Professional Documents
Culture Documents
Ebook Around The Unit Circle Mahler Measure Integer Matrices and Roots of Unity 1St Edition James Mckee Online PDF All Chapter
Ebook Around The Unit Circle Mahler Measure Integer Matrices and Roots of Unity 1St Edition James Mckee Online PDF All Chapter
Ebook Around The Unit Circle Mahler Measure Integer Matrices and Roots of Unity 1St Edition James Mckee Online PDF All Chapter
https://ebookmeta.com/product/precalculus-concepts-through-
functions-a-unit-circle-approach-to-trigonometry-4th-edition-
michael-sullivan/
https://ebookmeta.com/product/the-star-around-the-
sun-%e6%98%8e%e6%97%a5%e6%98%9f%e7%a8%8b-pt-1-1st-edition-
diamond-circle-work/
https://ebookmeta.com/product/the-future-of-think-tanks-and-
policy-advice-around-the-world-1st-edition-james-mcgann/
https://ebookmeta.com/product/basic-analysis-iv-measure-theory-
and-integration-1st-edition-james-k-peterson/
Primary Mathematics 3A Hoerst
https://ebookmeta.com/product/primary-mathematics-3a-hoerst/
https://ebookmeta.com/product/beauty-and-sadness-
mahler-s-11-symphonies-1st-edition-david-vernon-marina-mahler/
https://ebookmeta.com/product/linear-algebra-and-matrices-1st-
edition-shmuel-friedland/
https://ebookmeta.com/product/the-eighth-mahler-and-the-world-
in-1910-1st-edition-stephen-johnson/
https://ebookmeta.com/product/the-unity-of-science-1st-edition-
david-bensimon/
Universitext
James McKee
Chris Smyth
Around
the Unit
Circle
Mahler Measure, Integer Matrices and
Roots of Unity
Universitext
Series Editors
Carles Casacuberta, Universitat de Barcelona, Barcelona, Spain
John Greenlees, University of Warwick, Coventry, UK
Angus MacIntyre, Queen Mary University of London, London, UK
Claude Sabbah, École Polytechnique, CNRS, Université Paris-Saclay, Palaiseau,
France
Endre Süli, University of Oxford, Oxford, UK
Wojbor A. Woyczyński, Case Western Reserve University, Cleveland, OH, USA
Universitext is a series of textbooks that presents material from a wide variety of
mathematical disciplines at master’s level and beyond. The books, often well
class-tested by their author, may have an informal, personal even experimental
approach to their subject matter. Some of the most successful and established books
in the series have evolved through several editions, always following the evolution
of teaching curricula, into very polished texts.
Thus as research topics trickle down into graduate-level teaching, first textbooks
written for new, cutting-edge courses may make their way into Universitext.
123
James McKee Chris Smyth
Egham, UK Edinburgh, UK
Mathematics Subject Classification: 11C08, 11C20, 11R06, 05A05, 05C20, 05C22, 11R18, 11S05,
15B36
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
You shall see them on a beautiful quarto
page, where a neat rivulet of text shall
meander through a meadow of margin.
Sheridan, School for Scandal.
Preface
vii
viii Preface
Salem graphs are a special kind of graph that can be associated with a Salem
number or quadratic Pisot number in a specific way. In Chap. 20, families of these
are constructed using certain rational functions called rational interlacing quotients
and circular interlacing quotients. Special cases of these constructions have an
application to the following chapter on minimal polynomials of integer symmetric
matrices.
Exercises in the book are distributed throughout the text, including the appen-
dices. They are of very variable difficulty, ranging from the almost trivial to the
challenging. An example of an easy one is the following.
Exercise 0.1 (Serre [Ser19, (A.7.5)]) Show that when the complex numbers w
and z are of modulus 1 then
If we feel that an exercise is even trickier, and perhaps that we do not know the
answer, it is marked as a Problem. Some exercises require light use of a computer,
and some require more substantial programming: these are flagged as computational
exercises. Unsolved problems are presented either as Research Problems or Open
Problems, the former generally being more open-ended and less notorious.
Most chapters end with a section of Notes, followed by a Glossary of terms used
in that chapter. The Notes include references to the sources of some of the results
and proofs that have been presented, as well as references to related work. For
results that have been stated in the text, but not proved, references are given beside
the result.
We end this preface with a list of some books and survey articles containing
material related to that in our book, which could be consulted. First of all, we highly
recommend the recent text on Mahler measure by Brunault and Zudilin [BZ20]. It
contains some advanced material that we have referred to only briefly: for instance,
a detailed account of Deninger’s work, and treatment of Mahler measure of elliptic
curves in much greater depth than we have described. Other texts containing a
significant to amount of Mahler measure material are those of Schinzel [Sch00] and
Bombieri and Gubler [BG06]. For results on real and complex polynomials in
general, the scholarly book of Rahman and Schmeisser [RS02] is a very valuable
source of results and references.
While we treat some aspects of the spectra of graphs, the topic is well covered in
the books by Godsil and Royle [GR01] and Brouwer and Haemers [BH12]. We
refer to Pisot and Salem numbers only briefly—they are equal to their Mahler
measures! For a more extensive treatment, see the book [BDGGH+92] by Bertin
et al. The classic text [Sal63] by Salem is also well worth reading.
In our treatment of roots of unity and cyclotomic integers in Chap. 5, the
considerable literature on vanishing sums of roots of unity is not covered. See for
instance Zannier’s survey paper [Zan95] for an account of this area.
The idea of producing a book of this sort has been with us for several years.
Right from the start, Rémi Lodh at Springer gave us encouragement to pursue the
dream, and was patiently supportive of our slow progress. We owe him a huge debt
Preface xi
for nudging us at the right moments and continuing to believe in the project. We are
also appreciative of the substantial help and encouragement provided by the
reviewers of the manuscript. We thank Igor Pritsker for detailed feedback on the
chapter on the transfinite diameter. To all these and to all our friends and colleagues
who have helped us through this work, thank you: we are glad it is now finished!
xiii
xiv Contents
1.1 Introduction
In this chapter, we introduce some basic results about the Mahler measure of poly-
nomials in one variable and some of its applications. In later chapters, there are
further results for one-variable Mahler measures: Dobrowolski’s Theorem (Chap. 3)
and Mahler measures for restricted sets of polynomials (Chaps. 11 and 14) and for
nonreciprocal polynomials (Chap. 12).
