Zeros of some self-reciprocal polynomials

David Joyner∗
2010-11-12

1 Introduction
This talk is about zeros of a certain family of polynomials which arise nat-
urally in several areas of mathematics. We are especially interested in poly-
nomials which have all their zeros in the unit circle

S 1 = {z ∈ C | |z| = 1}.

Let p be a polynomial

p(z) = a0 + a1 z + . . . + an z n ai ∈ C, (1)
and let p∗ denote1 the reciprocal polynomial or reverse polynomial

p∗ (z) = an + an−1 z + . . . + a0 z n = z n p(1/z).

We say p is self-reciprocal if p = p∗ , i.e., if its coefficients are “symmetric.”

The types of polynomials we will be most interested in this talk are self-
reciprocal polynomials. They arise naturally in the theory of algebraic curves
over finite fields and in the theory of error-correcting codes, but are of inde-
pendent interest.
∗
Talk given to Department of Mathematics, Gettyburg College, Nov 18, 2010.
[email protected]
1
Some authors, such as Chen [Chen], have p∗ denote the complex conjugate of the
reverse polynomial. It will not matter for us, since we will eventually assume that the
coefficients are real.

1
2 “Smoothness” of roots
A natural question to ask about zeros of polynomials is how “smoothly” do
they vary as a function of the coefficients of the polynomial?
To address this, suppose that the coefficients ai of the polynomial p are
functions of a real parameter t. Abusing notation slightly, identify p(z) =
p(t, z) with a function of two variables (t ∈ R, z ∈ C). Let r = r(t) denote a
root of this polynomial, regarded as a function of t:

p(t, r(t)) = 0.

Using the two-dimensional chain rule,

d
0= p(t, r(t)) = pt (t, r(t)) + r0 (t) · pz (t, r(t)),
dt
so r0 (t) = −pt (t, r(t))/pz (t, r(t)). Since pz (t, r(t)) = p0 (r), the denominator
of this expression for r0 (t) is zero if and only if r is a double root of p (i.e., a
root of multiplicity 2 or more).
In answer to the above question, we have proven the following result on
the “smoothness of roots.”

Lemma 1 r = r(t) is smooth (i.e., continuously differentiable) as a function

of t, provided t is restricted to an interval on which p(t, z) has no double
roots.

Example 2 Let

p(z) = 1 + (1 + t) · z + z 2 ,

so we may take
p
−1 − t + (1 + t)2 − 4
r(t) = .
2
Note that r(t) is smooth provided t lies in an interval which does not contain
1 or −3. We can directly verify the lemma holds in this case. Observe (for
later) that if −3 < t < 1 then |r(t)| = 1.

2
 Let p(z) = p(t, z) and r = r(t) be as before. Consider the distance
function

d(t) = |r(t)|

of the root r. Another natural question is: How smooth is the distance
function of a root as a function of the coefficients of the polynomial p?
The analog to Lemma 1 holds, with one extra condition.

Lemma 3 d(t) = |r(t)| is smooth (i.e., continuously differentiable) as a

function of t, provided t is restricted to an interval one which p(t, z) has
no double roots and r(t) 6= 0.

proof: This is basically an immediate consequence of the above lemma and

the chain rule,  
d 0 d|x|
|r(t)| = r (t) · |x=r(t) .
dt dx


Example 4 This is a continuation of the previous Example 2. Figure 1 is a

plot2 of d(t) in the range −5 < t < 3.

We will find this “smoothness” comforting later when we perturb certain

self-reciprocal polynomials.

3 Self-reciprocal polynomials
Let

R[z]m = {p ∈ R[z] | deg(p) ≤ m}

denote the real vector space of polynomials of degree m or less. Let

Rm = {p ∈ R[z]m | p = p∗ }

denote the subspace of self-reciprocal ones.

2
The plot was created using Sage’s list plot command, though the axes labels were
modified using GIMP for ease of reading.

