Arctan
Arctan
Arctan
Jack S. Calcut
1 Introduction
We study the tangent analogues tan(n arctan x) of the Chebyshev polynomials from an algebraic
viewpoint. They are rational functions of a pleasant form and enjoy several noteworthy properties:
a useful composition law, their numerators pn (x) split into the minimal polynomials of the numbers
tan kπ/n, they define the elements of the Galois groups of these minimal polynomials, and their
algebraic and number theoretic properties strongly parallel those of the roots of unity. We give
a complete factorization of the polynomials pn (x) in Z[x], give several applications, and list some
open problems.
1
If n > 0, then
pn (x)
F−n (x) = −Fn (x) = − .
qn (x)
Therefore, for each integer n, Fn (x) is a rational function with integer coefficients. For example
x 2x 3x − x3
F1 (x) = , F2 (x) = , F3 (x) = ,
1 1 − x2 1 − 3x2
4x − 4x3 5x − 10x3 + x5
F4 (x) = 2 4
, and F5 (x) = .
1 − 6x + x 1 − 10x2 + 5x4
Clearly pn (x) is the alternating sum of the odd power terms in the binomial expansion (1 + x)n
and qn (x) is the alternating sum of the even power terms.
Remark 1. The Chebyshev polynomials of the first kind Tn (x) = cos(n arccos x) are integer poly-
nomials. The analogous functions for sine are not in general polynomials or rational functions.
r3 r2
r4 r1
r0 = 0
Figure 1: n = 5.
Similarly, qn (x) has roots tan kπ/2n where k = 1, 3, . . . , 2n − 1 if n is even and further k 6= n if
n is odd. Thus, for fixed n, pn (x) and qn (x) have real, distinct, and simple roots, and Fn (x) =
pn (x)/qn (x) is reduced.
2
The nth roots of unity form a multiplicative cyclic group. Analogously, the tangent functions
enjoy the following easily verified properties:
Fi ◦ Fj (x) = Fij (x) and (6)
Fi (rj ) = rij , (7)
where i and j are arbitrary integers and n ∈ N is fixed for the rj . Equation (6) will be called the
composition law. Note that the composition law does not hold precisely when Fj (x) does not exist
and i is even, in which case Fi ◦ Fj (x) does not exist and Fij (x) = 0. The composition law provides
a natural proof that pn (x) has only the obvious rational roots (compare [1]).
Proposition 1. The rational roots of pn (x) are x = 0 if 4 ∤ n and x = 0 and x = ±1 if 4|n.
Proof. The stated numbers are obviously roots and the result is clear if n ≤ 2. So, write n =
s1 s2 · · · sj > 2 a product of positive primes. We proceed by induction on j with the inductive
hypothesis: the possible rational roots of pn (x) are x = 0 and x = ±1 and of qn (x) are x = ±1.
If j = 1, then n is an odd prime and the rational root theorem (RRT) implies that the possible
rational roots of pn (x) are x = 0, x = ±1, and x = ±n. Substituting x = ±n into pn (x) = 0 implies
n|1, a contradiction. Similar reasoning proves the j = 1 case for qn (x).
3
Proof. If rj exists, then by (7)
rj = Fj (r1 ) ∈ Q(r1 ) (8)
since Fj is a rational function with integer coefficients. In particular, Q(r1 ) is the splitting field of
pn (x) over Q. Write ak + bn = 1 for integers a and b. Then
Equations (8) and (9) imply that Q(rk ) = Q(r1 ). Finally, ψ(x)|pn (x) in Q[x] since ψ(x) is minimal
and they share a root. The result follows.
Lemma 2. Let n > 2 and (k, n) = 1. Then each σ ∈ Gal (Q(rk )/Q) is induced by some Fm where
(m, n) = 1. In particular, Gal (Q(rk )/Q) is abelian. Additionally, if 4|n, then m ≡ 1 mod 4.
Proof. Let σ ∈ Gal (Q(rk )/Q). Such an automorphism permutes the roots of pn (x) and so
σ(r1 ) = rm = Fm (r1 )
n o
for some m. Now, Q(rk ) = Q(r1 ) has a basis 1, r1 , r12 , . . . , r1d−1 over Q where d = [Q(r1 ) : Q].
