Appendix.)
! !
Clearly, a property of L is that every member has a uniquely defined pth root (and,
!
for that matter, pmth root, for any m & 1) in L . Passing to L (which contains every
!
pmth root of unity), we obtain a field which contains all pmth roots of every member.
Now, for our specific word w and transcendental η, let ζ be any member of S(η).
We prepare to define a sequence of transcendentals (in S(η)) which begins with ζ and
ends with ζw. For this purpose, let us denote a power map x * xpm (for m & 1) by em
and a root map x * x"/pm (for m & 1) by rm. And, for any integer a, we denote the
translation x * x­a by ta. Then w can be expressed as a string of symbols
w ¯ Š … Šn, where the Ši alternate between a translation ta (with a 1 0) and either a
"
power em (for m & 1) or a root rm (for m & 1). (Here n is the length of w.) Then set
ζ ¯ ζ and ζj+ ¯ ζj Šj for j ¯ 1, …, n ; in particular, ζn+ ¯ ζw.
" " "
     311
We shall consider the sequence of fields K (ζ ), K (ζ , ζ ), …, K (ζ , ζn+ ), where, by
! " ! " # ! " "
K (ζ , ζj), we mean that both ζ and ζj have been adjoined to K , or, equivalently, that
! " " !
ζj has been adjoined to the simple transcendental extension K (ζ ) to yield an
! "
algebraic extension (a subfield of L ). Now, there is no immediate reason why this
!
sequence must form a tower
K (ζ ) X K (ζ , ζ ) X … X K (ζ , ζn+ ). (2)
! " ! " # ! " "
m
Indeed, if Šj ¯ em, a power, then ζj+ ¯ ζpj , so that K (ζ , ζj+ ) X K (ζ , ζj) ; and it is
" ! " " ! "
not at all obvious that these two fields are identical – but this is what the truth of (2)
would demand.
Suppose that we could prove that (2) holds. We claim that it would follow that
ζw 1 ζ. For, if there are no roots amongst Š , …, Šn, then (1) is merely a polynomial
"
over 1 and then ‘ w(ζ ) ¯ ζ ’ really is nonsense – remember, ζ is transcendental. So
suppose that the first root is Šj ¯ rm. Then ζj ` K (ζ ) and K (ζ , ζj+ ) ¯ K (ζ ) (ζ"j /pm)
! " ! " " ! "
and clearly ζ"j /p a K (ζ ). Thus K (ζ , ζj+ ) is a pure extension of K (ζ ) of degree pm and
! " ! " " ! "
so, certainly, strictly contains the latter field. Then (2) would imply that
K (ζ ) Z K (ζ , ζn+ ) (strictly). But, of course, if ζn+ ¯ ζ , then K (ζ , ζn+ ) ¯ K (ζ ),
! " ! " " " " ! " " ! "
contrary to the above, and our claim is justified.
We shall use standard field theory to prove (2). Unfortunately, we cannot do this
in isolation but only as part of a package of three results. Moreover, instead of (2)
itself, we shall build the corresponding tower with i adjoined to each field (that is, with
K replaced by K ). It will serve the same purpose.
!