3
Figure 1: Size of largest root of the polynomial 1 + (1 + t)z + z 2 , −5 < t < 3.

One of the simplest examples of a polynomial in Rm with all it zeros in

1
S is

cm (z) = 1 + z + . . . + z m .

If m is even then cm does not have ±1 as roots. Many results in the theory
fall into the following category.

Meta-theorem: If p ∈ Rm is “close” to cm then p has all its roots in the

unit circle S 1 .

For example, the example above satisfies this.

Here is a basic fact about even degree self-reciprocal polynomials. Let

p(z) = a0 + a1 z + . . . + a2n z 2n , ai ∈ R.

Lemma 5 ([DH], §2.1; see also [L2]) The polynomial p ∈ R[z]2n is self-
reciprocal if and only if it can be written

p(z) = z n · (an + an+1 · (z + z −1 ) + . . . + a2n · (z n + z −n )),

if and only if it can be written

4
 n
Y
p(z) = a2n · (1 − αk z + z 2 ), (2)
k=1

for some real αk ∈ R.

Example 6 Note

1 + z + z 2 + z 3 + z 4 = (1 + φ · z + z 2 )(1 + φ · z + z 2 ),
√ √
where φ = 1+2 5 = 1.618 . . . is the “golden ratio,” and φ = 1− 5
2
= −0.618...
is its “conjugate.”

The Chebyshev transformation T : R2n → R[x]n is defined on the subset3

of polynomials of degree 2n by
n
Y
Tp (x) = a2n (x − αk ),
k=1

where x = z + z −1 , and p and the αi ’s are as in (2).

The following statement is proven in Lakatos [L2].

Lemma 7 The Chebyshev transformation T : R2n → R[x]n is a vector space

isomorphism.

For any Xi ∈ C (1 ≤ i ≤ n), let

e0 (X1 , X2 , . . . , Xn ) =P
1,
e1 (X1 , X2 , . . . , Xn ) = 1≤j≤n Xj ,
P
e2 (X1 , X2 , . . . , Xn ) = 1≤j<k≤n Xj Xk ,
P
e3 (X1 , X2 , . . . , Xn ) = 1≤j<k<l≤n Xj Xk Xl ,
..
.
en (X1 , X2 , . . . , Xn ) = X1 X2 · · · Xn .

It is possible to describe explicitly how the αk ’s determine the aj ’s in (2).

The following result is proven in Losonczi [Los].
3
In this definition, we assume for simplicity a2n 6= 0 - see [L2] for the general definition
of p 7→ Tp .

5
Lemma 8 For each n ≥ 1 and αi ∈ C, we have
n
Y 2n
X
2
(z − αk z + 1) = c2n,k z k ,
k=1 k=1

where c2n,k = c2n,2n−k and

[k/2]  
k
X n − k + 2`
c2n,k = (−1) ek−2` (α1 , . . . , αn ),
`
`=1

for 0 ≤ k ≤ n.

Theorem 9 (Lakatos [L1]) Take the notation as in Lemma 5. The poly-

nomial p ∈ R2n has all its roots in S 1 if and only if −2 ≤ αk ≤ 2 for all
k.

Here’s another one of those “if it’s near cm then its roots are in S 1 ”-type
results.

Theorem 10 (Lakatos [L1]) The polynomial p ∈ Rm given by

m
X
p(z) = aj z j
j=0

has all its roots on S 1 , provided the coefficients satisfy the following condition
m
X
|am | ≥ |aj − am |.
j=0

Example 11 This is a continuation of the previous Example 2. In that

example, m = 2, p(z) = p(t, z) = c2 (z) + t · z. The theorem above says, in
this case, |t| ≤ 1 implies all roots of p(z) belong to S 1 .

4 Other characterizations
There are several other characterizations, due to Cohn, Chinen, Chen, and
Fell, which we discuss next.