Therefore σ is completely determined by [m]n and we say that σ is induced by Fm . Notice that
σ ∈ Gal (Q(rk )/Q) commutes with each rational function Fi . By (7), σ acts on each rj by
It follows that Gal (Q(rk )/Q) is abelian. The inverse of σ is induced by say Fl . Hence, rlm = r1
and (m, n) = 1. Finally if 4|n, then σ must fix 1 = rn/4 which implies m ≡ 1 mod 4.
Remark 3. The restriction (m, n) = 1 parallels the same restriction for the nth roots of unity.
The restriction m ≡ 1 mod 4 is special to the tangent numbers rk and comes from the line of slope
1. As we will show, these obvious restrictions are all of the restrictions. In other words, as with
the roots of unity, the Galois groups here are as large as possible.
If f (x) ∈ Z[x] and p > 1 is prime, then let f (x) denote the image of f (x) in Zp [x] under the
natural ring epimorphism.
Lemma 4. Let n be natural and p > 2 prime such that (p, n) = 1. Then pn (x) ∈ Zp [x] has simple
roots.
Proof. It suffices to show that the discriminant of pn (x) is nonzero. The discriminant of pn (x) equals
∆n mod p (see [17, pp. 82–88]) which is nonzero by Lemma 3 since p > 2 and (p, n) = 1.
4
Remark 4. The analogous result for xn − 1 is easier, since xn − 1 and its formal derivative nxn−1
are clearly coprime in Zp [x].
Lemma 5. Let n > 2 and p > 2 prime such that (p, n) = 1 and p ≡ 1 mod 4. Then rp is a root
of any minimal polynomial of r1 over Q.
It is remarkable that Dirichlet’s beautiful argument for roots of unity and cyclotomic polyno-
mials [12, pp. 51-52] may be recast in the present context.
Proof. Let h(x) be a minimal polynomial of r1 over Q. Then h(x)|pn (x) in Q[x] since pn (r1 ) = 0.
By Gauss’ lemma [6, p. 86] we may assume h(x) is primitive, that is h(x) ∈ Z[x] is not the zero
polynomial and the greatest common divisor of its coefficients is 1. So
for some f (x) ∈ Z[x]. The roots of pn (x) are distinct, so suppose by way of contradiction that
f (rp ) = 0. By (7), f (Fp (r1 )) = 0 where qp (r1 ) 6= 0. Let d = deg f and let g(x) ∈ Z[x] be the
polynomial obtained by clearing denominators in
Note that g(r1 ) = 0. Thus h(x)|g(x) in Q[x]. As h(x) is primitive, h(x)|g(x) in Z[x]. Write
k(x)h(x) = g(x)
where the latter uses p ≡ 1 mod 4 (this is a crucial point; if p ≡ 3 mod 4, then pp (x) = −xp and
the proof unravels.). Recall Fermat’s little theorem and exploit the Frobenius homomorphism to
obtain
p
k(x) h(x) = g(x) = f (xp ) = f (x) .
Note that h(x) has positive degree by (10) and since pn (x) has leading coefficient ±1 or ±n. Unique
factorization in Zp [x] implies that some irreducible factor of h(x) with positive degree divides f (x).
Hence, pn (x) = f (x) h(x) has a multiple root contradicting Lemma 4.
Up until this point all arguments have been elementary. The following lemma will employ
Dirichlet’s theorem [7, 15] on primes in arithmetic progressions.
Lemma 6. Let n and k be coprime naturals. If 4|n, then assume further that k ≡ 1 mod 4. Then
there exists a prime p > 2 such that (p, n) = 1, p ≡ k mod n, and p ≡ 1 mod 4.
Proof. First, assume 4|n so k ≡ 1 mod 4. By Dirichlet’s theorem there exist infinitely many primes
in the arithmetic progression 4na + k. Any such prime p > max {2, n} behaves as desired.
Next, assume n = 4m + 2. Case 1. k ≡ 1 mod 4. The previous argument proves this case.
Case 2. k ≡ 3 mod 4. Let K = k + n, so (K, n) = 1 and K ≡ 1 mod 4. Case 1 applied to K
5
yields a prime p > 2 such that (p, n) = 1, p ≡ K ≡ k mod n, and p ≡ 1 mod 4, proving Case 2.
Finally, assume n is odd. The system of congruences X ≡ 1 mod 4 and X ≡ k mod n has a
solution l > 0 by the Chinese remainder theorem. The progression 4na + l contains infinitely many
primes by Dirichlet’s theorem. Any such prime p > 2 behaves as desired.