(z − α1 )(z − α2 ) · · · (z − αd ) , (1.2)
Exercise 1.1 Prove that the minimal polynomial of an algebraic number is unique
and has no multiple zeros.
Exercise 1.2 Prove that two conjugate sets either coincide or are disjoint and that
the field of algebraic numbers Q is a countable disjoint union of conjugate sets.
1.1 Introduction 3
|α1 · · · αd | = |ad | ≥ 1 .
This shows that z is the only such polynomial having all its zeros in the open disc
|z| < 1. Furthermore, if |ad | = 1 and all α j are in the closed unit disc |z| ≤ 1, then
they are actually all on the unit circle |z| = 1. For the closed disc |z| ≤ 1, moreover,
we have the first of Kronecker’s two theorems, to which we now turn.
Pα (z) = z d + a1 z d−1 + · · · + ad ,
say. Then, because (−1)k ak is the sum of all possible distinct k-tuples of the zeros
of P, all of which have modulus at most 1, we have |ak | ≤ dk (k = 0, . . . , d). The
same bounds apply to the coefficients of the polynomial Pr whose zeros are the r th
powers of those of P. Because by Proposition A.22 αr is also an algebraic integer,
the coefficients of Pr are integers, so there are only finitely many possibilities for the
coefficients of these polynomials. Hence, there are finitely many possibilities for all
the zeros of all the Pr , and so in particular, there are only finitely many possibilities
for the αr . Therefore, two of them must be equal, say αr = α s with r < s. Then
α s−r = 1, and we see that α is a root of unity.
Theorem 1.4 (Kronecker’s Second Theorem [Kro57]) Suppose that β is a real alge-
braic integer that lies, with its conjugates, in the interval [−2, 2]. Then β is conjugate
to 2 cos(2π/) for some positive integer .
Proof Let Q(x) be the minimal polynomial of β, of degree d say. Then the poly-
nomial z d Q(z + 1/z) has all its zeros on |z| = 1, so, by the First Theorem, these
are all roots of unity. Hence, β = ω + ω−1 for some primitive th root of unity
ω . By Proposition 5.2, ω is conjugate to e2πi/ , showing that β is conjugate to
2 cos(2π/).
Kronecker’s theorems suggest that we should try to describe all ‘small’ (suitably
defined) algebraic integers that are not covered by his theorems.
One way to do this is using the Mahler measure.
The Mahler measure M(P) of a polynomial P(z) = a0 z d + · · · + a0 ∈ Z[z] with
a0 = 0, and zeros α1 , …, αd ∈ C, is defined as
4 1 Mahler Measures of Polynomials in One Variable
d
M(P) := |a0 | max(1, |αi |) .
i=1
So M(P) measures, in a single value, how far outside the unit circle the zeros of
P are. Thus, when a0 = 1, the Mahler measure M(P) is the product of the moduli
of those zeros of P that lie in the region |z| > 1 (empty products being 1). Clearly
M(P) ≥ |a0 | ≥ 1. The definition extends immediately to any Laurent polynomial
(negative powers of z allowed). Such polynomials can be written as z k P(z), where
P is a regular polynomial as in (1.1), and k is any (possibly negative) integer. Then
M(z k P(z)) is defined simply as M(P).
The nth cyclotomic polynomial n (z) is defined as the minimal polynomial of
all primitive nth roots of unity. (From Proposition 5.2(b), we know that they all
have the same minimal polynomial.) These polynomials are discussed in Chap. 5.
Throughout the book, we use the term cyclotomic polynomial to refer to any monic
integer polynomial whose zeros are all roots of unity. Thus, such polynomials are
products of polynomials n (z) for a multiset of values of n.
The following is almost immediate.
Exercise 1.5 Show that, for a nonzero algebraic integer α, the mean modulus of its
conjugates is at least 1, with equality if and only if α is a root of unity.
The nth Chebyshev polynomial (of the first kind) Tn (x), though usually defined
so that all its zeros are in [−1, 1], is more naturally defined so that they lie in [−2, 2].
So for n = 1, 2, 3, …we define Tn (x) by Tn (z + z −1 ) := z n + z −n , and then Tn (x)
is monic with integer coefficients. Its zeros are clearly at 2 cos((2 j − 1)π/n) ( j =
1, . . . , n). The following formula gives a convenient way of computing Tn (x)—for
a discussion of the resultant, see Appendix A.
Exercise 1.7 The nth Chebyshev polynomial of the second kind Un (x) is defined
by Un (z + z −1 ) := (z n − z −n )/(z − z −1 ). Find the zeros of Un (x) and show that
L(z) := z 10 + z 9 − z 7 − z 6 − z 5 − z 4 − z 3 + z + 1 ,
answered. That the answer to this question is ‘yes’ is now often called ‘Lehmer’s
Conjecture’. That the answer is ‘yes’ and with c = M(L) is the Strong Lehmer Con-
jecture. If true, the Strong Lehmer Conjecture would be an extension of Kronecker’s
First Theorem, as it would show that any irreducible monic integer polynomial P(z)
with M(P) < M(L) must be either z or a cyclotomic polynomial.
The best unconditional result in the direction of a proof of Lehmer’s Conjecture
is due to Dobrowolski [Dob79], who proved that if M(P) > 1, then for any ε > 0
there is a d0 (ε) so that
3
log log d
M(P) > 1 + (1 − ε) (1.4)
log d
for all P of degree d ≥ d0 (ε). A little later, Cantor and Straus [CS83] proved a
version of this inequality with (1 − ε) replaced by 1/1200, but valid for all d ≥ 2.
Later, Voutier [Vou96], improved the constant 1/1200 to 1/4, again valid for d ≥ 2.
A version of Dobrowolski’s result is proved in Chap. 3, based on the method of Cantor
and Straus.