6
Theorem 12 (Schur-Cohn4 ) Let p ∈ C[z]m be as in (1). Cohn showed that
p has all its zeros on S 1 if and only if
(a) there is a µ ∈ S 1 such that, for all k with 0 ≤ k ≤ n, we have an−k =
µ · ak , and
(b) all the zeros of p0 lie inside or on S 1 .

According to Chen [Chen], this result of Cohn, published in 1922, is

closely related5 to a result of Schur, published in 1918. The following result
is an immediate corollary of this theorem.

Corollary 13 p ∈ Rm has all its zeros on S 1 if and only if all the zeros of
p0 lie inside or on S 1 .

Remark 1 As a corollary to the corollary, by “version 2” of the Eneström-

Kakeya Theorem (see Remark 2 below), if p ∈ Rm is “near” cm then the
coefficients of p0 are increasing and positive, so all roots of p0 are in S 1 .

This result was proven by Chen [Chen] and later independently by Chi-
nen6 [Ch]. It provides a very large class of self-reciprocal polynomials having
roots on the unit circle.

Theorem 14 (Chen-Chinen) If p ∈ Rm has “decreasing symmetric form”

p(z) = a0 + a1 z + . . . + ak z k + ak z m−k + ak−1 z m−k+1 + . . . + a0 z m ,

with a0 > a1 > . . . > ak > 0 then all roots of p(z) lie on S 1 , provided m ≥ k.

We prove the following more general version of this.

Theorem 15 If g(z) = a0 + a1 z + . . . + ak z k and 0 < a0 < . . . < ak−1 < ak

then, for each r ≥ 0, the roots of z r g(z) + g ∗ (z) all lie on the unit circle.

proof: We shall adapt some ideas from Chinen [Ch] for our argument.
The proof requires recalling the following well-known theorem, discovered
independently by Eneström (in the late 1800’s) and Kakeya (in the early
1900’s).
4
See for example, Chen [Chen], §1.
5
In fact, both are exercises in Marden [Ma].
6
Actually, Chinen’s version is slightly stonger, and it is that version which we are
stating.

7
Theorem 16 (Eneström-Kakeya, version 1) Let f (z) = a0 +a1 z +. . .+ak z k
satisfy a0 > a1 > . . . > ak > 0. Then f (z) has no roots in |z| ≤ 1.

Remark 2 Replacing the polynomial by its reverse, here is “version 2” of

the Eneström-Kakeya Theorem: Let f (z) = a0 + a1 z + . . . + ak z k satisfy
0 < a0 < a1 < . . . < ak . Then f (z) has no roots in |z| ≥ 1.

Back to the proof. Write p(z) as p(z) = g(z) + h(z), where g is as

above and and h(z) = ak z m−k + ak−1 z m−k+1 + . . . + a0 z m . As usual, let
g ∗ (x) = xk g(1/x) denote the reciprocal polynomial. Note h(z) = z m g(z −1 ) =
z m−k g ∗ (z) and f ∗ (z) = f (z).
Claim: g ∗ (z) has no roots in |z| ≤ 1.
proof: This is equivalent to the statement of the Eneström-Kakeya Theorem
(Theorem 16). 
Claim: g(z) has no roots in |z| ≥ 1.
proof: This follows from the previous claim and the observation that the
roots of g(z) correspond to the inverse of the roots of g ∗ (z). 
Claim: |g(z)| < |g ∗ (z)| on |z| < 1.
proof: By the above claims, the function φ(z) = g(z)/g ∗ (z) is holomorphic on
|z| ≤ 1. Since g(z −1 ) = g(z) on |z| = 1, we have |g(z)| = |g ∗ (z)| on |z| = 1.
The claim follows from the maximum modulus principle. 
Claim: The roots of z r g(z) + g ∗ (z) all lie on the unit circle, r ≥ 0.
proof: By the previous claim, z r g(z) + g ∗ (z) has the same number of zeros as
r ∗ (z) z r g(z)
g ∗ (z) in the unit disc |z| < 1 (indeed, the function z g(z)+g
∗
g (z)
= 1 + g ∗ (z)
has
∗ r ∗
no zeros). Since g (z) has no roots in |z| < 1, neither does z g(z) + g (z).
But since z r g(z) + g ∗ (z) is self-reciprocal (in this case), it has no zeros in
|z| > 1 either. 
This proves Theorem 15. 
If P0 (z) and P1 (z) are polynomials, let