4. A subchain
Instead of regarding w as a string of symbols Š … Šn, we shall gather these into
"
syllables s , …, sk each of which ends in a root rm . Call a word f a polynomial word
" i
if f ¯ ta or is given by (1) with each qi a positiŠe power of p : thus f(x) is, indeed, a
polynomial. Then we can write w ¯ s … sk, where sj ¯ fj rm , and fj is a polynomial
" j
word. Here, generally, (with reference to (1)) all the integers a in fj are non-zero but
f (x) may be x or have a ¯ 0. Similarly, there may be no root at the end of sk and,
" "
additionally, the final a may be zero : in these cases, by convention, set mk ¯ 0. Now
define a subchain ²µ , …, µk+ ´ of ²ζ , …, ζn+ ´ by µ ¯ ζ ¯ ζ, µj+ ¯ µj sj, for 1 % j % k.
" " " " " " "
Then µk+ ¯ ζn+ ¯ ζw.
" "
Note that, granted (2), we would have the following tower :
K ( µ ) X K ( µ , µ ) X … X K (µ , µk+ ).
! " ! " # ! " "
We saw already that the first containment must be strict : the second item in our
package is that all are strict, that is,
K ( µ ) Z K ( µ , µ ) Z … Z K (µ , µk+ ) (3)
! " ! " # ! " "
(with the exception of the last of these if mk ¯ 0).
For the third item of our package, observe that, since µpm" ¯ f ( µ ), then
# " "
K ( µ ) (( f ( µ ))"/p) is a pure extension of K ( µ ) of degree p contained in K ( µ , µ ) and
! " " " ! " ! " #
so in K ( µ , µk+ ) (assuming (3)). Can there be another pure extension F of K ( µ ) of
! " " ! "
degree p contained in K ( µ , µk+ ) ? The answer (remarkably, we would again suggest)
! " "
is ‘ No ’. And, for this one, it is important that we replace K by K and allow F to be
!
a pure extension of degree p of K( µ ) (other than K( µ ) (( f ( µ ))"/p)) contained in
" " " "
K( µ , µk+ ). The answer is still negative.
" "
312  .   . . . 

5. The results
The three items of our package will be labelled ‘ Hypothesis H ’ ; they represent a
hypothesis that will be proved inductively on the length n of the word w. As we have
hinted already, we adjoin i to all the fields and work with K instead of K . We denote
!
each field K( µ , µj) by Kj. The theorem is stated for a general transcendental ζ but in
"
the proof we need consider only those ζ in our special orbit S(η).

T 1 (Hypothesis H). Let w ¯ Š … Šn ¯ s … sk be a word in Gp and ζ a

" "
real transcendental number. Then
H : K(ζ ) X K(ζ , ζ ) X … X K(ζ , ζn+ ) ;
" " " # " "
H : K Z K Z … Z Kk+ (omitting the last of these if mk ¯ 0) ;
# " # "
H : if F is a pure extension of K of degree p contained in Kk+ , then F X K .
$ " " #
C 2. With m , …, mk as in Section 4, the degree of the extension Kk+ of
" "
K is pm, where m ¯ m ­m ­…­mk.
" " #
The reader will easily check that Corollary 2 follows from Theorem 1.

C 3 (White’s Theorem [7]). The group Gp is a free group of rank 2.

Denote by H(n) the truth of Hypothesis H for words of length not exceeding n.
It is straightforward to check H(1) and, for polynomial words, to check H(n). Hence
we can suppose that k & 2. We shall deduce H(n­1) from H(n) according to the
scheme :
H(n) 3 H (n­1) 3 H (n­1) 3 H (n­1).
" # $
Since H(n­1) has to be tested we shall suppose that w ¯ Š … Šn+ (with
" "
corresponding transcendental chain ²ζ , …, ζn+ ´) but retain k for the number of
" #
syllables : thus w ¯ s … sk (with k & 2).
"
Another point : if w begins or ends with a translation, induction will easily yield
H(n­1) from H(n) (since, trivially, K(ζ , ζn+ ) ¯ K(ζ ­a, ζn+ ) and K(ζ , ζn+ ­a ) ¯
" " " " " "
K(ζ , ζn+ )). Similarly, in the course of a proof, it sometimes simplifies the
" "
discussion to make certain small adjustments (such as replacing ζ by ζps) that can be
" "
justified by induction or some simple reason. Our intention is to explain all but the
simplest adjustments.
Because, however, the induction depends crucially on a lemma about word
polynomials, we break off to introduce it. The reader may simply wish to note Lemma
5, and then skip to Section 7 and the induction proof before returning to Section 6.