In any study of Lehmer’s question, we of course need to make use of the fact
that P is not cyclotomic, so that the zeros αi of Pare not roots of unity. Thus, we
d
know that, for all positive integers n, the product i=1 (αin − 1), being nonzero and
a symmetric function of the αi , is a nonzero integer. Using only this arithmetical
information, Blanksby and Montgomery [BM71] and later Stewart [Ste78b] used
this fact (and very different methods!) to prove for some positive constant c and
d ≥ 2 that
c
M(P) > 1 + .
d log d
In their proof, Blanksby and Montgomery used the Fourier analysis, while Stewart,
borrowing an idea from transcendence proofs, constructed an auxiliary function.
Dobrowolski, in the proof of (1.4), made use of the slightly deeper fact that, for all
d d p
j=1 (αi − α j ) is a nonzero integer multiple of p .
d
primes p, we have that i=1
In Voutier’s proof, he uses some additional arithmetic information concerning the
modulus of integer polynomial discriminants.
If P is not self-reciprocal (see Sect. A.1), then it is known from Theorem 12.1 that
M(P) ≥ M(z 3 − z − 1) = 1.3247 . . . , so, when addressing Lehmer’s question, we
can assume that P(z) is self-reciprocal.
There are several ways of coming up with the polynomial L(z). One way (see
Exercise 1.8 below for another way) is to note that it comes from the graph E 10 :
We take its adjacency matrix A = (ai j ) which has 1 in the (i, j)th place if vertices
labelled i and j are joined by an edge, and 0 otherwise. Then
6 1 Mahler Measures of Polynomials in One Variable
√ √
L(z) = z 5 det ( z + 1/ z)I − A .
d
P(z) = a0 z d + · · · + ad = a0 (z − αi ) ,
j=1
so that M(P) = ±a0 m 1 say, where m 1 = |α j |≥1 α j = α1 . . . αk say. Now let k p
be the number of α j that are in |z| p > 1 in Q p . Then, by considering ±ak p /a0
as a sum of all products of k p of the α j , we see that |α j | p >1 |α j | p = |ak p /a0 | p ,
giving |a0 | p |α j | p >1 |α j | p = |ak p | p . Hence, |a0 | p times the product of any number
of the |α j | p (for distinct α j ) is at most |ak p | p ≤ 1. In particular, |M(P)| p ≤ 1 and
|M(P)∗ | p ≤ 1 for all conjugates M(P)∗ of M(P). Hence, by Exercise A.19, M(P)
is an algebraic integer.
An algebraic integer β is called a Perron number if it is real and positive, and all
its conjugates except itself (if there are any) have modulus less than β.
Exercise 1.10 Show that the Mahler measure M(P) of an integer polynomial P is
a Perron number.
The Mahler measure M(P) has an integral representation via Jensen’s Theorem
(see Exercise 1.11 below) as
1
M(P) = exp log |P(e2πit )| dt . (1.5)
0
1
log |a| if |a| > 1;
log |e2πit − a| dt =
0 0 otherwise.
d
a02 ad2
M(P)2 + ≤ a 2j .
M(P)2 j=0
d
Proof Factorise P over C as P(z) = a0 i=1 (z + αi ). Then
P(z) = a0 (z + αi ) · (1 + αi z) · f (z) ,
|αi |≥1 |αi |<1
where
z + αi
f (z) := .
|α |<1
1 + αi z
i
Also, put b := |αi |<1 αi and B := |αi |≥1 αi , so that a0 b = ad /B. Then P(z) =
a0 b f (z)Q(z), where
Q(z) := (z + αi ) · (z + 1/αi )
|αi |≥1 |αi |<1
= z d + · · · + B/b
a0 B 2
= zd + · · · + .
ad
Hence,
d 1
2
a 2j = P(e2πit ) dt
j=0 0
1
ad 2
= f (e2πit )Q(e2πit ) dt
0 B
1
ad2 2
= Q(e2πit ) dt (as | f (z)| = 1 on |z| = 1)
B2
0
2
a2 a0 B 2
≥ d2 1 + 2
B ad
a02 ad2
= + M(P)2 ,
M(P)2
8 1 Mahler Measures of Polynomials in One Variable
Exercise 1.15 Modify Gonçalves’ Inequality and its proof to cover polynomials
with complex coefficients.
Lemma 1.16 (Gauss–Lucas) For any polynomial P(z) ∈ C[z], the zeros of its
derivative P (z) lie in the convex hull of the zeros of P.
d
Proof Writing P(z) = i=1 (z − αi ), we have the well-known partial fraction expan-
sion
d
P (z) 1
= . (1.6)
P(z) i=1
z − αi
d d
1 1
=0= ,
i=1
β − αi i=1
β − αi
giving
d
β − αi
= 0,
i=1
|β − αi |2
d
so that β = i=1 wi αi for weights
1 1
wi = ,
W |β − αi |2
where
d
1
W = .
j=1
|β − α j |2
Proposition 1.17 Let P(z) be a real monic polynomial of degree d, with Mahler
measure M. Let its ‘monic derivative’ d1 P (z) have Mahler measure M . Then M ≤
M.
Proof From (1.6), we have
1 d
1 1
log M − log M = log dt .
0 d i=1
e2πit − αi
Now the right-hand side, considered as a function of one αi , with the others fixed,
is a subharmonic function of αi in the open set C \ {z : |z| = 1} (see Appendix C).
Hence, its maximum is on its boundary |z| = 1. Doing this for each αi shows that
log M − log M has its maximum when all the αi are on |z| = 1. But then M = 1
and, by Lemma 1.16, M = 1 too. So M /M ≤ 1 for all polynomials P.
A related result is the following.
Proposition 1.18 If P(z) is a real polynomial of degree d and self-reciprocal with
Mahler measure M(P) > 1, then also M( d1 P ) is greater than 1.
This is an immediate consequence of Theorem A.8.
Exercise 1.19 For a polynomial P(z) = a0 z d + a1 z d−1 + · · · + ad ∈ C[z], show
that for j = 1, . . . , d we have
d
|a j | ≤ M(P) .
j
Exercise 1.20 Show that if the polynomial Q, with leading coefficient q0 , is a factor
of the polynomial P, with leading coefficient p0 , then M(Q) ≤ |q0 / p0 |M(P).