Pa (z) = (1 − a)P0 (z) + aP1 (z),

for 0 ≤ a ≤ 1. Next, we recall an interesting characterization of polynomials

(not necessarily self-reciprocal ones) with roots on S 1 , due to Fell [Fe].

Theorem 17 (Fell) Let P0 (z) and P1 (z) be real monic polynomials of degree
n having zeros in S 1 − {1, −1}. Denote the zeros of P0 (z) by w1 , w2 , . . . , wn
and of P1 (z) by z1 , z2 , . . . , zn . Assume

8
 wi 6= zj ,

for 1 ≤ i, j ≤ n. Assume also that

0 < arg(wi ) ≤ arg(wj ) < 2π,

0 < arg(zi ) ≤ arg(zj ) < 2π,

for 1 ≤ i, j ≤ n. Let Ai be the smaller open arc of S 1 bounded by wi and zi ,
for 1 ≤ i ≤ n. Then the locus of Pa (z), 0 ≤ a ≤ 1, is contained in S 1 if and
only if the arcs Ai are all disjoint.

This theorem is used in the “heuristic argument” given at the end of the
paper.

5 The special case of Littlewood polynomials

This section discusses a very interesting class of polynomials named after the
late British mathematician J. E. Littlewood, famous for his collaboration
with G. Hardy in the early 1900’s. Although the main questions about these
polynomials do not involve their zeros, so this section is a bit tangential,
there are some aspects related to our main theme. Basically, we present just
enough to whet the readers’ taste to perhaps pursue the literature further on
their own.
This section recalls some relavant facts from Mercer’s thesis [M].
A polynomial p(z) as in (1) is a Littlewood polynomial if ai ∈ {±1}, for all
i, where ai = ai (p) is the i-th coefficient of the polynomial p. Let Ln denote
the set of all Littlewood polynomials of degree n.

Conjecture 18 (Littlewood’s “two-sided” conjecture) There are positive con-

stants K1 , K2 such that, for all n > 1, there exists a p ∈ Ln such that
√ √
K1 n ≤ |p(z)| ≤ K2 n. (3)

The autocorrelations of the sequence {ai }ni=0 are the elements of the se-
quence given by

9
 n−k
X
ck = ck (p) = aj aj+k , 0 ≤ k ≤ n. (4)
j=0

One can show that

p(z)p∗ (z) = cn + cn−1 z + . . . + c1 z n−1 + c0 z n + c1 z n+1 + . . . + cn z 2n .
Littlewood polynomials are studied in an attempt to gain further understand-
ing of pseudo-random sequences of ±1’s. In this connection, one is especially
interested in Littlewood polynomials with “small” autocorrelations. A Lit-
tlewood polynomial having the property that |ck | ≤ 1 is called a Barker
polynomial. It is an open problem to find a Barker polynomial for n > 13
(or show one does not exist). Let

b(n) = minp∈Ln max1≤k≤n |ck (p)|,

where ck is as in (4). If b(n) > 1 then there is no Barker polynomial of degree
n. The asymptotic growth of b(n), as n → ∞ is an open question, although
there is a conjecture of Turyn that
b(n) ∼ K log(n),
p
for some constant K > 0. It is known that b(n) = O( n log(n)).
The zeros of Littlewood polynomials on the unit circle are of tangential
interest to this question. It is known that self-reciprocal Littlewood poly-
nomials have at least one zero on S 1 . Such a polynomial would obviously
violate (3). A Littlewood polynomial p as in (1) is skew-reciprocal if, for all
j, ad+j = (−1)j ad−j , where d = m/2 (m even) or d = (m − 1)/2 (m odd). A
skew-reciprocal Littlewood polynomials has no zeros on S 1 . (The Littlewood
polynomials having small autocorrelations also tend to be skew-reciprocal.)
We refer to Mercer [M] for more details. Hopefully this was a worthwhile
side-trip.

