betaFeedback

WorksMathematics

The totient bound and extremal gaps in cyclotomic polynomials

Abstract

Let m>1m>1 be odd and squarefree, and let p∤mp\nmid m be prime. If p<mp<m, every gap in ΦmpΦ_{mp} has length strictly less than φ(m)φ(m). If p>mp>m, the maximum gap has length φ(m)φ(m) and occurs exactly 2⌊p/m⌋2\lfloor p/m\rfloor times. We determine its positions and endpoint coefficients. These results prove the equality criterion of Ambrosino, Hong and Lee and the multiplicity formula of Zhang. We also treat repeated prime factors, even cofactors and extremal zero runs in periodic coefficient blocks.

Introduction

A primitive nnth root of unity is a complex number of multiplicative order nn. The nnth cyclotomic polynomial is Φn(x)=∏ζ(x−ζ)\Phi_n(x) = \prod_{\zeta}(x-\zeta), where the product runs over all such roots. Its degree is Euler’s totient φ(n)=#{1≤k≤n:gcd⁡(k,n)=1}\varphi(n) = \#\{1 \le k \le n : \gcd(k,n) = 1\}. For F(x)=∑k=0DckxkF(x) = \sum_{k=0}^{D} c_kx^k, write supp⁡F={k:ck≠0}\operatorname{supp} F = \{k : c_k \ne0\}. A gap is a pair (a,b)(a,b) of consecutive elements of supp⁡F\operatorname{supp} F. Its length is b−ab-a, so it contains b−a−1b-a-1 intermediate zero coefficients. Write g(F)g(F) for the greatest gap length and N(F)N(F) for the number of gaps of that length.

For odd primes q<pq < p, Hong, Lee, Lee and Park [1] proved that g(Φqp)=q−1g(\Phi_{qp}) = q-1. For example, with q=3q = 3 and p=5p = 5,

Φ15(x)=1−x+x3−x4+x5−x7+x8.\Phi_{15}(x) = 1-x+x^3-x^4+x^5-x^7+x^8.

The consecutive exponent differences are 1,2,1,1,2,11,2,1,1,2,1, so g(Φ15)=2g(\Phi_{15}) = 2. Camburu, Ciolan, Luca, Moree and Shparlinski [2] and Zhang [3] independently showed that the length q−1q-1 occurs 2⌊p/q⌋2\lfloor p/q\rfloor times. In Φ15\Phi_{15} the two occurrences are (1,3)(1,3) and (5,7)(5,7), in agreement with 2⌊5/3⌋=22\lfloor5/3\rfloor= 2.

Two conjectures extend these results to products of distinct odd primes. In both, write such a product as n=mpn = mp, where pp is its largest prime factor.

Conjecture 1.1 (Ambrosino, Hong and Lee [4 Conjecture 7]). The maximum gap of Φn\Phi_n has length φ(m)\varphi(m) if and only if p>mp > m.

Conjecture 1.2 (Zhang [5 Conjecture 0.1]). If m>1m > 1 and p>mp > m, then

g(Φmp)=φ(m),N(Φmp)=2⌊p/m⌋.g(\Phi_{mp}) = \varphi(m), \qquad N(\Phi_{mp}) = 2\lfloor p/m\rfloor.

Al-Kateeb, Ambrosino, Hong and Lee [6 Theorem 1] proved the length formula for p>mp > m. Theorem 1.1(i) completes Conjecture 1.1 by proving strict inequality for p<mp < m. Part (ii) proves the count in Conjecture 1.2 for any number of prime factors and gives the positions and endpoint coefficients of all maximum gaps.

Theorem 1.1. Let m>1m > 1 be odd and squarefree, let pp be a prime with p∤mp \nmid m, and put d=φ(m)d = \varphi(m).

(i) p<mp < m. The maximum gap length satisfies g(Φmp)<dg(\Phi_{mp}) < d.