Theorem 1.21 Given a polynomial Q(z) ∈ Z[z] of degree e and an algebraic num-
ber α of degree d, then Q(α) = 0 or
max(1, |α|)e
|Q(α)| ≥ . (1.7)
L d−1 M(α)e
so that
|Q(α)|L d−1 M(α)e
1 ≤ p0e |Q(α)|L d−1 max(1, |α j |)e = ,
j>1
max(1, |α|)e
Corollary 1.22 For any rational number p/q with q > 0 and algebraic number
α = p/q of degree d, we have
p max(1, |α|)
α− ≥ .
q q(| p| + q)d−1 M(α)
In particular,
max(1, |α|)
|α − 1| ≥ .
2d−1 M(α)
B j ≤ N (Y − 1)M jN ( j = 1, . . . , d) , (1.8)
with M j := max(1, |α j |). Next, for j = 1, . . . , d divide the interval [−B j , B j ] into
L M jN intervals of equal length, and hence, divide the box dj=1 [−B j , B j ] ⊆ Rd
d
into smaller boxes. Since j=1 M j = M(P), this gives us at most L d M(P) N boxes,
together containing all the possible vectors v, with edge lengths at most
M jN 2N (Y − 1)
2N (Y − 1) < ( j = 1, . . . , d) . (1.9)
L M jN L −1
√
These lengths are less than 1/ 2 provided that
√
L ≥ 1 + 2 2N (Y − 1) . (1.10)
Now there are Y N possible vectors (y1 , . . . , y N ), so if Y N > L d M(P) N then, by the
Pigeonhole Principle, two vectors v1 , v2 , say, defined by distinct vectors (y1i ), (y2i )
say, must lie in the same box. Thus, it suffices to take
N /d
√ Y
1 + 2 2N (Y − 1) ≤ L < . (1.11)
M(P)
we certainly have
N i
must be 0. Thus, all α j are zeros of the nonzero polynomial i=1 yi z , which is an
integer polynomial multiple of P, given the α j are distinct.
Finally, we outline the modifications necessary to the proof when some of the α j
are equal. If α1 = α2 = · · · = αk = α say are zeros of P(z), then P(z) and its k th
Nk =i 1, 2, . . . , k − 1 are all zero at z = α. So we need to replace the
derivatives for
k copies of i=1 α yi appearing in v by
N N N N
i i−1 i i−2 i
α yi ,
i
α yi , α yi , . . . , α i−k+1 yi .
i=1 i=1
1 i=1
2 i=1
k − 1
Since k ≤ d, we can bound the modulus of each of these sums simply if rather crudely
by N d (Y − 1) max(1, |α|) N . The effect of this is that we need only replace the ‘N ’
appearing in (1.8), (1.9) (two places), (1.10) and the left-hand side of (1.11) by N d .
The argument is otherwise unaffected and adapts similarly if there are further sets of
equal zeros.
Exercise 1.24 The theorem tells us that for every n the polynomial (1 + z)n is a
factor of a polynomial with coefficients in {−1, 0, 1}. Find such a polynomial.
Exercise 1.25 Find an estimate for N in (1.11).
In this section, we give a lower bound for the minimum distance between conjugates
of an algebraic integer. We also give a lower bound for the modulus of the derivative
of the minimal polynomial of an algebraic integer, evaluated at its conjugates. Both
of these bounds involve the Mahler measure.
Let α be an algebraic integer with conjugates α = α1 , α2 , . . . , αd . Define the
separation of α, Sep(α), by
M(z) := M((z − z 1 ) · · · (z − z d )) .
2
0 1 1 ··· 1
1 z2 z3 ··· zd
z1 + z2 z 22 z 32 ··· z d2
| det(V )|2 = |z 1 − z 2 |2 z 12 + z 1 z 2 + z 22 z 23 z 33 ··· z d3
.. .. .
. . · · · ..
z 1d−2 + z 1d−3 z 2 + · · · + z 2d−2 z 2d−1 z 3d−1 · · ·
z dd−1
d−1 d d−1
≤ |z 1 − z 2 |2 |i z 1i−1 |2 |z j |2i by Lemma 3.2
i=1 j=2 i=0
d−1
d
≤ |z 1 − z 2 |2 i 2 max(1, |z 1 |)2(d−2) max(1, |z j |)2(d−1)
i=1 j=2
3
d d−1
≤ |z 1 − z 2 |2 d M(z)2(d−1) ,
3
d−1
using i=1 i 2 = 13 (d − 1)(d − 21 )d. This gives the required lower bound for |z 1 −
z 2 |.
We now apply this to the minimal polynomial of an algebraic integer α. For an
algebraic number α, we write M(α) for the Mahler measure M(Pα ), where Pα (z) ∈
Z[z] is the minimal polynomial of α.
Corollary 1.28 (Mahler [Mah64]) Let α be an algebraic integer of degree d. We
have √
Sep(α) > 3d −(d+2)/2 | (α)|1/2 M(α)−(d−1) .
⎩ z −(d−1)2 /2 if |z i | > 1.
14 1 Mahler Measures of Polynomials in One Variable
Here z := maxi=1,...,d |z i |.
d −(d−1)
min |Pα (αi )| ≥ 3(d−1)/2 d −(d +d−1)/2
| (α)|1/2 max M(α), α (d−1)/2
2
.
i=1
The length of a polynomial P(z) ∈ Z[z] is defined as usual to be the sum of the
absolute values of its coefficients. We define a short polynomial for P to be the
minimum-length product of P with a cyclotomic polynomial. We call this length the
shortness of P, denoted by sh(P). Note that a polynomial and its short polynomial
have the same Mahler measure. For an algebraic number α, we define the shortness
of α to be the shortness of its minimal polynomial. A nonreciprocal polynomial
clearly has shortness of at least 3, while a reciprocal noncyclotomic polynomial has
shortness of at least 5, by Exercise 1.34 below.