6 The Weil zeta function of a curve over a

finite field
7
Let X be a smooth projective curve of genus g over a finite field GF (q).
7
These terms will not be defined precisely here. Please see standard texts for a rigorous
treatment.

10
Example 19 The curve

y 2 = x7 − x,

over GF (7) is a curve of genus 3.

Suppose X is defined by a polynomial equation F (x, y) = 0, where F

is a polynomial with coefficients in GF (q). Let Nk denote the number of
solutions in GF (q k ) and create the generating function

G(z) = N1 z + N2 z 2 /2 + N3 z 3 /3 + · · · .

Define the zeta function of X by the formal power series

ζ(z) = ζX (z) = exp(G(z)) (5)

so ζ(0) = 1. In particular, the logaritmic derivative of ζ(z) has integral
coefficients. It is known that
P (z)
ζX (z) = ,
(1 − z)(1 − qz)

with P = PX a polynomial8 , of degree 2g where g is the genus of X. This

has a “functional equation” of the form
1
P (z) = q g z 2g P ( ).
qz
The logarithmic derivative of ζX is the generating function of the sequence
of counting numbers {N1 , N2 , . . .}. The Riemann hypothesis (RH) for curves
over finite fields states that the roots of P have absolute value q −1/2 . It
is well-known that the RH holds for ζX (this is a theorem of André Weil
from the 1940’s). Therefore, by a suitable change-of-variable (replacing z
√
by z/ q), we see that curves over finite fields give rise to a large class of
example of polynomials having roots on the unit circle.

Example 20 We use Sage to compute an example.

8
Sometimes called the reciprocal of the Frobenius polynomial, or the zeta polynomial.

11
 Sage

sage: R.<x> = PolynomialRing(GF(31))

sage: H = HyperellipticCurve(xˆ5 - x)
sage: time H.frobenius_polynomial()
CPU times: user 0.04 s, sys: 0.01 s, total: 0.05 s
Wall time: 0.16 s
xˆ4 + 62*xˆ2 + 961
sage: C.<z> = PolynomialRing(CC, "z")
sage: f = zˆ4+62*zˆ2+961
sage: rts = f.roots()
sage: [abs(z[0]) for z in rts]
[5.56776436283002, 5.56776436283002]
sage: RR(sqrt(31))
5.56776436283002

In other words, the zeta polynomial

PH (z) = 961z 4 + 62z 2 + 1

2 5
√ y = x − x over GF (31) satisfies the
associated to the hyperelliptic curve
RH. The polynomial p(z) = PH (z/ 31) is self-reciprocal, having all its zeros
on S 1 .
It can be shown that if X1 and X2 are “isomorphic” curves then the
corresponding zeta polynomials are equal. Therefore, these polynomials can
be used to help classify curves.

7 The Duursma zeta function of an error-

correcting code
Let F = GF (q) denote a finite field, for some prime power q.
Definition 21 Fix once and for all a basis for the vector space V = Fn . A
subset C of V = Fn is called a code of length n. A subspace of V is called
a linear code of length n. If F = GF (2) then C is called a binary code. If
F = GF (3) then C is called a ternary code. In general, when F = GF (q)
then C is called a q-ary code. The elements of a code C are called codewords.
Example 22 Let F = GF (2) and

C = {(0, 0, 0, 0), (1, 0, 0, 1), (0, 1, 1, 0), (1, 1, 1, 1)}.

This is a linear binary code which is a 2-dimensional subspace of V = GF (2)4 .