6. Polynomial lemmas
Let f be a polynomial word given by (1) with every q a positive power of p and
every a non-zero. Then f need not be square-free, that is, it may have roots (in K ) of
multiplicity exceeding 1. For example, if
f(x) ¯ ((x­a)p­b)p®bp with ab 1 0,
then x­a is a factor of multiplicity p : all other roots have multiplicity 1. More
generally, regard f as fF+ in a sequence of polynomial words f , …, fF+ built up from
" " "
     313
(1) (in an obvious way), with f (x) ¯ x­a . Differentiating, f « ¯ 0i= qi f qi i−", from
F
" " "
which it follows that any root of f of multiplicity exceeding 1 is also a root of one (or
more) of f , …, fF. In fact, if α ` K is a root, precisely, of fi , …, fi (with
" " s
i ! i ! … ! is % F) as well as of f, then one obtains qi … qi as the multiplicity of α as
" # " s
a root of f. All we need is the following consequence.

L 4. The multiplicity of a repeated root of a word polynomial is a power

of p.

We now prove the key lemma for Hypothesis H.

L 5. Let f be a word polynomial (with a … aF+ 1 0) and ω ( 1 1) be a pth

" "
root of unity (in K ). Then an identity of the form
f(x) f t(ωx) g(xp) ¯ hp(x), (4)
where t is an integer and g and h are polynomials (or eŠen rational functions) in K(x),
is impossible.

Proof. If (4) holds we can clearly suppose that 0 % t % p®1 (absorb the f p(ωx)
factors into hp(x)) and
f(x) f t(ωx) ¯ g(xp) hp(x), (5)
where g and h are polynomials.
Suppose that α is a root of f of multiplicity 1. Then any root of (x­a )q"®(α­a )q"
" "
is also a root of f of multiplicity 1. Thus, if f *(x) is the product of all factors x®α (for
α in K ) of multiplicity 1, then f *(x) ¯ f ((x­a )q") for some polynomial f .
! " !
Suppose first that t ¯ 0. Then, by Lemma 4, for some polynomial g ,
!
f *(x) ¯ f ((x­a )q") ¯ g (xp). (6)
! " !
But, if N ¯ deg f *, it is evident that the left side of (6) has a non-zero term in xN−" and
the right side does not. So we may assume that 1 % t % p®1.
In that case, again by Lemma 4, (5) must hold with f replaced by f *. Let α be a
root of f * and suppose, for each i ¯ 0, …, p®1, that ωiα has multiplicity mi as a root
of f *. Then each mi ¯ 0 or 1 and m ¯ 1. Moreover, ωiα has multiplicity mi− as a
! "
root of f(ωx) (or mp− when i ¯ 0). Since each ωiα has the same multiplicity as a root
"
of g(xp), we deduce from (5) that, modulo p, we have
m ­tmp− 3 m ­tm 3 … 3 mp− ­tmp− .
! " " ! " #
Adding these we conclude that either m ­…­mp− 3 0 (mod p) or t 3®1 (mod p)
! "
(that is, t ¯ p®1). But either of these imply that all the mi ¯ 1 and so f *(x) ¯ g (xp)
!
as at (6), which we already know to be impossible.