Exercise 1.31 Show that 0 has shortness 1, and all roots of unity have shortness 2.
Exercise 1.33 For an algebraic number α, show that its shortness is at least M(α)
(the smallest integer not less than M(α)).
1.7 The Shortness of a Polynomial 15
Exercise 1.34 Show that all nonconstant reciprocal noncyclotomic integer polyno-
mials have shortness at least 5.
Exercise 1.36 Show that the shortness of (z − 1)k is even, at most 2k , and is 6 for
k = 3.
Exercise 1.37 Let P be a cyclotomic polynomial and let k be the largest multiplicity
of any zero of P. Show that sh(P) ≤ sh((z − 1)k ).
Q(z)
= z 112 + z 111 + z 110 + z 109 − z 106 − 2z 105 − z 104 − z 103 + z 101 + z 100 + 2z 99 + 2z 98
+ z 97 − z 95 − 2z 94 − 2z 93 − 2z 92 − 2z 91 + z 89 + 2z 88 + 3z 87 + 2z 86 + 2z 85 + z 84 − z 83
− 2z 82 − 3z 81 − 3z 80 − 2z 79 − z 78 + 2z 76 + 3z 75 + 3z 74 + 3z 73 + z 72 − z 70 − 3z 69
− 3z 68 − 3z 67 − 2z 66 + z 64 + 2z 63 + 3z 62 + 3z 61 + 2z 60 − 2z 58 − 3z 57 − 3z 56 − 3z 55
− 2z 54 + 2z 52 + 3z 51 + 3z 50 + 2z 49 + z 48 − 2z 46 − 3z 45 − 3z 44 − 3z 43 − z 42 + z 40
+ 3z 39 + 3z 38 + 3z 37 + 2z 36 − z 34 − 2z 33 − 3z 32 − 3z 31 − 2z 30 − z 29 + z 28 + 2z 27
+ 2z 26 + 3z 25 + 2z 24 + z 23 − 2z 21 − 2z 20 − 2z 19 − 2z 18 − z 17 + z 15 + 2z 14 + 2z 13
+ z 12 + z 11 − z 9 − z 8 − 2z 7 − z 6 + z 3 + z 2 + z + 1
has length 179 and Mahler measure 1.24846635. However, its shortness is only 6,
with short polynomial
This is #170 in Table D.1 of Appendix D. The polynomial Q(z) is the noncy-
clotomic irreducible factor of P(z). In Chap. 5, we shall discover how to iden-
tify cyclotomic factors of polynomials, and in particular from Exercise 5.7, we
can see that, because P(z) has no factors that are polynomials in z 4 , we have
Q(z) := P(z)/ gcd(P(z), P(−z)P(z 2 )P(−z 2 )). For a discussion of a connection
between these short polynomials and related polynomials in 2 or 3 variables, see
Sect. 2.6.1.
There are 236 known integer polynomials with Mahler measures less than 1.25,
as shown in Table D.1. Computation shows that 15 of these are Mahler measures
16 1 Mahler Measures of Polynomials in One Variable
Lemma 1.39 If the polynomial P T has length and leading coefficient v, then
≥1+ α and v≤ . (1.12)
1+ α
To give a coefficient of modulus greater than 1 (other than the leading coefficient),
some di must be 0, and the signs of ±z n−d1 −d2 −···−di−1 and ±z n−d1 −d2 −···−di must be
the same, to disallow cancellation. Then
Now assume that P T satisfies (1.12). Then from the proof of Lemma 1.39, we
have
log(/v − 1)
d1 ∈ 1, 2, . . . .
log |α|
| ± α n−d1 −d2 ± α n−d1 −d2 −d3 · · · ± α n−d1 −d2 −···−d−v−1 ± 1| < ( − v − 1)|α|n−d1 −d2
< |vα n ± α n−d1 |
−v−1
if |α|d2 > α d1 ±1
. Hence,
log( − v − 1) − log |vα d1 ± 1|
d2 ∈ 0, 1, 2, . . . , ,
log |α|
where the range of values for d2 depends on the choice of the ± sign. Continuing in
this way, we have, for i = 2, , . . . , , − v, that
if
−v−i
|α|di+1 > .
|vα d1 +···+di ± α d2 +···+di ± · · · ± α di ± 1|
18 1 Mahler Measures of Polynomials in One Variable
Note that the term vα d1 +···+di ± α d2 +···+di ± · · · ± α di ± 1 has length i + v < so,
by assumption, cannot be zero. Thus, for i = 1, . . . , − v, each di is bounded in
terms of d1 , . . . , di−1 . If a polynomial of the form (1.13) is found that is 0 at z = α
then, because P is irreducible, the polynomial will be divisible by P, and so be of the
form P T . Then n is given by (1.14). Thus, the search is a finite one. It will eventually
succeed in finding P T of minimum length, as it will find P itself as a last resort.
This completes the algorithm description.
If T turns out to be cyclotomic, then we have found a short polynomial P T for
P.
Problem 1.41 Modify the above algorithm so that it works when P is nonmonic
and irreducible and has all its complex zeros in the disc |z| ≤ 1.
As before, for an algebraic number α, we write M(α) for the Mahler measure M(Pα ),
where Pα (z) ∈ Z[z] is the minimal polynomial of α. Three variants of Mahler mea-
sure are used in the literature:
• The Weil height
log M(α)
h(α) := log M(α) = . (1.15)
deg α
• The absolute Mahler measure M(α) := M(α)1/ deg α . Also called the exponential
Weil height.
• The logarithmic Mahler measure m(P) := log M(P) is often used for polyno-
mials P in several variables.
1.8 Variants of Mahler Measure 19
The Mahler measure results for algebraic numbers α have an alternative formulation
using the (absolute logarithmic) Weil height h(α) defined in (1.15). In particular, the
following are easily proved.