(ii) p>mp > m. The maximum gap length is dd. Put D=d(p−1)D = d(p-1) and Q=⌊p/m⌋Q = \lfloor p/m\rfloor. There are exactly 2Q2Q gaps of length dd, whose pairs of endpoints are

(jm−d,jm)and(D−jm,D−jm+d),1≤j≤Q.(1)(jm-d,jm) \qquad\text{and}\qquad(D-jm,D-jm+d), \qquad1 \le j \le Q. \tag*{(1)}

In the first pair, the coefficient of xjm−dx^{jm-d} is −1-1 and that of xjmx^{jm} is +1+1. In the second, the coefficient of xD−jmx^{D-jm} is +1+1 and that of xD−jm+dx^{D-jm+d} is −1-1. Every coefficient strictly between the two endpoints of either pair is zero.

Example. Take m=15m = 15, so d=8d = 8. For p=13<15p = 13 < 15, the maximum in Φ195\Phi_{195} is 6<86 < 8 and occurs four times [5 Theorem 0.1]. For p=17>15p = 17 > 15, we have D=128D = 128 and Q=1Q = 1. The two gaps of length 88 in Φ255\Phi_{255} run from −x7-x^7 to +x15+x^{15} and from +x113+x^{113} to −x121-x^{121}.

2020 Mathematics Subject Classification. 11C08.

Key words and phrases. Cyclotomic polynomial, maximum gap, Euler totient function, roots of unity.

When n=pn=p, Conjecture 1.1 follows directly from Φp(x)=1+x+⋯+xp−1\Phi_p(x)=1+x+\cdots+x^{p-1} and φ(1)=1\varphi(1)=1. For m>1m>1, Theorem 1.1 allows any prime p∤mp\nmid m. Sanna [7, §6] records the proposed upper bound for p<mp<m. Zhang [5, Theorem 0.1] determined the maximum and its multiplicity for m=15m=15. Benoist and Perucca [8] proved the predicted count for m=3qm=3q, with q>3q>3 prime, p>3qp>3q and p≡±1,±2(mod3q)p\equiv\pm1,\pm2\pmod{3q}.

The proof of Theorem 1.1 uses the product Φm(x)Φmp(x)=Φm(xp)\Phi_m(x)\Phi_{mp}(x)=\Phi_m(x^p), which also occurs in Kaplan’s binary argument in [2 §2.1]. Truncating it at the right endpoint of a gap reduces the equality case to a question about a partial sum of Φm\Phi_m.

Corollary 1.2 (Odd cofactors). Let m>1m>1 be odd and let p∤mp\nmid m be prime. Put s=rad⁡(m)=∏q∣m, q primeqs=\operatorname{rad}(m)=\prod_{q\mid m,\ q\ \mathrm{prime}}q, d=φ(m)d=\varphi(m) and D=d(p−1)D=d(p-1). Then

g(Φmp)≤d,g(Φmp)=d⟺p>s.g(\Phi_{mp})\le d,\qquad g(\Phi_{mp})=d\quad\Longleftrightarrow\quad p>s.

The gaps of length dd are exactly

(jm−d,jm)and(D−jm,D−jm+d),1≤j≤⌊ps⌋.(jm-d,jm)\quad\text{and}\quad(D-jm,D-jm+d),\qquad1\le j\le\left\lfloor\frac{p}{s}\right\rfloor.

Their endpoint coefficients are (−1,1)(-1,1) and $(1,-1),respectively.Thusthereare$2⌊p/s⌋, respectively. Thus there are \$2\left\lfloor p/s\right\rfloor gaps of length $d.

Proof. For n>1n>1, put R=rad⁡(n)R=\operatorname{rad}(n) and v=n/Rv=n/R. In the Möbius product for Φn\Phi_n, a divisor t∣nt\mid n contributes precisely when n/t∣Rn/t\mid R, that is, when t=vet=ve with e∣Re\mid R. This gives the classical radical reduction [1 §2]

Φn(x)=∏e∣R(1−xve)μ(R/e)=ΦR(xv).(2)\Phi_n(x)=\prod_{e\mid R}(1-x^{ve})^{\mu(R/e)}=\Phi_R(x^v). \tag*{(2)}