7. Proof of H (n­1)
"
We assume the discussion of Section 5 using the notation of that and earlier
sections. We assume H(n). One further convention, whose usefulness will be apparent
immediately, is to employ notation such as (ζi, ζj) for the subword Ši … Šj− of w whose
"
action sends ζi to ζj. We make a number of (apparently) insignificant initial
observations and adjustments.
314  .   . . . 
By H (n) (applied to (ζ , ζn+ ) and (ζ , ζn+ )), we have
" " " # #
K(ζ ) X K(ζ , ζ ) X … X K(ζ , ζn+ ), (7)
" " # " "
K(ζ ) X K(ζ , ζ ) X … X K(ζ , ζn+ ). (8)
# # $ # #
Suppose, however, that K(ζ , ζn+ ) does not contain K(ζ , ζn+ ). Then, obviously,
" # " "
Šn+ is a power and mk ¯ 0. Trivially, ζn+ ¯ ζn+ Šn+ ` K(ζn+ ) and hence K(ζ , ζn+ ) is
" # " " " " #
strictly contained in K(ζ , ζn+ ). Further, Š is a root rm because, otherwise, ζ ` K(ζ )
" " " # "
and the inconsistent conclusion K(ζ , ζn+ ) X K(ζ , ζn+ ) is a consequence of adjoining
" " " #
ζ to the final two fields in the tower (8). An adjustment (replacing ζ ¯ ζpm by ζps, where
" " # #
1 % s % m, and replacing s by m) allows us to suppose that K(ζ , ζn+ )fK(ζ ,ζ )
" # " #
¯ K(ζ ). Since ζn+ ` K(ζn+ ) and ζ ` K(ζ ), we deduce from (7) and (8) that
" # " " #
K(ζ , ζn+ ) ¯ K(ζ , ζn+ ) ¯ K(ζ , ζn+ ), this field strictly containing K(ζ , ζn+ ).
" " # # # " " #
In terms of syllables the above yields the following (for which we note that
µ ¯ ζ ¯ µq, q ¯ pm) :
" " #
K( µ , µk) ¯ K( µ , µk+ ) ¯ K( µ , µk),
" # " #
a field which strictly contains Kk+ ¯ K( µ , µk+ ). Moreover, Kk+ fK ¯ K and
" " " " # "
Kk ¯ Kk+ ( µ ) is a pure extension of Kk+ . Since µ ` K( µ ) and µ ` Kj (for j ¯ 2, …, k)
" # " " # #
by (7), it follows that Kj ¯ K( µ , µj) for 2 % j % k. Thus, by an adjustment, Kk can be
#
assumed to be of degree q over Kk+ .
"
The situation is represented in Figure 1.

Kk = K( l2, lk ) Kk = Kk
= Kk +1( l2 )
l2

deg q Kk –1 Kk –1
Kk –1
Kk + 1

Kk – 2
Kk –2
Kk –2
…

K3 = K( l2, l3 )
K3
K3

K2 = K( l2)

deg q K2 = K2

K1 = K( l1)
K1

Fig. 1 Fig. 2
     315
Suppose that k & 3. We see that Kk+ is an extension of K contained in Kk, but
" "
is not necessarily pure.
Since Kk ¯ Kk+ (µ ) is a pure extension of Kk+ of degree q and we have ensured
" # "
that Kk contains a pth root of unity ω (1 1) (this was the point of using K instead of
K ), then, as noted in Section 2, we can define a Kk+ -automorphism τ of Kk that sends
! "
µ * ωµ . (Of course, τ sends the underlying real field K ( µ , µk) onto a genuinely
# # ! "
complex field.) Set µa ¯ τ(µ ) ` Kk and let the second syllable s of w be frF. An
$ $ F F #
application of τ to the expression µp ¯ f( µ ) yields µa p ¯ f(ωµ ). Thus K ¯ K( µ , µ )
$ # $ # $ # $
and Ka ¯ K( µ , µa ) are both pure extensions of degree p of K ¯ K( µ ) contained in
F
# $ # #
Kk ¯ K( µ , µk). (In fact, with τ(µj) ¯ µa j and Ka j ¯ K( µ , µa j), we have a second tower
# #
K ¯ Ka Z Ka Z … Z Ka k ¯ Kk, which ends in Kk (see Figure 2) ; but we only need to
# # a $
consider K .)
$
If F ¯ 1, the crucial application of H (n) to the word ( µ , µk) (evidently of length
$ #
less than n) yields immediately that Ka ¯ K . If F " 1, a supplementary argument
$ $
yields the same conclusion. For example, if K fKa ¯ K , then K ( µp −") is a pure
F

$ $ # # $
extension of K( µ ) of degree p contained in K( µ , µk) but not in K , in contradiction
# # $
to H (n).
$
Now, from Section 2, K( µ ) ( µ ) ¯ K( µ ) ( µa ) implies that, for some t (indivisible
# $ # $
by p), µ µa t ` K(µ ). Raise this element to the pth power and replace the transcendental
$ $ # F
µ by x to yield an identity f(x) f t(ωx) ¯ hp (x). This is impossible by Lemma 5 (with
#
g constant). Thus k < 3.
Finally, if k ¯ 2 and w ¯ rm f, with f a polynomial word (necessarily of the form
F
f p , where Šn+ ¯ eF), we must have K ¯ K and so µ ¯ f( µ ) ` K ¯ K( µq). This
! " $ " $ # " #
quickly leads to f (x) ¯ g(xp) and a contradiction.
!