Corollary 2. Let n > 2 and let h(x) be a minimal polynomial of r1 over Q. Then rk is a root of
h(x) provided (k, n) = 1 and, in case 4|n, k ≡ 1 mod 4.
Proof. Without loss, assume k > 0. By Lemma 6 there exists a prime p > 2 such that (p, n) = 1,
p ≡ k mod n, and p ≡ 1 mod 4. Lemma 5 implies rp = rk is a root of h(x).
If 4|n, then
Hn = {[m]n | (m, n) = 1 and m ≡ 1 mod 4} (11)
is a well-defined subgroup of Z×
n of index 2 and order
and (
Z× if 4∤n
Gal (Q(rk )/Q) ∼
= n
Hn if 4|n.
Proof. We have degQ rk = [Q(rk ) : Q] = |Gal (Q(rk )/Q)|. The stated degree (13) is an upper
bound for degQ rk by Lemma 2 and equations (11) and (12) and is a lower bound by Lemma 1 and
Corollary 2. This proves (13). By Lemma 2 each σ ∈ Gal (Q(rk )/Q) is induced by some Fm so
naturally we define
Φ : AutQ Q (rk ) → Z×
n
σ 7→ [m]n
Remark 5. In [13, pp. 33–41] Niven proved that if n > 2, n 6= 4, and (k, n) = 1, then
2kπ
ϕ(n) if 4 ∤ n
degQ tan = ϕ(n)/2 if n ≡ 4 mod 8 (14)
n
ϕ(n)/4 if n ≡ 0 mod 8.
The reader may verify that (13) and (14) agree. Niven’s proof has the advantage of being indepen-
dent of a difficult result (for us, Dirichlet’s theorem), but it is somewhat ad hoc and lacks insight,
and the resulting statement (14) is clumsier than (13). With the present approach the Galois groups
Gal (Q(rk )/Q) and (13) emerge naturally, the factor of 2 in (13) is transparent (slope 1 must be
6
fixed), and deep parallels are revealed between the tangent numbers rk and roots of unity especially
in the proof of Lemma 5.
Pedagogically minded readers may be interested in the following natural approach to Theorem 1
which avoids Dirichlet’s theorem. The material on the functions Fn (x) preceding Remark 3, which
is intuitive and elementary, should be presented first (Proposition 1 may be safely omitted). At
this point, one has naturally motivated necessary restrictions on the relevant Galois groups and
it remains only to prove they are sufficient. One option is to employ Niven’s approach [13, pp.
33–41]. Here is another option that applies to cosine and sine as well. Fix n > 2 and let ζ =
exp 2πi/n. Then Gal (Q(ζ)/Q) ∼ = Z+ by Dirichlet’s argument [12, pp. 51-52]. Complex conjugation
is a nontrivial automorphism in Gal (Q(ζ)/Q) since n > 2. Let (k, n) = 1. The cosine proof is the
simplest: as cos 2kπ/n = (ζ k + ζ −k )/2 we see that Q(cos 2kπ/n) ⊂ Q(ζ) is the fixed field of complex
conjugation and so degQ cos 2kπ/n = ϕ(n)/2. Next we outline the proof for tangent, the proof for
sine being similar. If n 6= 4, then
1 ζ k − ζ −k
tan 2kπ/n = .
i ζ k + ζ −k
Define
ζ k − ζ −k
F =Q
ζ k + ζ −k
and consider the lattice of fields
Q(ζ, i) (15)
iii
iiii1,2
Q(ζ)
1,2
h F (i)
hhhh
hhhh 1,2
ϕ(n) F 2
Q(r2k )
hhh
hhhh
Q h
Strictly speaking, the node Q(ζ, i) is unnecessary, but it is the natural overfield and its inclusion
makes a pretty picture. In each of the three cases 4 ∤ n, n ≡ 0 mod 8, and n ≡ 4 mod 8 one
determines the upper left vertical degree and the middle diagonal degree using elementary field
theory. The result is equation (14). This proves (13) and the proof of Theorem 1 follows as above
verbatim.
7
Fix n > 2. For each positive divisor d 6= 2 of n define
Y
ψd (x) = (x − rj·n/d ). (17)
[j]d ∈Z+
d
Notice that rj·n/d = tan jπ/d. This explains the condition d 6= 2 in (17) and shows that the
polynomials (17)–(19) are actually independent of n. Lemma 7 and Theorem 1 imply that if 4 ∤ d,
then (17) is the monic minimal polynomial of rn/d and if 4|d, then (18) and (19) are the monic
minimal polynomials of rn/d and r−n/d respectively.