Set h=m/sh=m/s. Since p∤mp\nmid m, we have rad⁡(mp)=sp\operatorname{rad}(mp)=sp and hence Φmp(x)=Φsp(xh)\Phi_{mp}(x)=\Phi_{sp}(x^h). Every gap (a,b)(a,b) of Φsp\Phi_{sp} corresponds to the gap (ha,hb)(ha,hb) with the same endpoint coefficients. Theorem 1.1 applied to ss now gives the claim, since hs=mhs=m and hφ(s)=dh\varphi(s)=d. □\square

For example, take m=9m=9 and p=5p=5. In Φ45(x)=Φ15(x3)\Phi_{45}(x)=\Phi_{15}(x^3) the gaps (1,3)(1,3) and (5,7)(5,7) of Φ15\Phi_{15} become (3,9)(3,9) and (15,21)(15,21). Both have length 6=φ(9)6=\varphi(9), as rad⁡(9)=3<5<9\operatorname{rad}(9)=3<5<9.

Section 4 treats every even cofactor by the radical and sign identities of [1 §2]. For squarefree even m>2m>2 and p∤mp\nmid m, the threshold is p>m/2p>m/2. The final section describes the cyclic runs of zeros for both parities.

Truncation and complementary partial sums

Fix m>1m>1. Write Φm(T)=∑j=0dujTj\Phi_m(T)=\sum_{j=0}^{d}u_jT^j, where d=φ(m)d=\varphi(m), and put Pi(T)=∑j=0iujTjP_i(T)=\sum_{j=0}^{i}u_jT^j for 0≤i<d0\le i<d. We have u0=ud=1u_0=u_d=1.

Proposition 2.1. For every prime p∤mp\nmid m one has g(Φmp)≤dg(\Phi_{mp})\le d. Suppose that (a,b)(a,b) is a gap of length dd, and let c=[xb]Φmp(x)c=[x^b]\Phi_{mp}(x). Then for every primitive mmth root ζ\zeta,

Pi(ζp)=−cζb,i=⌊b−1p⌋.(3)P_i(\zeta^p)=-c\zeta^b,\qquad i=\left\lfloor\frac{b-1}{p}\right\rfloor. \tag*{(3)}

Proof. Set F=ΦmpF=\Phi_{mp}. A root of unity of order nn has ppth power of order n/gcd⁡(n,p)n/\gcd(n,p). Since p∤mp\nmid m, this order equals mm precisely when n=mn=m or n=mpn=mp. The roots of Φm(xp)\Phi_m(x^p) are simple because this polynomial divides xmp−1x^{mp}-1. Both factors on the left below are monic, so comparison of roots and then degrees gives

Φm(x)F(x)=Φm(xp),D=deg⁡F=pd−d=d(p−1).(4)\Phi_m(x)F(x)=\Phi_m(x^p),\qquad D=\deg F=pd-d=d(p-1). \tag*{(4)}

For any gap (a,b)(a,b), let A(x)A(x) be the sum of the terms of F(x)F(x) of degree less than bb. Since aa is the last support exponent before bb, we have deg⁡A=a\deg A=a and F≡A(modxb)F\equiv A\pmod{x^b}. On the right of (2.2), the term ujxpju_jx^{pj} survives truncation modulo xbx^b precisely if pj<bpj<b. The indices in this range are exactly 0≤j≤i=⌊(b−1)/p⌋0\le j\le i=\lfloor(b-1)/p\rfloor. Moreover, b≤D=d(p−1)b\le D=d(p-1), since bb is an exponent of FF. Thus ip≤b−1<dpip\le b-1<dp, which gives i<di<d. Consequently

Φm(x)A(x)≡∑j=0iujxpj=Pi(xp)(modxb).(5)\Phi_m(x)A(x)\equiv\sum_{j=0}^{i}u_jx^{pj}=P_i(x^p)\pmod{x^b}. \tag*{(5)}

If b−a>db-a>d, the left side has degree d+a<bd+a<b, and the right side has degree at most pi≤b−1p i\le b-1. Their difference is divisible by xbx^b, hence is zero. Evaluation at a primitive mmth root ζ\zeta gives Pi(ζp)=0P_i(\zeta^p)=0. But ζp\zeta^p is again primitive of order mm, so its minimal polynomial over Q\mathbb{Q} has degree dd. This contradicts Pi(0)=1P_i(0)=1 and deg⁡Pi≤i<d\deg P_i\le i<d. The gap bound follows.

