Overview

  • Which reciprocal factors must be separated in a large-gap theorem?
  • What information is preserved when a factor is reversed?
  • Under what hypotheses is the nonreciprocal part eventually irreducible?

Large gaps and reciprocal factors

Consider the family

fN(x)=xNc(x−1)+d(x),N>deg⁡c+deg⁡d,f_N(x)=x^N c(x^{-1})+d(x),\qquad N>\deg c+\deg d,

where c,d∈Z[x]c,d\in\mathbb Z[x] are fixed and c(0)d(0)≠0c(0)d(0)\ne0. The high and low blocks of coefficients are separated by a gap. Write h∗(x)=xdeg⁡hh(1/x)h^*(x)=x^{\deg h}h(1/x). A nonconstant irreducible factor is reciprocal when h∗=±hh^*=\pm h; the nonreciprocal part is obtained by removing all such irreducible factors, including their multiplicities.

Cyclotomic polynomials are reciprocal up to sign, but other reciprocal factors can occur. Thus irreducibility of the nonreciprocal part proves irreducibility of the original polynomial only after all reciprocal factors have been excluded.

References 2

Schinzel’s theorem

Call (c,d)(c,d) Capellian if −d(x)/c(x−1)-d(x)/c(x^{-1}) is a pp-th power in Q(x)\mathbb Q(x) for some prime pp, or if d(x)/c(x−1)d(x)/c(x^{-1}) is four times a fourth power. These power conditions lead to factorizations for infinitely many exponents.

Theorem · Schinzel, in large-gap form

Assume c(0)d(0)≠0c(0)d(0)\ne0, gcd⁡Z[x](c∗,d)=1\gcd_{\mathbb Z[x]}(c^*,d)=1, c≠±dc\ne\pm d, and that (c,d)(c,d) is not Capellian. Then there is an effective bound N0N_0, depending on c,dc,d, such that for N>N0N>N_0 the nonreciprocal part of fNf_N is irreducible over Q\mathbb Q.

This is the formulation in Sawin–Shusterman–Stoll, Theorem 1.2, deduced there from Schinzel’s work. Filaseta–Ford–Konyagin improve explicit thresholds. Sawin–Shusterman–Stoll give further bounds and algorithms under an additional robustness hypothesis (Theorems 1.5–1.6).

References 3

Factor reversal

For f(0)≠0f(0)\ne0 and a factorization f=ghf=gh, form G=gh∗G=gh^*. The reciprocal product does not change:

GG∗=ff∗.GG^*=ff^*.

Equivalently, as Laurent polynomials, one may use g(x)h(x−1)g(x)h(x^{-1}); multiplication by a power of xx only changes the presentation. On the unit circle, a real-coefficient polynomial satisfies ∣h∗(z)∣=∣h(z)∣|h^*(z)|=|h(z)|. Consequently GG and ff have the same squared coefficient norm, by Parseval's identity:

∑j∣[xj]G∣2=12π∫02π∣G(eit)∣2 dt=∑j∣[xj]f∣2.\sum_j |[x^j]G|^2=\frac1{2\pi}\int_0^{2\pi}|G(e^{it})|^2\,dt=\sum_j |[x^j]f|^2.

For a sparse polynomial with small integral coefficients, the right side is small. Only finitely many multisets of nonzero integer coefficients have that norm, although their positions may vary. Comparing the full identity GG∗=ff∗GG^*=ff^* then restricts the possible positions as well as the coefficients. In suitable sparse families, these restrictions exclude nonreciprocal factorizations. Ljunggren’s method and its later extensions use this restriction.

References 3

Example

Example · xN+x+1x^N+x+1

Let ζ\zeta be a primitive cube root of unity. Since ζ+1=−ζ2\zeta+1=-\zeta^2,

ζN+ζ+1=0⟺N≡2(mod3).\zeta^N+\zeta+1=0\quad\Longleftrightarrow\quad N\equiv2\pmod3.

Thus x2+x+1x^2+x+1 divides xN+x+1x^N+x+1 exactly in that residue class. In degree 55, for example,

x5+x+1=(x2+x+1)(x3−x2+1).x^5+x+1=(x^2+x+1)(x^3-x^2+1).

The cubic has no rational root, hence is irreducible over Q\mathbb Q. Thus infinitely many members are reducible because of a recurring cyclotomic factor.

The periodicity is general for a fixed cyclotomic factor: evaluation at a root of unity depends on NN modulo its order. The large-gap literature combines this elementary observation with bounds that control other factors.

References 2
  • Ljunggren1960Sparse trinomial factorization and reciprocal obstructions; the displayed root-of-unity calculation is given here.Source
  • sawinshustermanstoll2020 · Corollary 1.4Effective eventual description by finitely many exceptional exponents and arithmetic progressions.Source

A family with an exceptional degree

Sawin–Shusterman–Stoll prove that xN−7x2+1x^N-7x^2+1 is irreducible for every N≥5N\ge5, and also for N=3N=3. The degree 44 is exceptional:

x4−7x2+1=(x2−3x+1)(x2+3x+1).x^4-7x^2+1=(x^2-3x+1)(x^2+3x+1).

Their general theory supplies an eventual threshold, and a finite check handles the remaining exponents in this example.

References 1

Questions

OpenIrreducibility methods

How small can a uniform large-gap threshold be?

Scope. Pairs satisfying the hypotheses of the large-gap theorem; degree sum δ and squared coefficient weight at most w

Sawin–Shusterman–Stoll define Nopt(δ,w)N_{\mathrm{opt}}(\delta,w) as the smallest uniform threshold for their Theorem 1.2 over pairs with deg⁡c+deg⁡d=δ\deg c+\deg d=\delta and ∥c∥22+∥d∥22≤w\|c\|_2^2+\|d\|_2^2\le w. Their Question 1.7 asks, for fixed δ>1\delta>1, whether this is Oδ(log⁡w)O_\delta(\log w); Question 1.8 asks for the uniform bound O(δlog⁡w)O(\delta\log w). This records the questions as posed in that paper; it is not a claim that all later work has been excluded.

References 1
Discuss this question →

Related topics

References

  1. Polynomials with Special Regard to Reducibility

    Schinzel, Andrzej. Cambridge University Press, Encyclopedia of Mathematics and its Applications, vol. 77, 2000.

    Read the source
  2. Irreducibility of polynomials with a large gap

    Sawin, Will and Shusterman, Mark and Stoll, Michael. Acta Arithmetica, 192(2), 111–139, 2020.

    Read the source
  3. On an irreducibility theorem of A. Schinzel associated with coverings of the integers

    Filaseta, Michael and Ford, Kevin and Konyagin, Sergei. Illinois Journal of Mathematics, 44(3), 633–643, 2000.

    Read the source
  4. On the irreducibility of certain trinomials and quadrinomials

    Ljunggren, Wilhelm. Mathematica Scandinavica, 8, 65–70, 1960.

    Read the source
  5. On the factorization of polynomials with small Euclidean norm

    Filaseta, Michael. Number Theory in Progress, Vol. 1, de Gruyter, 143–163, 1999.

    Read the source