8. Proof of H (n­1)
#
We can now assume H (n­1) (in addition to H(n)) : it is pervasive and we do
"
not stop to point out every use of it. We shall use the induction hypothesis on the
inŠerse of a subword of w. But first we make some preliminary observations and
adjustments.
By H (n) we must prove that Kk Z Kk+ when mk & 1. For reductio ad absurdum,
# "
assume that Kk+ ¯ Kk. Since this necessitates that µk+ ` Kk, then, certainly,
" "
µpk+mk−" ` Kk : so replacing µk+ by µpk+mk−" we can assume that mk ¯ 1. If µk+ ` K(ζ , µk),
" " " " #
then K(ζ , µk) ¯ K(ζ , µk+ ), contradicting H (n) applied to (ζ , ζn+ ). Hence
# # " # # #
µk+ a K(ζ , µk) and, in particular, w must begin with a power Š ¯ eq.
" # "
Now, by assumption and H (n­1), we have that
"
K(ζ , µk) ( µk+ ) ¯ K(ζ , µk+ ) X K(ζ , µk+ ) ¯ K(ζ , µk) ¯ K(ζ , µk) (ζ ).
# " # " " " " # "
Actually, by an adjustment to ζ , we can assume that q ¯ 1 and that equality holds
"
in the above, that is, ζ a K(ζ , µk), yet K(ζ , µk+ ) ¯ K(ζ , µk) ¯ K( µ , µk+ ).
" # # " " " "
Write the end of the word w as

w ¯ … em g−" rF fr for m, F & 1, (9)

"
where f and g are both polynomial words (beginning with a genuine translation) and
g−" denotes the inverse of g. That is, if g is the polynomial word

tb eF tb eF … eF tb with b … bj+ 1 0,
" " # # j j+" " "
316  .   . . . 
then g−" ¯ t−b rF … rF t−b rF t−b ; of course g(x) may simply be a translation x­b
j+" j # # " "
(b 1 0), in which case g−"(x) ¯ x®b. In similar vein, let u ¯ (s … sk− )−" be the inverse
" "
in Gp of the bijection s … sk− . We could easily write down its symbol representation
" "
from that of w ; clearly it has the same length as s … sk− (which is less than n) and
" "
ends in r . In particular, µk u ¯ ζ , so u ¯ ( µk, ζ ). Let ²ν ¯ µk, ν , …´ be the syllable
" " " " #
transcendental chain of u.
Set F ¯ K( µk, µk+ ). Since µpk+ ¯ f( µk), F is a pure extension of K(ν ) (¯ K( µk)) of
" " "
degree p contained in K( µk, ζ ) but not K( µk, ν ). Apply H (n) to the word u with
" # $
respect to K(ν ) X F X K( µk, ζ ). Then F X K(ν , ν ). Moreover, unless u is a
" " " #
monosyllable, by H (n) applied to u, we have K(ν , ν ) X K(ν , ζ ) ¯ K( µk, ζ ), which
" " # " # #
yields the contradiction µk+ ` F X K(ζ , µk). Hence u is monosyllabic with ζp ` K( µk) ;
" # "
thus K( µk) (ζ ) ¯ K(µk) ( µk+ ) ¯ F. Therefore in (9), m ¯ 1, and the entire word w is
" "
e g−"rF fr . Further, for some t (indivisible by p) µk+ µt ` K( µk). Raising this to the pth
" " " "
power and again replacing the specific transcendental µk by the indeterminate x, we
F
get f(x) gt(xp ) ¯ hp(x), which contradicts Lemma 5 (with t ¯ 0 in (4)).
One bonus, now that we have established H (n­1), is that we can assume that
#
Corollary 2 holds for words of length not exceeding n­1.