Now suppose b−a=db-a=d. By (5), there is a polynomial f(x)f(x) such that Φm(x)A(x)−Pi(xp)=xbf(x)\Phi_m(x)A(x)-P_i(x^p)=x^bf(x). The difference has degree exactly a+d=ba+d=b, since deg⁡Pi(xp)<b\deg P_i(x^p)<b. Its leading coefficient is c=[xa]F(x)c=[x^a]F(x), since Φm\Phi_m is monic. The quotient ff therefore has degree zero. Comparing the coefficients of xbx^b gives f=cf=c, so

Φm(x)A(x)=Pi(xp)+cxb.\Phi_m(x)A(x)=P_i(x^p)+cx^b.

At x=ζx=\zeta the left side vanishes, giving (3).

Lemma 2.2. Suppose mm is odd and squarefree. For a primitive mmth root η\eta and 1≤i≤d−21\le i\le d-2, one has ∣Pi(η)∣2∉Q|P_i(\eta)|^2\notin\mathbb{Q}.

Proof. Inversion permutes the primitive mmth roots, so Φm\Phi_m is reciprocal and uj=ud−ju_j=u_{d-j}. Put k=d−1−ik=d-1-i. Reversing the terms after index ii in Φm(η)=0\Phi_m(\eta)=0 gives

Pi(η)=−ηdPk(η−1),∣Pi(η)∣=∣Pk(η)∣.P_i(\eta)=-\eta^dP_k(\eta^{-1}),\qquad|P_i(\eta)|=|P_k(\eta)|.

Here η−1=η‾\eta^{-1}=\overline{\eta} and the coefficients of PkP_k are real. Let j=min⁡(i,k)j=\min(i,k) and B=PjB=P_j. Both ii and kk are positive, and 2j≤i+k=d−12j\le i+k=d-1. The product formula Φm(T)=∏s∣m(1−Ts)μ(m/s)\Phi_m(T)=\prod_{s\mid m}(1-T^s)^{\mu(m/s)}, with μ\mu the Möbius function, gives u1=−μ(m)≠0u_1=-\mu(m)\ne0, since only the factor with s=1s=1 contributes a linear term and mm is squarefree. Thus e=deg⁡Be=\deg B satisfies 1≤e≤j1\le e\le j, and ue≠0u_e\ne0 because ee is the largest index t≤jt\le j for which ut≠0u_t\ne0.

Suppose that C=∣Pi(η)∣2C=|P_i(\eta)|^2 belongs to Q\mathbb{Q}. By the equality of moduli above, C=∣B(η)∣2C=|B(\eta)|^2 for either choice of jj. Set

H(T)=Te(B(T)B(T−1)−C).H(T)=T^e\left(B(T)B(T^{-1})-C\right).

The exponents in B(T)B(T−1)B(T)B(T^{-1}) range from −e-e to ee, so H∈Q[T]H\in\mathbb{Q}[T] and deg⁡H≤2e<d\deg H\le2e<d. Since η−1=η‾\eta^{-1}=\overline{\eta} and BB has real coefficients, B(η−1)=B(η)‾B(\eta^{-1})=\overline{B(\eta)}. Hence

H(η)=ηe(B(η)B(η)‾−C)=ηe(∣B(η)∣2−C)=0.H(\eta)=\eta^e\left(B(\eta)\overline{B(\eta)}-C\right)=\eta^e\left(|B(\eta)|^2-C\right)=0.

The coefficient of T−eT^{-e} in B(T)B(T−1)B(T)B(T^{-1}) is u0ueu_0u_e. Since e>0e>0, it is also the constant coefficient of HH, and it is nonzero. Thus HH is a nonzero polynomial over Q\mathbb{Q} of degree less than dd vanishing at η\eta, contrary to the minimality of Φm\Phi_m.