12
 If · denotes the usual inner product,

v · w = v1 w1 + . . . + vn wn ,

where v = (v1 , . . . , vn ) ∈ V and w = (w1 , . . . , wn ) ∈ V , then we define the

dual code C ⊥ by

C ⊥ = {v ∈ V | v · c = 0, ∀c ∈ C}.

We say C is self-dual if C = C ⊥ .
For each vector v ∈ V , let

supp(v) = {i | vi 6= 0}

denote the support of the vector. The weight of the vector v is wt(v) =
|supp(v)|. The weight distribution vector or spectrum of a code C ⊂ Fn is the
vector

A(C) = spec(C) = [A0 , A1 , ..., An ]

where Ai = Ai (C) denote the number of codewords in C of weight i, for

0 ≤ i ≤ n. Note that for a linear code C, A0 (C) = 1, since any vector space
contains the zero vector. The weight enumerator polynomial AC is defined
by
n
X
AC (x, y) = Ai xn−i y i = xn + Ad xn−d y d + . . . + An y n .
i=0

Denote the smallest non-zero weight of any c ∈ C by d = dC and the smallest

non-zero weight of any c ∈ C ⊥ by d⊥ = dC ⊥ .

The connection between the weight enumerator of C and that of its dual
is very close, as the following well-known result shows.

Theorem 23 (MacWilliams’ identity) If C is a linear code over GF (q) then

AC ⊥ (x, y) = |C|−1 AC (x + (q − 1)y, x − y).

13
Definition 24 A polynomial P = PC for which
(xT + (1 − T )y)n AC (x, y) − xn n−d
P (T ) = . . . + T + ... .
(1 − T )(1 − qT ) q−1
is called a Duursma zeta polynomial of C. The Duursma zeta function is
defined in terms of the zeta polynomial by means of
P (T )
ζC (T ) = ,
(1 − T )(1 − qT )

It can be shown that if C1 and C2 are “equivalent” codes then the cor-
responding zeta polynomials are equal. Therefore, these polynomials can be
used to help classify codes.

Proposition 25 The Duursma zeta polynomial P = PC exists and is unique,

provided d⊥ ≥ 2.

proof: This is proven in the appendix to Chinen [Ch]. Here is the rough
+y(1−T ))n
idea. Expand (xT(1−T )(1−qT )
in powers of T . The Duursma polynomial is
a polynomial of degree n + 2 − d − d⊥ . Provided d⊥ ≥ 2, the Duursma
polynomial can be written as P (T ) = a0 + a1 T + . . . + an−d T n−d . Now, use

(xT + y(1 − T ))n F (x, y) − xn n−d

P (T ) = . . . + T + ... ,
(1 − T )(1 − qT ) q−1
~ given
to express the coefficients by means of the matrix equation B · ~a = A
by

 
b0,0 b1,0 . . . bn−d,0    
an−d An /(q − 1)
 0 b1,1 . . . bn−d,1 
   an−d−1   An−1 /(q − 1) 

 0 0 b2,2 ... 
 .. =
 
.. .

(6)
 .. .. ..  . .
.
  
 . . 
a0 Ad /(q − 1)
0 . . . 0 bn−d,n−d

Thanks to an explicit computation (omitted), we know the diagonal entries

 n 
of this matrix are binomial coefficients, bi,i = , hence are non-zero.
i
Therefore the matrix is invertible and the existence is established. 

14
 If C is self-dual (i.e., C = C ⊥ ), the Duursma zeta polynomial satisfies a
functional equation of the form
1
P (T ) = q g T 2g P ( ).
qT
√
After a suitable change-of-variable, replacing T by T / q, these polynomials
are self-reciprocal.
In general, the analog of the Riemann hypothesis for curves does not hold
for the Duursma zeta polynomials of self-dual codes. However, when it does
√
hold, after a suitable change-of-variable (replacing T by T / q), we see that
self-dual codes give rise to a class of example of polynomials having roots on
the unit circle.