9. Proof of H (n­1)
$
We may assume H (n­1), H (n­1), H (n) and Corollary 2.
" # $
Let F be a pure extension of degree p of K contained in Kk+ but not in K . By
" " #
H (n­1) and H (n), we can suppose that sk ends in a root, that is, mk & 1. Again by
" $
H (n) we can suppose that F \ Kk. Set F ¯ F( µk). Since both µ and µk are in F , then
$ " " "
Kk X F ; indeed, F is a pure extension of Kk of degree p contained in Kk+ (see Figure
" " " m
3). By Corollary 2 we know that Kk+ is a pure extension of Kk of degree p k and so
"
Kk +1 Kk +1
l
p
deg
lk deg p
F1 deg p mk F0 Kk

deg deg p l
p deg
p
Kk Kk – 1

Kk –1
…
…

F
F K2 K2
deg
p
K1 K1

Fig. 3 Fig. 4
     317
pmk−"
F ¯ Kk( µk+ ). By an adjustment (to µk+ ), we may suppose that mk ¯ 1 and
" " "
F ¯ Kk+ .
" "
Again write the final section of w as in (9). Put F ¯ F( µk− ), a subfield of F ¯ Kk+ .
! " " "
Since µ and µk− are in F , then F contains Kk− and, indeed, is an extension of it of
" " ! ! "
degree p. On the other hand, by Corollary 2, Kk+ is an extension of Kk− of degree
" "
pF+". Hence (see Figure 4), F ¯ F ( µk) is a pure extension of F of degree pF (that is,
F−" " ! !
µpk a F ). Hence there is an F -automorphism τ of Kk+ which maps µk * ωµk, where
! ! "
ω (1 1) is a pth root of unity. Set µa k+ ¯ τ( µk+ ) ` Kk+ . Then µa pk+ ¯ f(ωµk) ` K( µk). So
" " " "
Kk( µa k+ ) is a pure extension of Kk of degree p contained in Kk+ and therefore must
" "
be identical to Kk+ ¯ Kk( µk+ ).
" "
It follows that µk+ µa tk+ ` Kk for some t (indivisible by p). Further,
" "
K( µk, µk+ µa tk+ ) X K(µk, µ ),
" " "
yet (µk+ µa tk+ )p ¯ f( µk) f t(ωµk) ` K(µk). Thus K( µk, µk+ µa tk+ ) is a pure extension of
" " " "
K( µk) of degree p contained in Kk ¯ K( µk, µ ). As in Section 8, by applying H (n) to
" $
u ¯ (s … sk− )−" with syllable transcendental chain ²ν ¯ µk, ν , …´ we deduce that
"t " " # m−
µk+ µa k+ ` K(ν , ν ) ¯ K( µk) ( ν ). Hence, for some s (divisible by p " but not pm), we
" " " # #
have
µk+ µa tk+ νs ` K(µk).
" " #
Taking pth powers and (as usual) replacing µk by x, we obtain
F
f(x) f t(ωx) gs(xp ) ¯ hp(x),

for some polynomial h. This contradicts Lemma 5 in its most general form.
With this the induction is complete and Theorem 1 is established. White’s
Theorem follows.