Lemma 8. If d 6= e, then ψd (x) and ψe (x) have no common roots. If 4|d, then ψd+ (x) and ψd− (x)
have no common roots and ψd (x) = ψd+ (x)ψd− (x).
Proof. First write d = (d, e)d′ , e = (d, e)e′ , and n = (d, e)d′ e′ n′ . If ψd (x) and ψe (x) share the root
rjn/d = rkn/e where (j, d) = 1 and (k, e) = 1, then (d, e)d′ e′ |je′ − kd′ . Hence d′ = e′ = 1 proving the
first part. Next, let 4|d. If rjn/d = r−kn/d is a common root where j, k ≡ 1 mod 4, then j ≡ −k
mod 4 a contradiction. The last part is clear.
Remark 6. Equations (17)–(19) admit equivalent definitions with the products over suitable classes
in Z+
n . In the tower of fields
Q ⊆ Q(rn/d ) ⊆ Q(r1 ) (20)
Q(rn/d ) is a stable intermediate field, Gal Q(rn/d )/Q
permutes the roots of a minimal polynomial
of rn/d transitively, and each τ ∈ Gal Q(rn/d )/Q extends to some σ ∈ Gal (Q(r1 )/Q) by the
fundamental theorem of Galois theory. In the present cases, extendability may be shown directly.
For instance if d|n are naturals and (j, d) = 1, then there exists an integer l such that [l]d = [j]d
and (l, n) = 1 (proof: l must have the form l = j + md for some m; let m be the product of the
primes dividing n but not dividing d or j.).
Theorem 2. A complete factorization of pn (x) into irreducibles in Q[x] is
Y
pn (x) = ± ψd (x) if (n, 4) = 1, (21)
d|n
Y
pn (x) = ±n ψd (x) if (n, 4) = 2, and (22)
26=d|n
Y Y Y
pn (x) = ±n ψd (x) ψd+ (x) ψd− (x) if (n, 4) = 4. (23)
4∤d|n 4|d|n 4|d|n
d6=2
8
5 Complete Factorization of pn (x) in Z[x]
We begin by stating some elementary results that will be used below without explicit mention.
A nonzero polynomial is primitive provided it lies in Z[x] and the greatest common divisor of
its coefficients is 1. The product of primitive polynomials is primitive [6, p. 85]. Every nonzero
f (x) ∈ Q[x] may be written uniquely in the form f (x) = cfb(x) where c ∈ Q+ and fb(x) is primitive
[6, p. 86]; c is called the content of f (x) and content is multiplicative.
Proposition 2. Each ψd (x) and ψd± (x) is primitive, unless d = 2pa for some odd prime p and
a > 0 in which case ψd (x) 6∈ Z[x] and pψd (x) is primitive.
Theorem 3. The factorization (21) is a factorization into irreducibles in Z[x]. If n = 2pa11 pa22 · · · pakk
where each aj > 0 and the pj are distinct odd primes, then
Y Y
pn (x) = ±2 ψd (x) pj ψ2pai (x) (24)
j
26=d|n 1≤j≤k
d6=2pbj 1≤i≤aj
is a factorization into irreducibles in Z[x]. If n = 2a pa11 pa22 · · · pakk where a > 1, each aj > 0, and
the pj are distinct odd primes, then
Y Y Y Y
pn (x) = ±2a ψd (x) pj ψ2pai (x) ψd+ (x) ψd− (x) (25)
j
4∤d|n 1≤j≤k 4|d|n 4|d|n
d6=2pbj 1≤i≤aj
Proof. First, pk (x) is primitive for every odd k and qk (x) is primitive for every k. Next
Lemma 11. Let k > 1 be an integer and f (x) = g(x) · kh(x) where f (x) and g(x) are primitive
and h(x) ∈ Q[x]. Then h(x) 6∈ Z[x] and kh(x) is primitive.
9
Proof. Both results follow from simple content arguments.
Lemma 12. Let n = 2pa where p > 0 is an odd prime and a > 0. Then for each d = 2pb , 1 ≤ b ≤ a,
we have ψd 6∈ Z[x] and pψd (x) is primitive.
where ψ1 (x) = x and ψp (x) are primitive by Lemma 9 and the content of pn (x) is 2 by Lemma 10.