Before giving its proof, we need a preliminary lemma, with corollaries. Let us
extend the p-adic valuation on Q p to an algebraic closure Q p of Q p .
d
a0 z d + a1 z d−1 + · · · + ad−1 z + ad = a0 (z − α j ) , (1.16)
j=1
Corollary 1.45 Assume further that the coefficients ai in (1.16) are in Z, with a0 > 0,
and gcd(a0 , . . . , ad ) = 1. Then
log+ |α j | p = log a0 ,
p j
Proof (of Lemma 1.43) The proof is based on the obvious inequality
1 1 1
log+ |(αβ)k | ≤ log+ |αi | + log+ |β j | .
dαβ k
dα i
dβ j
Doing the same for each prime p, except embedding N into Q p instead of C, we
obtain
1 1 1
log+ |(αβ)k | p ≤ log+ |αi | p + log+ |β j | p .
dαβ k
dα i
dβ j
Then adding all these inequalities and applying Corollary 1.46 gives the result.
1.9 Notes 21
1.9 Notes
The first result in the direction of Lehmer’s Conjecture where, for a nonzero non-
cyclotomic integer polynomial P of degree d, M(P) − 1 was bounded below by a
function not exponentially small in d, was due to Blanksby and Montgomery [BM71].
They proved that M(P) ≥ 1 + 1/(52d log(6d)) for such P. Fourier analysis, in par-
ticular the nonnegativity of the Fejér kernel, played a large part in the proof. A little
later Stewart [Ste78b] proved a result of similar strength, but by a transcendence-type
argument, using an auxiliary function. The following result, which we state as an
exercise, formed part of Blanksby and Montgomery’s proof.
Exercise 1.47 ([BM71, Lemma 4]) For given real ρ satisfying 0 < ρ ≤ 1 and z ∈ C
with ρ ≤ |z| ≤ ρ −1 , show that
z
|z − 1| ≤ ρ −1 ρ −1 .
|z|
1.10 Glossary
The integer part, or floor function: for real x, x denotes the greatest integer
not exceeding x.
α . The house of α.
22 1 Mahler Measures of Polynomials in One Variable
and this integral version of the definition will be used later to define the Mahler
measure of polynomials in more than one variable.
minimum length, minimum-length multiple. Let P ∈ Z[z]. Among all integer
polynomial multiples of P, say P T where T ∈ Z[z], the smallest possible length
24 1 Mahler Measures of Polynomials in One Variable
is called the minimum length of P, and an example P T that achieves that length
is a minimum-length multiple of P.
Perron number. A Perron number is a real and positive algebraic integer β such
that |β | < β for any conjugate β = β.
Pisot number. A Pisot number is a real algebraic integer greater than 1, all of
whose conjugates except itself lie in the open disc |z| < 1.
separation. The separation of an algebraic integer α is the smallest modulus of
the difference between two of its conjugates.
short polynomial. Let P(z) ∈ Z[z]. A short polynomial for P is a polynomial of
minimum length of the shape P(z)Q(z), where Q(z) is a cyclotomic polynomial.
shortness. The shortness of a polynomial P(z) ∈ Z [z] is the length of a short
polynomial for P. The shortness of an algebraic integer α is the shortness of its
minimal polynomial.
Strong Lehmer Conjecture. A more precise version of Lehmer’s Conjecture,
with c = M(L(z)).
subharmonic. A real-valued continuous function of a complex variable, f (z), is
subharmonic on an open subset of the complex plane if for every closed disc
in that region the function value at the centre is bounded above by the average
1
value on the boundary of the disc: f (z) ≤ 0 f (z + r e2πit ) dt, where z is at the
centre of the disc and r is the radius. Extending f (z) to its boundary by continuity
(allowing −∞ as a value), the maximum of f (z) is attained on the boundary.
Weil height. The Weil height of an algebraic number is the logarithm of its absolute
Mahler measure.
Chapter 2
Mahler Measures of Polynomials
in Several Variables
2.1 Introduction
While the set of Mahler measures of polynomials with real coefficients clearly con-
sists of the whole of the positive real line, the situation for polynomials with integer
coefficients is far more interesting. Consider the sequence {M(z n − z − 1)}n≥2, where
1
log M(z n − z − 1) = log |e2πint − e2πit − 1| dt .
0
As n → ∞, the term e2πint becomes increasingly uncorrelated with e2πit , so that (as
we shall see)
1 1
lim log M(z n − z − 1) = log |e2πit1 − e2πit2 − 1| dt1 dt2 .
n→∞ 0 0
Thus, if we take this integral to be the definition of log M(z 1 − z 2 − 1), then we have
a convergent sequence of one-variable Mahler measures converging to a two-variable
Mahler measure. This suggests a more general definition of the Mahler measure. It
applies also to Laurent polynomials, so that negative exponents of the variables are
allowed.
Let k ≥ 1, zk = (z 1 , . . . , z k ) and F(zk ) be a nonzero Laurent polynomial with
integer coefficients. Then its Mahler measure M(F) is defined as
1 1
M(F) = exp ··· log |F(e 2πit1
,...,e 2πitk
)| dt1 · · · dtk . (2.1)
0 0
Conjecture 2.1 (Boyd’s Conjecture) The set L is a closed subset of the real line.
M(F(ρ1 z 1 , . . . , ρk z k )) = M(F(z 1 , . . . , z k )) .
The closure M1 (F) in R of this set can be described explicitly, as follows. Given
≥ 0 and an × k integer matrix A = (ai j ), define the k-tuple zA by
Camptopteris.
Hausmannia.
A critical and exhaustive account of this genus has been given by
Prof. Von Richter[966] based on an examination of specimens found in
the Lower Cretaceous rocks of Quedlinburg in Germany. The name
was proposed by Dunker[967] for leaves from the Wealden of
Germany characterised by a deeply dissected dichotomously
branched lamina. Andrae subsequently instituted the genus
Protorhipis[968] for suborbicular leaves with dichotomously branched
ribs from the Lias of Steierdorf. A similar but smaller type of leaf was
afterwards described by Zigno[969] from Jurassic beds of Italy as P.
asarifolius, and Nathorst[970] figured a closely allied form from Rhaetic
rocks of Sweden. While some authors regarded Hausmannia and
Protorhipis as ferns, others compared them with the leaves of Baiera
(Ginkgoales); Saporta suggested a dicotyledonous affinity for leaves
of the Protorhipis type. The true nature of the fossils was recognised
by Zeiller[971], who called attention to the very close resemblance in
habit and in soral characters to the recent genus Dipteris. A
comparison of the different species of Dipteris, including young
leaves (fig. 231, p. 297), with those of the fossil species reveals a
very striking agreement[972]. There can be no doubt, as Richter points
out, that the names Hausmannia and Protorhipis stand for one
generic type.