10. Concluding remarks and further reading

The reader may wonder how many algebraic ζ satisfy the equation ζw ¯ ζ. for any
w. For e (¯ e( p)) it can be shown that, if w ¯ ta" ek" … tann ekn, with n & 1,
"
k , …, kn, a , …, an integers and k , …, kn− , a , …, an non-zero, then ²ζ ` 2 : ζw ¯ ζ ´
" " " " #
has size at most pk, where k ¯ rk r­…­rknr. This is exactly what we would expect :
"
if all of k , …, kn were positive, then w would be a polynomial of degree pk and so
"
would have at most pk real roots.
In fact, White’s Theorem (Corollary 3) holds for any (not necessarily prime) odd
integer p (positive or negative) or ‘ odd ’ rational number other than ³1 (see [1, 3]).
There is also a version over # for all positiŠe rationals other than 1 (see [2]). We
believe that the more intricate techniques of [2] will prove the result for all negative
rationals p (1®1) too, but we have not checked the details. White’s Theorem fails
for p ¯­1 (since e(1) is the identity) and also for p ¯®1 since e(®1)# and
(t e(®1) t−" e(®1) t e(®1))# are identity maps. See also [4] for an analogue in prime
" " "
characteristic and [5] for an informal introduction to the general question of the
concrete realisation of free groups and free products of groups.
Finally, we remark that, despite the above, it seems much harder to prove
Corollary 3 with p replaced by an irrational number (algebraic or transcendental) and
we have so far failed in that endeavour. It could prove an interesting challenge !
318  .   . . . 

Appendix
At the request of the referee we justify some of the preliminary non-standard facts
stated in Sections 2 and 3. Of course these can be extended to more general fields of
characteristic 0 and to degrees that are not prime powers.

L A. Let E be a subfield of 2 and p an odd prime. Also let m, n be integers

with 1 % m % n and a, b ` E such that a"/p a E and b"/p a E. Suppose that
E(a"/pm) X E(b"/pn). Then E(a"/pm) ¯ E(b"/pm) and abt ¯ cpm for some c ` E and integer
t indiŠisible by p.

Proof. It is a standard fact that [E(a"/pm) : E ] ¯ pm (under the above hypotheses) ;

this follows from [6, Theorem 13.1.5], for example. More generally, if A ` E and F is
F
the largest integer satisfying 0 % F % n such that A"/p ` E, then [E(A"/pn) : E ] ¯ pn−F.
We use these truths in what follows.
Assume that E(a"/pm) X E(b"/pn). We prove the lemma by induction on n. If
n ¯ m ¯ 1, then since E(a"/p) X E(b"/p) and both fields have degree p over E, it follows
that E(a"/p) ¯ E(b"/p). Hence F(a"/p) ¯ F(b"/p), where F ¯ E(ζ ) with ζ a primitive pth
root of unity. Hence abt ¯ cp, for some t coprime to p and c ` F by [6, Theorem
13.2.8]. Thus cp ` E ; so [E(c) : E ] divides p, yet [F : E ] is a divisor of p®1 [6, Theorem
10.2.8] that exceeds 1, since p is odd and ζ a E. It follows that c ` E and the result
holds.
Now suppose that n " 1. We first show that E(a"/pm) ¯ E(b"/pm) for which purpose
we can assume that n " m. By induction we may also suppose that
E(a"/pm) \ E(b"/pn−"). Put E ¯ E(B), where B ¯ b"/pn−m so that E (B"/pm) ¯ E(b"/pn).
" "
Then, evidently, E (a"/pm) X E (B"/pm). Let F (with 0 % F % m) be maximal such that
F " F " F
a"/p ` E , and A ¯ a"/p . Then E (A"/pm− ) X E (B"/pm), where A"/p, B"/p a E . By
" F " F " F "
induction, E (a"/pm) ¯ E (A"/pm− ) ¯ E (B"/pm− ) ¯ E(b"/pn− ). It follows that
F " " "
a"/pm ` E(b"/pn− ) and so, by assumption, F ¯ 0 ; that is, E (a"/pm) ¯ E (B"/pm). Thus, by
" "
induction, a"/pmBu/pm ¯ d ` E for some u coprime to p ; that is,
"
a"/pmbu/pn ¯ d ` E(b"/pn−m).
Now dpn ¯ apn−mbu ` E and so, since E(d ) : E ] divides [E(b"/pn−m) : E ] ¯ pn−m, then
e ¯ dpn−m ` E and E(e"/pn−m) X E(b"/pn−m), where e"/pn−m ¯ d. By induction,
e"/pn−m bv/pn−m ¯ c ` E for some Š (possibly p r Š). We conclude that
a"/pmbu/pnbv/pn−m ¯ c ` E, that is, a"/pmbt/pn ¯ c ` E,
where t ¯ u­pmŠ, so that pit (since p i u). It follows that
E(a " /pm
) ¯ E(bt/pn) ¯ E(b"/pn), since b ` E and 1 ¯ tx­pny for some integers x and y.
This is a contradiction since [E(a"/pm) : E ] ¯ pm and [E(b"/pn) : E ] ¯ pn. We have
therefore shown that E(a"/pm) ¯ E(b"/pm) for all m ` ²1, …, n´. Hence for all such m, we
have E(a"/pm−") ¯ E(b"/pm−"). Moreover, by induction we have, if m " 1, then
a"/pm−"bu/pm−" ¯ d ` E, for some u coprime to p. Hence,
d"/p ¯ a"/pmbu/pm ` E(a"/pm, b"/pm) ¯ E(b"/pm), that is, E(d"/p) X E(b"/pm).
By the first part we have E(d"/p) X E(b"/p) and d"/pbv/p ¯ c ` E, for some Š, where
possibly p r Š. Hence
a"/pmbt/pm ¯ c ` E, where t ¯ u­pm−"Š and p i t.
This completes the proof.
     319
L B. Let E be a subfield of 2 and pm be an odd prime power. Suppose that
a ` E where a"/p a E. Then a subfield F of E(a"/pm) containing E necessarily has the form
F
E(a"/p ) for some F with 0 % F % m.

