Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Short-arc flat root relations and what they force

../


An explicit degree-49 flat relation

Bzdęga's jump-one theorem for ternary cyclotomic coefficients, Theorem 1.5 of Bounds on ternary cyclotomic coefficients (printed page 6, proof in Section 5, pages 14--15, checked in the held copy), implies that (1−x)Φpqr(x)(1-x)\Phi_{pqr}(x), and hence (1+x)Φ2pqr(x)(1+x)\Phi_{2pqr}(x), are flat for distinct odd primes p,q,rp,q,r.

The particular polynomial F=(1+x)Φ210F=(1+x)\Phi_{210} has positive exponents

{0,3,5,7,8,10,12,37,39,41,42,44,46,49}\{0,3,5,7,8,10,12,37,39,41,42,44,46,49\}

and negative exponents

{18,20,21,22,23,24,25,26,27,28,29,31}.\{18,20,21,22,23,24,25,26,27,28,29,31\}.

Its elementary exact certificate is

F(x)(1+x15)(1+x21)(1+x35)=(1+x105)(1+x3)(1+x5)(1+x7).F(x)(1+x^{15})(1+x^{21})(1+x^{35}) =(1+x^{105})(1+x^3)(1+x^5)(1+x^7).

Evaluation at a primitive 210th root is legitimate because the three factors on the left other than FF are nonzero there, and 1+x105=01+x^{105}=0. This gives a 26-point signed relation in an arc of angular length 2π⋅49/210=7π/152\pi\cdot49/210=7\pi/15. Polynomial multiplication checks the identity exactly; it is short exposed arithmetic and needs no separate checker.

Degree 49 is minimal for a nonzero flat multiple of Φ210\Phi_{210}: deg⁡Φ210=48\deg\Phi_{210}=48, its height is two, and a degree-48 multiple would be a scalar multiple with scalar ±1\pm1, by its constant coefficient. The same degree argument applies to Φ105\Phi_{105}. This is a finite fact, not an o(n)o(n)-degree family. Within the displayed ternary-derived even-conductor family, the degree/conductor ratio is minimized at (p,q,r)=(3,5,7)(p,q,r)=(3,5,7), giving 49/21049/210: (φ(pqr)+1)/(2pqr)(\varphi(pqr)+1)/(2pqr) increases in each prime separately.

What a short-arc relation actually forces

For a proper coloring c:Z/n→[q]c:\mathbb Z/n\to[q], put ρc(t)=n−1∣{x:c(x+t)≠c(x)}∣\rho_c(t)=n^{-1}|\{x:c(x+t)\ne c(x)\}|. If SS supports a signed relation and TT is any spanning tree on SS, then

∑{a,b}∈Tρc(b−a)≥1.\sum_{\{a,b\}\in T}\rho_c(b-a)\geq1.

Every translated support is nonmonochromatic, so at least one of its tree edges is bichromatic; averaging proves the inequality. If S⊂[0,d]S\subset[0,d] and B=nρc(1)B=n\rho_c(1), ordering its exponents and using ρc(t)≤tB/n\rho_c(t)\leq tB/n gives B≥n/dB\geq n/d. Thus hypothetical arbitrarily short flat relations force unbounded angular color variation, not impossibility of arbitrary finite coloring.

Neither a finite palette nor averaging implies angular continuity. Write a root as e2πixe^{2\pi ix}, let x2∈[0,1)x_2\in[0,1) be its 2-primary component, and color by ⌊2x2⌋\lfloor2x_2\rfloor. For every odd prime pp, e−πi/p→1e^{-\pi i/p}\to1 has 2-primary component 1/21/2, so multiplication by it flips every color. This example is not asserted proper; it refutes only the proposed automatic continuity inference. Bare measurability also adds nothing on this countable group: every coloring has a Borel-measurable extension to the circle, assigning all nonroots one color.

No o(n)o(n)-degree flat family or unrestricted-coloring contradiction was proved on this route.

Exact ranges of the exploratory searches

All runs below were bounded finite searches and supplied no new negative-solution ingredient. No script is retained.

  • For n=105,1155,15015n=105,1155,15015, exhaustively tested Φn(x)∏p∈U(1−xp)(1−x)ϵ\Phi_n(x)\prod_{p\in U}(1-x^p)(1-x)^\epsilon, with U⊆{p:p∣n}U\subseteq\{p:p\mid n\}, ϵ∈{0,1}\epsilon\in\{0,1\}. Minimum heights were respectively 1, 3, 14. The next conductor 255255 was interrupted during polynomial construction after the completed cases ceased to discriminate, and has no search conclusion.
  • At n=1155n=1155, one integer-feasibility experiment used multiplier degree at most 48, constant coefficient 1, all remaining multiplier coefficients bounded in magnitude by 1000, product coefficients constrained to [−1,1][-1,1], and a 20-second solver limit. It timed out without a feasible point. This is not infeasibility.
  • At n=1155n=1155, tested all 465 squarefree products of proper cyclotomic factors of total multiplier degree at most 58. Eligible divisors were 1,3,5,7,11,15,21,33,35,55,1051,3,5,7,11,15,21,33,35,55,105. None gave a flat product. The minimum height was two, attained by Φ1Φ3Φ5\Phi_1\Phi_3\Phi_5, degree seven.
  • Consecutive-power differences (1−xs)Φm(x)(1-x^s)\Phi_m(x) for m∈{105,165,195,1155}m\in\{105,165,195,1155\} and s∈{1,2,3,5,7}s\in\{1,2,3,5,7\}, a fixed total of 24 polynomials, were formed exactly to test whether a consecutive-power recurrence could yield a useful bounded-height mandatory root configuration. Height means the maximum absolute coefficient.
mmDegree of Φm\Phi_mHeightDifference heights, in the displayed ss-order
1054821, 1, 2, 3, 3
1658021, 2, 3, 3, 3
1959621, 2, 3, 3, 2
115548036, 4, 4, 4, 4

The low-height multiples did not supply a general recurrence construction; the prime staircase supplied the needed circuits directly. These numbers are recorded observations, not an asymptotic assertion.