The extremal gaps

Proof of Theorem 1.1. Set F=ΦmpF=\Phi_{mp}, D=d(p−1)D=d(p-1), and Q=⌊p/m⌋Q=\lfloor p/m\rfloor. Proposition 2.1 gives g(F)≤dg(F)\le d. For a gap (a,b)(a,b) of length dd, take i,c,ζi,c,\zeta as in (3). Since ∣Pi(ζp)∣2=c2∈Q|P_i(\zeta^p)|^2=c^2\in\mathbb{Q}, Lemma 2.2 applied to η=ζp\eta=\zeta^p gives i=0i=0 or i=d−1i=d-1.

If i=0i=0, then 1=−cζb1=-c\zeta^b and 0<b≤p0<b\le p. The only rational root of unity of odd order is 11. Thus m∣bm\mid b and c=−1c=-1. Writing b=jmb=jm gives the first family in (1.1).

If i=d−1i=d-1, then Pi(ζp)=−ζdpP_i(\zeta^p)=-\zeta^{dp}, so c=ζdp−b=1c=\zeta^{dp-b}=1 and m∣dp−bm\mid dp-b. The inequalities (d−1)p<b≤D(d-1)p<b\le D give d≤dp−b<pd\le dp-b<p. Write dp−b=jmdp-b=jm. Since D=dp−dD=dp-d, we obtain (a,b)=(D−jm,D−jm+d)(a,b)=(D-jm,D-jm+d), as in the second family.

In both cases jj is a positive integer with jm≤pjm\le p, hence j≤Qj\le Q.

(i) Suppose p<mp<m. Then Q=0Q=0, so neither case can yield a gap of length dd. By Proposition 2.1, all gaps therefore have length less than dd.

(ii) Suppose p>mp>m. Then Q≥1Q\ge1. To realize the candidates in (1.1), put U(x)=(1−xm)/Φm(x)U(x)=(1-x^m)/\Phi_m(x). This polynomial has degree m−dm-d, constant coefficient 11, and leading coefficient −1-1. In Z[[x]]\mathbb{Z}[[x]] we have

1Φm(x)=∑h≥0xhmU(x).(6)\frac{1}{\Phi_m(x)}=\sum_{h\ge0}x^{hm}U(x). \tag*{(6)}

By (4), F(x)F(x) and 1/Φm(x)1/\Phi_m(x) agree in all degrees less than pp. For 1≤j≤Q1\le j\le Q we have jm<pjm<p, since m>1m>1 and p∤mp\nmid m. The (j−1)(j-1)st translate in (6) ends at exponent jm−djm-d, and the jjth starts at jmjm. This gives the first family with endpoint coefficients (−1,1)(-1,1). The reciprocity of FF gives the second family and reverses the endpoint coefficients. The right endpoints in the two families are congruent to 0 and dpdp modulo mm, respectively. These classes are distinct since gcd⁡(p,m)=1\gcd(p,m)=1 and 0<d<m0<d<m. Each family therefore contributes QQ distinct gaps, giving 2Q2Q maximum gaps in total.

Example 3.1. Take m=3⋅5⋅7=105m=3\cdot5\cdot7=105 and p=211p=211. Then d=48d=48, D=10080D=10080, and Q=2Q=2. In increasing order the four maximum gaps are (57,105),(162,210),(9870,9918),(9975,10023)(57,105),\quad(162,210),\quad(9870,9918),\quad(9975,10023).

Even values of the cofactor

For an even integer mm, let rr be the product of its distinct odd prime divisors, with the empty product equal to 1. Write

m=2hr,L=m/2=hr,d=φ(m)=hφ(r).m=2hr,\qquad L=m/2=hr,\qquad d=\varphi(m)=h\varphi(r).

Throughout this section pp is a prime with p∤mp\nmid m, so pp is odd. Put D=d(p−1)D=d(p-1).

Theorem 4.1. Let mm be even and let p∤mp\nmid m be a prime. Then

g(Φmp)=h g(Φrp)≤d,g(Φmp)=d⟺p>r.g(\Phi_{mp})=h\,g(\Phi_{rp})\le d,\qquad g(\Phi_{mp})=d\Longleftrightarrow p>r.