Proof. Suppose that F is a field such that E X F X E(a"/pm) that is not of the
claimed form. Replacing E by an appropriate field of the form E(a"/pr), replacing a by
a"/pr and m by a suitable integer which is at most m, we can assume that a"/p a F and
F is still not of the claimed form. Note, however, that F(a"/pm) ¯ E(a"/pm), where
a"/p a F. It follows that [E(a"/pm) : F ] ¯ pm ¯ [E(a"/pm) : E ] which yields the contradiction
that F ¯ E.

Finally, the following simple result is quoted in Section 3.

L C. Let E be a subfield of 2 with algebraic closure A X #. Set A ¯ Af2.

!
Then A ¯ A (i).
!
Proof. Evidently A (i) X A. Suppose that α ¯ a­ib ` A, where a, b ` 2. Since A
!
is a field that is closed under complex conjugation (because f(α) ¯ 0 with f(x) ` E [x]
implies that f(αa ) ¯ 0), then a ¯ "(α­αa ) ` A and b ¯ (1}2i) (α®αa ) ` A . Thus α ` A (i).
# ! ! !

References
1. S. A. A, A. M. W. G and L. M, ‘ Arithmetic permutations ’, J. London Math. Soc.
(2) 43 (1991) 255–268.
2. S. D. C, ‘ The group of translations and positive rational powers is free ’, Quart. J. Math. Oxford
Ser. (2) 46 (1995) 21–93.
3. S. D. C and A. M. W. G, ‘ Composites of translations and odd rational powers act freely ’,
Bull. Austral. Math. Soc. 51 (1995) 73–81.
4. S. D. C and A. M. W. G, ‘ Free products in prime characteristic : a representation of
Fuchsian groups ’, J. Algebra, 179 (1996) 92–114.
5. A. M. W. G, ‘ The ubiquity of free groups ’, Math. Intelligencer 14 (1992) 54–57.
6. S. R, Field theory, Graduate Texts in Mathematics 158 (Springer, Heidelberg, 1995).
7. S. W, ‘ The group generated by x * x­1 and x * xp is free ’, J. Algebra 118 (1988) 408–422.

Department of Mathematics Department of Mathematics and Statistics

University of Glasgow Bowling Green State University
Glasgow G12 8QW Bowling Green
Scotland Ohio 43403-0221
USA

View publication stats