Hence
1
± pn (x) = [ψ1 (x)ψp (x)] · pψ2p (x)
2
and the result follows by Lemma 11.
where each polynomial in the products in brackets is primitive by Lemma 9 or induction. The
result follows by Lemma 11.
Taking stock, Lemmas 9 and 12 prove Proposition 2 for odd d and d = 2pa respectively. These
results, (21), and (22) prove Theorem 3 for odd n and n = 2pa11 .
Q
Lemma 13. Let f (x) = g(x) ki=1 hi (x) where f (x) and g(x) are primitive and each hi (x) ∈ Q[x]
is monic. Then each hi (x) is primitive.
Proof. The hypotheses imply that f (x) and g(x) have equal leading coefficients. ForQeach iwrite
Q
hi (x) = ci hbi (x) where ci ∈ Q+ and hbi (x) is primitive. Then f (x) = ( ci ) g(x) hbi (x) and
Q Q
ci = 1. Comparing leading coefficients in f (x) = g(x) hbi (x), we see that each hbi (x) is monic.
As hi (x) = ci hbi (x) and hi (x) is monic, each ci = 1. The result follows.
Equation (22) implies (24) holds as a polynomial identity in Q[x]. We have already shown each
term in the second product in (24) is primitive, and therefore so is their product. Divide (24) by
2, the content of pn (x), and note that ±pn (x)/2 and the second product in (24) have equal leading
coefficients. Lemma 13 now implies that each term in the first product in (24) is primitive. An
almost identical argument applies to each term in the first, third, and fourth products in (25). This
completes the proof of Theorem 3 and Proposition 2.
Remark 7. Proposition 2 provides an effective and rigorous procedure to produce these integer
minimal polynomials using suitably precise floating point arithmetic on a computer: evaluate the
desired product (17), (18), or (19), if d = 2pa for some odd prime p and a > 0, then further
multiply the polynomial by p, and finally round each coefficient to the nearest integer. In practice,
an alternative is to use a computer algebra system such as MAGMA to factor an appropriate pn (x)
over Z and then pick out the correct irreducible factor by inspection.
10
6 Applications and Questions
6.1 Rational and Quadratic Irrational Values tan kπ/n
We compute exactly the values tan kπ/n of degree 1 or 2 over Q. We may p assume (k, n) = 1
and 0 ≤ kπ/n < π. Theorem 1 and the well-known inequality ϕ(n) ≥ n/2 reduce this to a
finite problem. The trivial cases n = 1 and n = 2 are inspected directly. The rational values are
tan 0 = 0, tan π/4 = 1, and tan 3π/4 = −1, in agreement with Corollary 1 above. For degree 2, we
find n = 2, 3, 6, 8, 12. The pertinent values of k and corresponding minimal polynomials are
x2 − 3 for n = 3 and k = 1, 2,
3x2 − 1 for n = 6 and k = 1, 5,
x2 + 2x − 1 for n = 8 and k = 1, 5,
x2 − 2x − 1 for n = 8 and k = 3, 7,
x2 − 4x + 1 for n = 12 and k = 1, 5, and
x2 + 4x + 1 for n = 12 and k = 7, 11.
Recently Cha [5, p. 16] used the above to give a nontrivial example where the so-called Grand
Simplicity Hypothesis for function fields may be verified.
This raises the question: which numbers tan kπ/n are expressible by real radicals? The Galois
groups Gal (Q(rk )/Q) are all abelian by Lemma 2 and hence are solvable. Therefore the numbers
tan kπ/n may all be expressed by radicals. However, expressions by real radicals are rarely possible.
If an irreducible polynomial f (x) ∈ Q[x] has all real roots and any root can be constructed from
11
Q by a combination of field operations and real mth roots, then deg f (x) is a power of 2 and the
Galois group of f (x) over Q is a 2-group [11]. If n = 7, then deg tan π/7 = ϕ(7) = 6 and tan π/7 is
not expressible by real radicals. This question is related to compass and straightedge constructions
and further investigation is left to the reader.
Therefore, 2 tan π/50 is not an algebraic integer either by the following lemma.
Remark 8. Even if h(x) is primitive, g(x) need not be primitive: take α = 1/2 and m = 2.
Corollary 3. The number rk = tan kπ/n with (k, n) = 1 is an algebraic integer if and only if n
is not of the form 2pa for some odd prime p and a > 0. Further, if n = 2pa and 0 6= m ∈ Z, then
mrk is an algebraic integer if and only if p|m.
Proof. The first assertion is immediate by Proposition 2. For the second assertion, rk has primitive
minimal polynomial pψ2pa (x) by Proposition 2 and (17). The leading coefficient of pψ2pa (x) is p
and some coefficient is not divisible by p. By Lemma 14, a minimal polynomial g(x) ∈ Z[x] of mrk
is obtained from pψ2pa (x) by multiplying each non-leading coefficient by a positive power of m.
Therefore p factors out of g(x) yielding a monic integer minimal polynomial of mrk if and only if
p|m.
Remark 9. Carlitz and Thomas [4] obtained the partial result that tan kπ/n with (k, n) = 1 is an
algebraic integer provided n is odd or 4|n.
where each εi ∈ {−1, 1}. That is, expand (1+x)m , collect like powers, and change any subset of the
m + 1 signs. Two infinite families of SBPs emerge naturally from the functions Fn (x). Before we
12
present them let us make some preliminary observations. Define R(m) to be the number of SBPs
of degree m that are reducible in Q[x]. The interesting point is that there are no known general
irreducibilty criteria for polynomials that take the signs of the coefficients into account. Bounds
on R(m) must be obtained by other means.
If m > 1, then R(m) ≥ 4 by taking all the same signs and strictly alternating signs. If m is
odd, then one may exploit the symmetry of the binomial coefficients to obtain reducible SBPs with
factors x − 1 or x + 1 as follows. Let m = 2j + 1 > 1 and consider the SBP (27). By symmetry, (27)
has 1 as a root if and only if
Xj
m
(εi + ε2j+1−i ) =0 (28)
i
i=0
with each δi ∈ {−1, 0, 1} correspond to SBPs (27) having 1 or −1 as a root. This correspondence
is a bijection except for the possibility that some solution δ to (31) may correspond to a SBP
having 1 and −1 as roots (i.e. both (28) and (29) are satisfied). By (30), this does not occur if
some δi = 0. Equation (31) has the obvious solution δ = ~0 (i.e. all δi = 0) which corresponds to
2j+1 distinct solutions to each of (28) and (29). Hence, if m = 2j + 1 > 1 is odd, then R(m) ≥ 2j+2 .
It is not clear how to produce more reducible or irreducible SBPs. Using MAGMA and a few
weeks computing time on multiple Sun Fire V20z nodes of an AMD64 cluster we collected the data
in Table 1. In particular the lower bound R(m) ≥ 2j+2 for m = 2j + 1 is tight if 1 < m ≤ 27,
m 1 3 5 7 9 11 13 15 17 19 21 23 25 27
R(m) 0 8 16 32 64 128 288 512 1024 2048 4096 8192 16384 32768
m 2 4 6 8 10 12 14 16 18 20 22 24 26 28
R(m) 4 4 12 12 8 8 32 4 12 16 8 104 16 8
Table 1: The number R(m) of reducible signed binomial polynomials of degree m ≤ 28.
except when m = 13. This anomaly is completely explained by the fact that if m = 13, then (31)
has, in addition to the obvious solution δ = ~0, the two solutions δ = ±[0, 0, 0, −1, 1, 1, −1] each
13
corresponding to 23 distinct solutions to (28) and (29) for a total of 2 · 27 + 2 · 23 + 2 · 23 = 288
reducible SBPs.
Again computing with MAGMA for several weeks we obtained all exceptional solutions to (31)
for m = 2j + 1 ≤ 23 as in Table 2. It appears that these exceptional solutions are more than a
j ±δ
6 [0, 0, 0, 1, −1, −1, 1]
14 [0, 0, 0, 0, 0, 0, 0, 1, −1, 0, 1, 1, −1, 0, 0]
15 [0, 0, 0, 1, −1, 0, 0, −1, −1, 0, −1, 1, 1, −1, −1, 1]
[0, 0, 0, 1, −1, 0, 0, −1, −1, 1, 0, 1, 0, 1, 0, −1]
16 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, −1, −1, 1, 0]
17 [1, −1, 1, −1, 1, −1, 1, −1, 1, −1, 1, 1, 0, −1, −1, 1, 0, 0]
[1, −1, 1, −1, 1, −1, 1, −1, 1, −1, 1, −1, 1, −1, −1, 1, 0, 0]
20 [0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, −1, 1, 0, −1, 0, −1, 1, 0, 0]
[0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, −1, −1, 1, −1, 0, −1, 1, 0, 0]
[0, 0, 0, 0, 0, 1, −1, 1, 1, 1, 0, −1, 0, 1, 0, −1, −1, 1, 0, 0, 0]
[0, 0, 0, 0, 0, 1, −1, 1, 1, 1, 0, −1, 0, −1, 1, −1, −1, 1, 0, 0, 0]
23 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, −1, 0, −1, −1, 1, 0, 0, 0, 0, 0]
Having introduced SBPs, let us explain their relation to the tangent functions. The two infinite
families of SBPs in Table 3 emerge naturally from the functions Fn (x). Moreover, we may determine
exactly when each is irreducible.
Lemma 15. The SBPs qn (x) ± pn (x) are irreducible if and only if n = 2j for some j.
Therefore
p4n (x) = 2p2n (x)(qn (x) − pn (x))(qn (x) + pn (x)). (32)
14
As qn (x) is even and pn (x) is odd, α is a root of qn (x) − pn (x) if and only if −α is a root of
qn (x)+pn (x). Therefore (32) implies that tan π/4n is a root of one of the polynomials qn (x)−pn (x)
or qn (x) + pn (x) and tan(4n − 1)π/4n is a root of the other. Both of these numbers have degree
ϕ(4n)/2 over Q by (13) and both of these polynomials has degree n. So, they are irreducible if and
only if and only if ϕ(4n)/2 = n and this occurs if and only if n is a power of 2.
In particular, when n is not a power of two these polynomials yield four new reducible SBPs
in addition to the obvious four. Hence if n 6= 2j , then R(m) ≥ 8. Table 1 shows that for
m = 10, 12, 22, 28 these are the only reducible SBPs. That is, of the 536, 870, 912 SBPs of de-
gree 28 the reducible ones are exactly the ones with the same signs, strictly alternating signs, and
doubly alternating signs as in Table 1.
Notice that ±1 are roots of (32). If n is odd, then ±1 are roots of the polynomials in Table 3.
If n is even, then ±1 are roots of p2n (x) and (32) shows that ±1 are not roots of the polynomials
in Table 3. In particular, these yield examples of reducible SBPs without 1 or −1 as a root.
Conjecture 1. There does not exist a SBP of odd degree having both 1 and −1 as roots. Equiva-
lently, every signed sum of the first half of the elements in an odd row of Pascal’s triangle (com-
pare (31)) is nonzero.
Conjecture 2. A SBP of odd degree is reducible over Q if and only if it has 1 or −1 as a root.
R(m)
Conjecture 3. limm→∞ 2m+1
= 0.
Conjecture 4. There exist infinitely many even m for which R(m) = 8. In particular, R(m) 6→ ∞
as m → ∞.
Y
2n−1 n−1
X Y
n−1
x2n − 1 = x − ζk =⇒ x2j = x − ζk x − ζ n+k .
k=0 j=0 k=1
Y
n−1 Y
n−1
n= 1 − ζ 2k = ζ n(n−1)/2 ζ −k − ζ k .
k=1 k=1
15
Recalling that
kπ 1 ζ k − ζ −k
rk = tan = ,
n i ζ k + ζ −k
we obtain the modulus of the product of the nonzero roots of pn (x):
n−1
Y n−1
Y n−1
Y
ζ k − ζ −k ζ −k − ζ k
rk = = . (34)
ζ k + ζ −k ζ −k + ζ k
k=1 k=1 k=1
k6=n/2 k6=n/2 k6=n/2
The value of (34) equals the absolute value of the constant coefficient of the monic polynomial
±pn (x)/x if n is odd and ±pn (x)/nx if n is even. The value of the former is n and the value of the
latter is 1. These values, equations (33) and (34), and ζ −n/2 − ζ n/2 = 2 imply that
(
n−1
Y 1 if n is odd
ζ −k + ζ k = (35)
k=1
n/2 if n is even.