Hausmannia may be defined as follows:
Rhizome creeping, slender, dichotomously branched; leaf-stalks slender (2–
25 cm. long), bearing a leathery lamina (1–12 cm. long and broad), wedge-
shaped below, occasionally cordate or reniform, entire or more or less deeply
lobed into broad linear segments. The leaf is characterised by dichotomously
branched main ribs which arise from the summit of the rachis as two divergent
arms and radiate in a palmate manner, with repeated forking, through the
lamina. Lateral veins are given off at a wide angle, and, by subdivision, form a
fairly regular network similar to that in Dictyophyllum, Clathropteris, and
Dipteris.
Ptychocarpus.
This generic name, proposed by the late Professor Weiss[988], is
applied to a type of fructification illustrated by the plant which
Brongniart named Pecopteris unita, a species common in the Upper
Coal-Measures of England[989]. It is adopted by Kidston for fertile
specimens from Radstock which he describes as Ptychocarpus
oblongus[990], but the precise nature of the fertile pinnules of this
species cannot be determined.
Danaeites.
This generic name, instituted by Goeppert[993], has been used by
authors without due regard to the nature of the evidence of affinity to
Danaea. The type named by Stur Danaeites sarepontanus[994] (fig.
291, E) bears small pecopteroid pinnules with ovoid sporangia in
groups of 8–16 in two contiguous series on the lower face of the
lamina. The sporangia dehisce by an apical pore and are more or
less embedded in the mesophyll of the segments. No figures have
been published showing any detailed sporangial structure, and such
evidence as we have is insufficient to warrant the conclusion that the
resemblance to Danaea is more than an analogy.
Parapecopteris.
Asterotheca.
Certain species of Pecopteris fronds from Carboniferous strata are
characterised by circular sori or synangia consisting of a small
number (3–8) of exannulate sporangia attached to a central
receptacle and free only at their apices. Strasburger[997] suggested a
Marattiaceous affinity for Asterotheca and Stur[998] describes the
species Asterotheca Sternbergii Goepp. (fig. 291, C, D) as an
example of a Marattiaceous fern. The latter author retains Corda’s
genus Hawlea[999] for the fertile fronds of the common Coal-
Measures species Pecopteris Miltoni, while on the other hand
Kidston[1000] includes this type in Asterotheca.
Hawlea.
Stur[1002] retains this generic name for sori in which the sporangia
are free and united only by the proximal end to a central receptacle
(fig. 291, F, G). He describes the individual sporangia as possessing
a rudimentary annulus, a comparatively strong wall, and terminating
in a pointed distal end. He emphasises the greater degree of
cohesion between the sporangia of Asterotheca as the distinguishing
feature of that genus; but this is a character difficult to recognise in
some cases, and from the analogy of recent ferns one is disposed to
attach little importance to the greater or less extent to which
sporangia are united, at least in such cases as Asterotheca and
Hawlea when the cohesion is never complete.
Scolecopteris.
Zenker[1003] gave this name to detached fertile pinnules from the
Lower Permian of Saxony, which he described as Scolecopteris
elegans. He recognised the fern nature of the sori and suggested
that the pinnules might belong to the fronds of one of the
“Staarsteinen” (Psaronius), a view which subsequent investigations
render far from improbable. The sori, which occur in two rows on the
lower surface of the small pecopteroid segments with strongly
revolute margins (fig. 291, H–K), contain 4–5 sporangia attached to
a stalked receptacle comparable with that of Marattia Kaulfussii.
These pedicellate synangia were fully described by Strasburger[1004],
who decided in favour of a Marattiaceous alliance. The lower
portions of the distally tapered sporangia are concrescent, the distal
ends being free (fig. 291, H). Stur includes in Scolecopteris the
common species Pecopteris arborescens (fig. 376), but Kidston[1005]
states that the British example of Scolecopteris is S. polymorpha,
Brongn. from the Upper Coal-Measures.
Scolecopteris elegans Zenk. furnishes an example of a plant, or
plant fragment, which has been assigned to the animal kingdom.
Geinitz[1006] described silicified pinnules as Palaeojulus dyadicus, the
generic name being chosen because of the resemblance to
Millipedes such as the genus Julus. The mistake is not surprising to
anyone who has seen a block of siliceous rock from Chemnitz
crowded with the small pinnules with their concave surfaces formed
by the infolding of the edges. Sterzel[1007], who pointed out the
confusion between Myriapods and Filices, has published figures
which illustrate the deceptive resemblance of the pinnules, with their
curved lamina divided by lateral veins into segments, to the body of
a Millipede (fig. 291, K). He points out that Geinitz searched in vain
for the head and legs of Palaeojulus and expressed the hope that
further examination would lead to fresh discoveries: the examination
of sections revealed the presence of sporangia and demonstrated
the identity of Palaeojulus and Scolecopteris.
Discopteris.
Stur[1008] instituted this genus for fertile fronds from the Upper
Carboniferous Schatzlarer beds, including two species Discopteris
karwinensis and D. Schumanni. He described the small
Sphenopteroid pinnules as characterised by disc-shaped sori made
up of 70–100 sporangia attached to a hemispherical receptacle: the
absence of a true annulus led him to refer the genus to the
Marattiaceae. In his memoir on the coal-basin of Heraclea (Asia
Minor), Zeiller[1009] instituted the species Sphenopteris (Discopteris)
Rallii and figured sporangia resembling those described by Stur in
the possession of a rudimentary “apical annulus.” He compared the
sporangia with those of recent Osmundaceae and Marattiaceae. In
the later memoir on the Upper Carboniferous and Permian plants of
Blanzy and Creusot, Zeiller[1010] gives a very full and careful
description of fertile specimens of Sphenopteris (Discopteris)
cristata, a fern originally described by Brongniart as Pecopteris
cristata[1011]. Many of the Sphenopteroid pinnules of this
quadripinnate fern frond show the form and structure of the sori with
remarkable clearness in the admirable photographs reproduced in
Plates i.–iii. of Zeiller’s Blanzy memoir. The lobed pinnules of this
species are of oval-triangular form, 5–15 mm. long and 2·5–6 mm.
broad[1012]. An examination of the type-specimens of Discopteris from
Vienna enabled Zeiller to correct Stur’s original description of the
sori: he found that the Austrian and French specimens, though
specifically distinct, undoubtedly belong to one genus. The sori in
Discopteris cristata are globular, as in the recent genera Cyathea
and Alsophila, and frequently cover the whole face of the lamina.