If r>1r>1, the gaps of length dd and the coefficients at their endpoints, in increasing order of exponent, are given by

gap of length ddendpoint coefficients
(jL−d,jL)(jL-d,jL)((−1)j+1,(−1)j)\bigl((-1)^{j+1},(-1)^j\bigr)
(D−jL,D−jL+d)(D-jL,D-jL+d)((−1)j,(−1)j+1)\bigl((-1)^j,(-1)^{j+1}\bigr)
1≤j≤⌊p/r⌋1\le j\le\lfloor p/r\rfloor

Table 1.

There are exactly 2⌊p/r⌋2\lfloor p/r\rfloor such gaps. If r=1r=1, all p−1p-1 gaps have length d=hd=h. They are (kh,(k+1)h)(kh,(k+1)h) with coefficients ((−1)k,(−1)k+1)\bigl((-1)^k,(-1)^{k+1}\bigr), for $0\le k\le p-2.

Proof. We combine the radical reduction (2) with the sign identity of [1 §2]. For odd k>1k>1, negation bijects primitive kkth roots with primitive 2k2kth roots. Since φ(k)\varphi(k) is even, this gives

Φ2k(x)=Φk(−x).\Phi_{2k}(x)=\Phi_k(-x).

Now rad⁡(mp)=2rp\operatorname{rad}(mp)=2rp and rp>1rp>1, whence

Φmp(x)=Φ2rp(xh)=Φrp(−xh).(7)\Phi_{mp}(x)=\Phi_{2rp}(x^h)=\Phi_{rp}(-x^h). \tag*{(7)}

Thus a nonzero coefficient cc at degree aa in Φrp\Phi_{rp} gives the coefficient (−1)ac(-1)^ac at degree haha in Φmp\Phi_{mp}. The ordered supports correspond bijectively, so every gap is scaled by hh with its multiplicity preserved.

If r>1r>1, apply Theorem 1.1 with cofactor rr. For p<rp<r, part (i) gives strict inequality. For p>rp>r, part (ii) gives the maximum gaps of Φrp\Phi_{rp}. Writing δ=φ(r)\delta=\varphi(r), the two exponents of the first gap (jr−δ,jr)(jr-\delta,jr) both have the parity of jj, since rr is odd and δ\delta is even. Multiplication of both coefficients by (−1)j(-1)^j gives the first row. Reflection in degree DD reverses the coefficients and gives the second row. Distinctness and the range of jj follow from the support bijection. If r=1r=1, the same identity gives Φmp(x)=∑k=0p−1(−1)kxhk\Phi_{mp}(x)=\sum_{k=0}^{p-1}(-1)^kx^{hk}.

For squarefree even m>2m>2, one has h=1h=1 and r=m/2r=m/2. Thus equality already holds for m/2<p<mm/2<p<m, and the count is 2⌊2p/m⌋2\lfloor2p/m\rfloor. For p<rp<r, all gaps are multiples of hh, hence g(Φmp)≤d−hg(\Phi_{mp})\le d-h. In that range the actual maximum is still given by the reduction (4.1). For example, if r=qr=q is prime, apply Theorem 1.1(ii) with the smaller of p,qp,q as cofactor. The actual maximum is then h(min⁡(p,q)−1)h(\min(p,q)-1) and occurs 2⌊max⁡(p,q)/min⁡(p,q)⌋2\lfloor\max(p,q)/\min(p,q)\rfloor times.

Example 4.2. For m=30m=30, p=17p=17, we have r=15r=15, d=8d=8, and D=128D=128. The extreme pairs are (7,15)(7,15) and (113,121)(113,121), with coefficients (1,−1)(1,-1) and (−1,1)(-1,1), respectively. For m=60m=60 and the same prime, rr is unchanged and h=2h=2. The extreme pairs become (14,30)(14,30) and (226,242)(226,242) with the same coefficients. For m=30m=30, p=13p=13, the reduction to Φ195\Phi_{195} instead gives four maximum gaps of length 6 discussed after Theorem 1.1.