Example 26 We use Sage to compute an example.

Sage

sage: MS = MatrixSpace(GF(2),4,8)
sage: G = MS([[1,1,1,1,0,0,0,0],[0,0,1,1,1,1,0,0],[0,0,0,0,1,1,1,1],[1,0,1,0,1,0,1,0]])
sage: C = LinearCode(G)
sage: C == C.dual_code()
True
sage: C.zeta_polynomial()
2/5*Tˆ2 + 2/5*T + 1/5
sage: C.<z> = PolynomialRing(CC, "z")
sage: f = (2*zˆ2+2*z+1)/5
sage: rts = f.roots()
sage: [abs(z[0]) for z in rts]
[0.707106781186548, 0.707106781186548]
sage: RR(sqrt(2))
1.41421356237310
sage: RR(1/sqrt(2))
0.707106781186548

In other words, the Duursma zeta polynomial

PC (T ) = (2T 2 + 2T + 1)/5

associated to “the” binary self-dual code of length 8 satisfies the analog of

the RH.

15
7.1 Duursma’s conjecture
There is an infinite family of Duursma zeta functions for which Duursma
conjectures that the analog of the RH always holds. The linear codes used
to construct these zeta functions are so-called “extremal self-dual codes.”
Although the construction of these codes is fairly technical (see [JK] for
an expository treatment), we can give some examples.

Example 27 Let

p(z) = a0 + a1 z + . . . + an z n
√
denote the normalized Duursma zeta polynomial, so p(z) = P (z/ q) is self-
reciprocal.
Some examples of the lists of coefficients a0 , a1 , . . ., computed using Sage.
We have normalized the coefficients so that they sum to 10 and represented
the rational coefficients as decimal approximations to give a feeling for their
relative sizes.

• Case Type I:
m = 2: [1.1309, 2.3990, 2.9403, 2.3990, 1.1309]
m = 3: [0.45194, 1.2783, 2.0714, 2.3968, 2.0714, 1.2783, 0.45194]
m = 4: [0.18262, 0.64565, 1.2866, 1.8489, 2.0724, 1.8489, 1.2866,
0.64565, 0.18262]

• Case Type II:

m = 2: [0.43425, 0.92119, 1.3028, 1.5353, 1.6129, 1.5353, 1.3028,
0.92119, 0.43425]
m = 3: [0.12659, 0.35805, 0.63295, 0.89512, 1.1052, 1.2394, 1.2854,
1.2394, 1.1052, 0.89512, 0.63295, 0.35805, 0.12659]
m = 4: [0.037621, 0.13301, 0.28216, 0.46554, 0.65783, 0.83451, 0.97533,
1.0656, 1.0967, 1.0656, 0.97533, 0.83451, 0.65783, 0.46554, 0.28216,
0.13301, 0.037621]

• Case Type III:

m = 2: [1.3397, 2.3205, 2.6795, 2.3205, 1.3397]
m = 3: [0.58834, 1.3587, 1.9611, 2.1836, 1.9611, 1.3587, 0.58834]

16
 m = 4: [0.26170, 0.75545, 1.3085, 1.7307, 1.8874, 1.7307, 1.3085,
0.75545, 0.26170]
• Case Type IV:
m = 2: [2.8571, 4.2857, 2.8571]
m = 3: [1.6667, 3.3333, 3.3333, 1.6667]
m = 4: [0.97902, 2.4476, 3.1469, 2.4476, 0.97902]

Hopefully it is clear that, at least in these examples, these “extremal” Du-

ursma zeta functions are self-reciprocal and in “increasing symmetric form.”