The individual sporangia are 0·4–0·5 mm. long and 0·15–0·2 mm. in
diameter; they are exannulate, but for the annulus is substituted a
group of thicker-walled and larger cells in the apical and dorsal
region. The description by Stur of a hemispherical receptacle
seemed to indicate an important difference between the Austrian and
French species; but Zeiller found that this feature does not actually
exist and that it was so described as the result of misinterpretation.
Zeiller succeeded in isolating spores, 40–50 μ in diameter, from
some of the sporangia of D. cristata and found that they exhibited
the three-rayed pattern characteristic of fern-spores and which is
indicative of their formation in tetrads. The conclusion arrived at is
that the genus Discopteris, as represented by D. karwinensis, D.
cristata etc., may be regarded as a true fern and included in the
Marattiaceae. As Zeiller points out, the sori of Discopteris differ from
those of recent Marattiaceae in their pluriseriate construction and
agree in this respect with those of the Cyatheaceae. The comparison
already made[1013] between the sporangia of D. Rallii and those of
recent Osmundaceae holds good: the genus affords another
example of a generalised type, in this case probably a fern,
combining features which are now distributed among the
Marattiaceae, Osmundaceae and Cyatheaceae.
• • • • •
In addition to genera founded on true synangia or groups of free or
partially united sporangia, the literature of Palaeozoic ferns contains
several generic names applied to sporangia which occur singly on
Sphenopteroid or Pecopteroid pinnules. The following may serve as
examples; but it should be stated that these will probably be
transferred eventually to the Pteridosperms. It is, however,
immaterial whether they are dealt with here or in the chapter devoted
to the seed-bearing “ferns.”
Dactylotheca.
Zeiller[1014] created this genus for fertile fronds of Pecopteris
dentata Brongn. (= P. plumosa Artis[1015]), a common British species
in the Upper and Middle Coal-Measures. Stur[1016] included P. dentata
in his list of species of Senftenbergia, the genus to which reference
was made under the Schizaeaceae.
Pecopteris (Dactylotheca) plumosa (Artis). Figs. 290, E, 292, 293.
1825. Filicites plumosus, Artis, Antedil. Phyt. p. 17, Pl. xvii.
1828. Pecopteris plumosa, Brongniart, Hist. vég. foss. p. 348, Pls.
cxxi. cxxii.
— P. dentata, Brongniart, ibid. Pls. cxxiii. cxxiv.
— P. delicatulus, Brongniart, ibid. Pl. cxvi. fig. 6.
1832. Sphenopteris caudata, Lindley and Hutton, Foss. Flor. Vol. i.
Pl. xlviii.; Vol. ii. Pl. cxxxviii.
1834. Pecopteris serra, Lindley and Hutton, ibid. Vol. ii. Pl. cvii.
1834. Schizopteris adnascens, Lindley and Hutton, ibid. Vol. i. Pls.
c. ci.
1836. Aspidites caudatus, Goeppert, Syst. fil. foss. p. 363.
1838. Steffensia silesiaca, Presl, in Sternberg, Flor. Vorwelt, Vers.
ii. p. 122.
1869. Pecopteris silesiacus, Schimper, Trait. pal. vég. Vol. i. p. 517.
— Cyathocarpus dentatus, Weiss, Flora der jüngst. Stk. und
Roth. p. 86.
1877. Senftenbergia plumosa, Stur, Culm Flora, ii. p. 187 (293).
— S. dentata, ibid.
1886. Dactylotheca plumosa, Kidston, Cat. Palaeozoic Plants, p.
128.
1888. Dactylotheca dentata, Zeiller, Flor. Valenc. Pls. xxvi.–xxviii.
For a fuller synonymy reference should be made to Kidston’s
account of this species[1017], from which the above list is compiled.
The large fronds of this species are tri- or quadripinnate. The
pinnules vary much in shape and size and in degree of lobing,
according to their position on the frond (fig. 293). The primary pinnae
are subtended by two Aphlebiae (fig. 293, A) appressed to the
rachis, like the delicate leaves of the recent fern Teratophyllum
aculeatum (see page 301). The sporangia (0·5–0·65) are oval and
exannulate and are attached parallel to the lateral veins; they may
occupy the whole of the space between the midrib and the edge of
the pinnules. This species occurs in the Upper, Middle, and Lower
Coal-Measures of Britain, reaching its maximum in the Upper Coal-
Measures. The aphlebiae undoubtedly served to protect the young
fronds, as shown by a specimen figured by Kidston (fig. 293, B); they
may also have served other purposes, as suggested by the above
comparison with Teratophyllum, in the mature frond. Lindley and
Hutton regarded the aphlebiae as leaves of a fern climbing up the
rachis; which they named Schizopteris adnascens, a confusion
similar to that already mentioned in the description of Hemitelia
capensis (see p. 304).
Renaultia.
This name was proposed by Zeiller[1018] for Upper Carboniferous
fertile pinnae of the Sphenopteroid type, bearing ovoid sporangia
either singly or in marginal groups of 2 to 5 at the ends of the veins.
The appearance of the apical cells occasionally suggests the
presence of a rudimentary annulus. Kidston has recorded this type of
fructification in Britain[1019]. Stur describes fertile pinnules of the same
type under the generic name Hapalopteris[1020].
Zeilleria.