Zero runs in the periodic blocks

The partial-sum argument also identifies the blocks in the coefficient decomposition of [6] that can contain an extremal run of zeros. Continue with the notation and hypotheses of Theorem 1.1, and put D=d(p−1)D=d(p-1). Work in R=Q[x]/(xm−1)R=\mathbb{Q}[x]/(x^m-1) and put

Gi=U(x)Pi(xp) in R,0≤i<d.G_i=U(x)P_i(x^p)\text{ in }R,\qquad0\leq i<d.

For G∈RG\in R, let G[t]G[t] denote the coefficient at the residue of tt modulo mm in the representative of degree less than mm. Multiplication by xx rotates this list of mm coefficients. For 0≤t≤w≤D0\leq t\leq w\leq D, the coefficients in a short interval ending at ww satisfy

[xt]Φmp(x)=G⌊w/p⌋[t]if w−t≤d−1.(8)[x^t]\Phi_{mp}(x)=G_{\lfloor w/p\rfloor}[t]\qquad\text{if }w-t\leq d-1. \tag*{(8)}

Indeed, (2.2) and (3.1) give the formula with w=tw=t. Passing from ⌊t/p⌋\lfloor t/p\rfloor to ⌊w/p⌋\lfloor w/p\rfloor adds terms ujU[t−jp]u_jU[t-jp] with t<jp≤wt<jp\leq w. Each vanishes because 1≤jp−t≤d−11\leq jp-t\leq d-1 and deg⁡U=m−d\deg U=m-d. Thus (5.1) also holds when the interval crosses several multiples of pp.

Proposition 5.1. No GiG_i contains dd cyclically consecutive zeros. A run of d−1d-1 cyclically consecutive zeros occurs precisely for i=0i=0 and i=d−1i=d-1, once per period in each case. In G0G_0 it ends at residue −1-1, and in Gd−1G_{d-1} it ends at residue dp−1dp-1.

Proof. Since xm−1=−UΦmx^m-1=-U\Phi_m, cancellation in Q[x]\mathbb{Q}[x] gives

UA=0 in R⟺Φm∣A(A∈Q[x]).(9)UA=0\text{ in }R\quad\Longleftrightarrow\quad\Phi_m\mid A\qquad(A\in\mathbb{Q}[x]). \tag*{(9)}

In particular Gi≠0G_i\ne0, because Pi(ζp)≠0P_i(\zeta^p)\ne0 for a primitive mmth root ζ\zeta. The polynomial representative of every rotation of GiG_i is divisible by UU. Rotate a run of ℓ\ell zeros to the final ℓ\ell positions. The representative then has degree at most m−1−ℓm-1-\ell. Since deg⁡U=m−d\deg U=m-d, this excludes ℓ=d\ell=d. For ℓ=d−1\ell=d-1 the representative must equal cUcU for some c∈Q×c\in\mathbb{Q}^{\times}. Undoing the rotation yields Gi=cxsUG_i=cx^sU in RR. By (5.2), Pi(ζp)=cζsP_i(\zeta^p)=c\zeta^s. Lemma 2.2 forces i=0i=0 or i=d−1i=d-1.

The last d−1d-1 entries of G0=UG_0=U are zero. If a rotation also has these final zeros, it equals cUcU by the same degree bound. Equation (5.2) then gives ζs=c\zeta^s=c, so m∣sm\mid s. Thus the run in UU is unique. Finally, (2.2) and (5.2) give Gd−1=−xdpUG_{d-1}=-x^{dp}U in RR. Its unique run is therefore shifted by dpdp residues.

Blocks for an even cofactor

Let mm be even and p∤mp\nmid m prime, and use the notation of Section 4. Define UU, PiP_i, GiG_i with this even mm by the same formulas as above, and use the representative modulo xm−1x^m-1 to read each block.

Proposition 5.2. Let m>2m>2 be even. No GiG_i contains dd cyclically consecutive zeros. A run of length d−1d-1 occurs precisely for the indices

0≤i<hord−h≤i<d.0\leq i<h\qquad\text{or}\qquad d-h\leq i<d.