k6=n/2
Next, we compute the following product that is intimately related to the discriminant (see [17,
pp. 82–88]): Y
δn = (rk − rl )2 ,
rk 6=rl
where the product is taken over pairs of distinct roots of pn (x). The roots of pn (x) are real and
distinct so δn > 0. Thus, we may compute δn using the modulus. If n is odd, then
Y
δn = |rk − rl |2
0≤k<l≤n−1
Y 2
2 ζ k−l − ζ l−k
=
i (ζ k + ζ −k ) (ζ l + ζ −l )
0≤k<l≤n−1
Qn−1 −k 2(n−k)
k=1 ζ − ζk
= 2(n−1)(n−2) Q 2n−2 (36)
n−1 k −k )
k=1 (ζ + ζ
Y
n−1 n
(n−1)(n−2)
=2 ζ −k − ζ k (37)
k=1
(n−1)(n−2) n
=2 n . (38)
Equation (36) combined the obvious n(n − 1) factors of 2 with the 2(n − 1) factors of ζ 0 + ζ −0 = 2
from the denominator and collected the remaining equal terms in the numerator and denominator
respectively. Equation (37) collected some equal terms in the numerator, namely ζ −(n−k) − ζ n−k =
ζ −k − ζ k , and noted that the denominator in (36) equals 1 by (35). Equation (38) used (33). If n
16
is even, then
Y
δn = (rk − rl )2
0≤k<l≤n−1
k,l6=n/2
Y 2
2 ζ k−l − ζ l−k
=
i (ζ k + ζ −k ) (ζ l + ζ −l )
0≤k<l≤n−1
k,l6=n/2
Qn−1 n−2
k=1 ζ −k − ζ k
= 2(n−1)(n−2) 2n−4 (39)
Qn−1
k=0 (ζ −k + ζ k )
k6=n/2
nn−2
= 2(n−1)(n−2) (40)
22n−4 (n/2)2n−4
= 2(n−1)(n−2) n2−n .
Equation (39) was obtained similarly to equations (36) and (37). Equation (40) used (33) in the
numerator and collected factors of 2 and used (35) in the denominator. Summarizing, we have
(
2(n−1)(n−2) nn if n is odd
δn = (n−1)(n−2) 2−n
2 n if n is even.
The discriminant ∆n of pn (x) is obtained by multiplying δn by the leading coefficient of pn (x) raised
to the power 2 deg pn (x) − 2 (see [17, p. 82]). If n is odd, then this has no effect. If n is even, then
∆n = (±n)2n−4 δn
= 2(n−1)(n−2) nn−2 .
Acknowledgments. The author thanks Gregory Johnson, Vince Vatter, and Sophie Huczynska
for helpful conversations.
References
[1] J. S. Calcut, Single rational arctangent identities for π, Pi Mu Epsilon J., 11 (1999) 1–6.
[2] J. S. Calcut, Gaussian integers and arctangent identities for pi, Amer. Math. Monthly to appear
1–20.
17
[3] J. S. Calcut, Grade school triangles, submitted preprint (2008).
[4] L. Carlitz and J. Thomas, Rational tabulated values of trigonometric functions, Amer. Math.
Monthly 69 (1962) 789–793.
[6] A. Clark, Elements of Abstract Algebra, Dover, New York, 1971, reprinted 1984.
[7] L. Dirichlet, Beweis des Satzes, dafs jede unbegrentze arithmetische Progression, deren
erstes Glied und Differenz ganze Zahlen ohne gemeinschaftlichen Factor sind, un-
endlich viele Primzahlen enthält, Abhandlungen der Königlichen Preussischen Akademie
der Wissenschaften zu Berlin (1837) 45-81; English translation by R. Stephan (2008)
arXiv:0808.1408v1.
[9] C. F. Gauss, Werke. Band II, Königlichen Gesellschaft der Wissenschaften zu Göttingen,
Göttingen, 1863.
[11] I. Isaacs, Solution of polynomials by real radicals, Amer. Math. Monthly 92 (1985) 571–575.
[12] J. S. Milne, Fields and Galois Theory (Version 4.21), Notes (2008) available at
http://www.jmilne.org/math/.
[13] I. Niven, Irrational Numbers, Carus Mathematical Monographs, no. 11, Mathematical Associ-
ation of America, Washington, DC, 1997; fourth printing, 1956.
[14] J. Olmsted, Rational values of trigonometric functions, Amer. Math. Monthly 52 (1945) 507–
508.
[15] A. Selberg, An elementary proof of Dirichlet’s theorem about primes in an arithmetic progres-
sion, Ann. of Math. (2) 50 (1949) 297–304.
[16] J. Stillwell, Mathematics and Its History, Springer-Verlag, New York, 1989.
[17] B. L. van der Waerden, Modern Algebra, Vol. 1, Second Ed., Ungar, New York, 1949.
18