7.2 Heuristic argument

Are there conditions under which self-reciprocal polynomials with in “in-
creasing symmetric form” have all their zeros on S 1 ?
We know that self-reciprocal polynomial with “decreasing symmetric form”
have all their roots on S 1 . Under what conditions is the analogous statement
true for functions with “increasing symmetric form?” The remainder of this
section considers this question for polynomials of even degree.
Let d be an odd integer and let f (z) = f0 + f1 z + . . . fd−1 z d−1 ∈ Rd−1 be
a self-reciprocal polynomial with “increasing symmetric form”

0 < f0 < f1 < . . . < f d−1 .

2

For each c ≥ f d−1 , the polynomial

2

g(z) = c · (1 + z + . . . + z d−1 ) − f (z) = g0 + g1 z + . . . gd−1 z d−1 ∈ Rd−1 ,

is a self-reciprocal polynomial having non-negative coefficients with “decreas-
ing symmetric form.” If c > f d−1 , the Chen-Chinen theorem (Theorem 15)
2
implies, all the zeros of g(z) are on S 1 . Let

P0 (z) = g(z)/gd−1 , P1 (z) = f (z)/fd−1 , Pa (z) = (1 − a)P0 (z) + aP1 (z),

for 0 ≤ a ≤ 1. By the Chen-Chinen theorem, there is a t0 ∈ (0, 1) such that
all zeros of Pt (z) are on S 1 for 0 ≤ t < t0 . In fact, if
f d−1 − fd−1
2
t= ,
f d−1
2

17
then Pt (z) is a multiple of 1 + z + . . . + z d−1 .
Do any of the polynomials Pt (z) have multiple roots (0 < t < 1)? Using
the notation of §2, in the case p(t, z) = Pt (z), we have

P1 (r(t)) − P0 (r(t))
r0 (t) = −pt (t, r(t))/pz (t, r(t)) = .
Pt0 (r(t))

If no Pt (z) has a multiple root, then by the second “smoothness of roots

lemma” (Lemma 3), all the roots of f (z) are also on S 1 .

References
[Chen] W. Chen, On the polynomials with all there zeros on the unit circle,
J. Math. Anal. and Appl. 190 (1995)714-724.

[Ch] K. Chinen, An abundance of invariant polynomials satisfying the Rie-

mann hypothesis, preprint. Available:
http://arxiv.org/abs/0704.3903

[Fe] H. J. Fell, On the zeros of convex combinations of polynomials, Pac. J.

Math. 89 (1980)43-50.

[DH] S. DiPippo, E. Howe, Real polynomials with all roots on the unit circle
and abelian varieties over finite fields, J. Number Theory 78(1998)426-
450.
Available: http://arxiv.org/abs/math/9803097

[D] Iwan Duursma, Weight distributions of geometric Goppa codes, Trans-

actions of the AMS, vol. 351, pp. 3609-3639, September 1999.
Available: http://www.math.uiuc.edu/~duursma/pub/

[Fe] H. J. Fell, On the zeros of convex combinations of polynomials, Pac. J.

Math. 89 (1980)43-50.

[JK] D. Joyner and J.-L. Kim, Selected Unsolved Problems in Coding

Theory, to appear in Birkhäuser, 2011.

[L1] P. Lakatos, On polynomials having zeros on the unit circle, C. R. Math.

Acad. Sci. Soc. R. Can. 24 , (2), (2002), 9196.

18
[L2] ——, On zeros of reciprocal polynomials, Publ. Math. Debrecen 61,
(2002), 645661.

[Los] L. Losonczi, On reciprocal polynomials with zeros of modulus one Math-

ematical Inequalities & Applications 9 (2006), 286-298.
http://mia.ele-math.com/09-29/On-reciprocal-polynomials-with-zeros-of-modulus-one

[Ma] M. Marden, Geometry of Polynomials, American Mathematical So-

ciety, 1970.

[M] I. D. Mercer, Autocorrelation and Flatness of Height One Polynomials,

PhD thesis, Simon Fraser University, 2005.
http://www.idmercer.com/publications.html

[S] W. Stein and the Sage group, Sage - Mathematical Software, version
4.5, 2010.
http://www.sagemath.org/.

19