In each such block there are exactly two runs of length d−1d-1 per period of mm. In the first range they end at residues −1-1, L−1L-1, and in the second at dp−1dp-1, dp+L−1dp+L-1, all taken modulo mm. For a power of two the two ranges coincide.

Proof. The degree and cancellation argument in Proposition 5.1 applies to every m>1m>1. It excludes dd consecutive zeros, and a run of length d−1d-1 forces Gi=cxsUG_i=cx^sU with c∈Q×c\in\mathbb{Q}^{\times}. Hence Pi(ζp)=cζsP_i(\zeta^p)=c\zeta^s at any primitive mmth root ζ\zeta.

Suppose r>1r>1. As in (4.1) we have Φm(T)=Φr(−Th)\Phi_m(T)=\Phi_r(-T^h). Thus the prefix Pi(ζp)P_i(\zeta^p) is the prefix of Φr\Phi_r through index ⌊i/h⌋\lfloor i/h\rfloor, evaluated at −ζhp-\zeta^{hp}, a primitive rrth root. Its squared modulus is c2c^2. Lemma 2.2 forces ⌊i/h⌋\lfloor i/h\rfloor to be 00 or φ(r)−1\varphi(r)-1, which gives the asserted ranges. In the first range Pi=1P_i=1 and Gi=UG_i=U. In the second Pi=Φm−TdP_i=\Phi_m-T^d, whence Gi=−xdpUG_i=-x^{dp}U by (5.2) and Φm(x)∣Φm(xp)\Phi_m(x)\mid\Phi_m(x^p).

The last d−1d-1 positions of UU vanish. A rotation with the same final zeros must equal a rational multiple of UU, by the degree argument. Cancellation implies ζs∈Q\zeta^s\in\mathbb{Q} for its shift ss, so s=0,Ls=0,L (mod mm). The second shift occurs because Φm∣xL+1\Phi_m\mid x^L+1 gives xLU≡−Ux^LU\equiv-U modulo xm−1x^m-1. This gives the two listed end residues for UU and their translates for −xdpU-x^{dp}U.

Finally, if r=1r=1, then m=2hm=2h, d=hd=h, Φm=1+xh\Phi_m=1+x^h and every Gi=U=1−xhG_i=U=1-x^h. The two runs end immediately before 00 and hh. As dp=hp≡h(mod2h)dp=hp\equiv h\pmod{2h}, both lists of end residues agree.

:::

For m=2m = 2 the only block is 1−x1 - x; both cyclic positions are nonzero.

Declaration

This paper was written by the author in collaboration with GPT 6 Astra.

References

  1. [1]H. Hong, E. Lee, H.-S. Lee and C.-M. Park, Maximum gap in (inverse) cyclotomic polynomial, J. Number Theory 132 (2012), 2297–2315.arxiv.org/abs/1101.4255
  2. [2]O.-M. Camburu, E.-A. Ciolan, F. Luca, P. Moree and I. E. Shparlinski, Cyclotomic coefficients: gaps and jumps, J. Number Theory 163 (2016), 211–237.arxiv.org/abs/1507.07098
  3. [3]B. Zhang, Remarks on maximum gap in binary cyclotomic polynomials, Bull. Math. Soc. Sci. Math. Roumanie 59(107) (2016), no. 1, 109–115.
  4. [4]M. Ambrosino, H. Hong and E. Lee, Lower bounds for maximum gap in (inverse) cyclotomic polynomials, arXiv:1702.07650 (2017).arxiv.org/abs/1702.07650
  5. [5]B. Zhang, Maximum gaps in a class of ternary cyclotomic polynomials, Adv. Math. (China) 48 (2019), 419–429.DOI
  6. [6]A. Al-Kateeb, M. Ambrosino, H. Hong and E. Lee, Maximum gap in cyclotomic polynomials, J. Number Theory 229 (2021), 1–15.arxiv.org/abs/1911.11667
  7. [8]A. Benoist and A. Perucca, The ternary cyclotomic polynomials Φ₃pq, Involve 19 (2026), no. 1, 51–72.DOI

Paper details

vlnc:W_s9qdcnmb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents