Wiki
Wiki

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

Updated


Source. Section 6, pp. 8--10, with Lemma 6.1 and Theorem 6.2 on p. 9 and Remark 6.4 on p. 10, and Appendix A, pp. 10--14, of Venkata Siddharth Pendyala, Radial access for polynomial lemniscates: the cubic theorem and a degree-six obstruction, SSRN preprint (2026), doi:10.2139/ssrn.6850818, the edition named on the source card.

Statement

Admissible polynomials, safe radii, Δ\Delta, B\mathcal B and N∗N^* are as on the Theorem 1.1 page.

The polynomial (pp. 8--9). Let

ρ=19999992000000,S=(927,927,317,317,697,547),qρ(z)=∏s∈S(ρ+e−2πis/960z),\rho=\frac{1999999}{2000000},\qquad S=(927,927,317,317,697,547),\qquad q_\rho(z)=\prod_{s\in S}\bigl(\rho+e^{-2\pi is/960}z\bigr),

the product running over SS with its repetitions, so that

qρ(z)=(ρ+e−2πi 927/960z)2(ρ+e−2πi 317/960z)2(ρ+e−2πi 697/960z)(ρ+e−2πi 547/960z).q_\rho(z)=\bigl(\rho+e^{-2\pi i\,927/960}z\bigr)^2\bigl(\rho+e^{-2\pi i\,317/960}z\bigr)^2 \bigl(\rho+e^{-2\pi i\,697/960}z\bigr)\bigl(\rho+e^{-2\pi i\,547/960}z\bigr).

Its leading coefficient is ∏s∈Se−2πis/960=e−2πi 3732/960=e−2πi 71/80\prod_{s\in S}e^{-2\pi is/960}=e^{-2\pi i\,3732/960}=e^{-2\pi i\,71/80}, so

Pρ(z)=e2πi 71/80 qρ(z)P_\rho(z)=e^{2\pi i\,71/80}\,q_\rho(z)

is monic of degree six, and ∣Pρ∣=∣qρ∣|P_\rho|=|q_\rho|. Its zeros are −ρe2πis/960-\rho e^{2\pi is/960} for s∈Ss\in S: double at s=927s=927 and s=317s=317, simple at s=697s=697 and s=547s=547, all of modulus ρ<1\rho<1; and ∣Pρ(0)∣=ρ6<1|P_\rho(0)|=\rho^6<1. As fractions of a turn, their directions are 447/960447/960 and 797/960797/960 for the double zeros and 217/960217/960 and 67/96067/960 for the simple ones, the four directions Remark 6.4 names (p. 10). For z=re2πiuz=re^{2\pi iu} the paper's (6.1) is

∣qρ(re2πiu)∣2=∏s∈S(ρ2+r2+2ρrcos⁡2π(u−s960)).|q_\rho(re^{2\pi iu})|^2=\prod_{s\in S}\Bigl(\rho^2+r^2+2\rho r\cos2\pi\bigl(u-\tfrac{s}{960}\bigr)\Bigr).

Lemma 6.1 (Three-radius certificate, p. 9). Let R={1,1/300,439/1000}R=\{1,1/300,439/1000\}. For every u∈[0,1]u\in[0,1],

max⁡r∈R∣qρ(re2πiu)∣>1,\max_{r\in R}\lvert q_\rho(re^{2\pi iu})\rvert>1,

and, more quantitatively,

min⁡0≤u≤1 max⁡r∈R∣qρ(re2πiu)∣2>1+2.327036066010500056098356⋅10−6.\min_{0\le u\le1}\ \max_{r\in R}\lvert q_\rho(re^{2\pi iu})\rvert^2>1+2.327036066010500056098356\cdot10^{-6}.

Theorem 6.2 (Strict degree-six radial obstruction, p. 9). There is a monic polynomial PρP_\rho of degree six, with all its zeros in the open disk D\mathbb D, such that

min⁡θ∈R max⁡0≤r≤1∣Pρ(reiθ)∣>1.\min_{\theta\in\mathbb R}\ \max_{0\le r\le1}\lvert P_\rho(re^{i\theta})\rvert>1 .

Consequently N∗≤6N^*\le6. The polynomial is the PρP_\rho above: every radius carries a point at distance 11, 1/3001/300 or 439/1000439/1000 from 00 where ∣Pρ∣>1|P_\rho|>1, so no radius is safe, B(Pρ)>1\mathcal B(P_\rho)>1, and Proposition 2.2 gives Δ(Pρ)>0\Delta(P_\rho)>0 (p. 10).

The certificate

Appendix A (pp. 10--12) states the certificate for Lemma 6.1 as a finite datum and checks it by rational inequalities.

  • Intervals. [0,1][0,1] is cut into Ij=[j/8192,(j+1)/8192]I_j=[j/8192,(j+1)/8192] for 0≤j<81920\le j<8192, and each IjI_j gets a radius rj∈Rr_j\in R from Table 1 (p. 11), which the paper calls the certificate; no search or numerical optimization enters the check.
  • Table 1, the blocks of jj with their radius: 00 to 363363, radius 11; 364364 to 795795, 439/1000439/1000; 796796 to 16271627, 11; 16281628 to 20582058, 439/1000439/1000; 20592059 to 33173317, 11; 33183318 to 42604260, 1/3001/300; 42614261 to 63536353, 11; 63546354 to 72967296, 1/3001/300; 72977297 to 81918191, 11. The outer radius 11 blocks every direction except four short corridors centered at the zero directions; the middle radius 439/1000439/1000 closes the corridors at 67/96067/960 and 217/960217/960 of the simple zeros, and the inner radius 1/3001/300 those at 447/960447/960 and 797/960797/960 of the double zeros (Remark 6.4, p. 10).
  • Cosine bound. For a phase ss let Jj,s=Ij−s/960J_{j,s}=I_j-s/960 and let δj,s\delta_{j,s} be the distance from Jj,sJ_{j,s} to Z+1/2\mathbb Z+1/2. If δj,s=0\delta_{j,s}=0 then cj,s=−1c_{j,s}=-1; otherwise cj,s=max⁡{−1,−U(2π0δj,s)}c_{j,s}=\max\{-1,-U(2\pi_0\delta_{j,s})\}, with π0=103993/33102<π\pi_0=103993/33102<\pi and U(x)=∑k=010(−1)kx2k/(2k)!+x21/21!U(x)=\sum_{k=0}^{10}(-1)^kx^{2k}/(2k)!+x^{21}/21!. Each cj,sc_{j,s} is rational, and cos⁡2π(u−s/960)≥cj,s\cos2\pi(u-s/960)\ge c_{j,s} for u∈Iju\in I_j, the paper's (A.1).
  • Inequalities. With multiplicities (m927,m317,m697,m547)=(2,2,1,1)(m_{927},m_{317},m_{697},m_{547})=(2,2,1,1), let Mj=∏s∈{927,317,697,547}max⁡{0,ρ2+rj2+2ρrjcj,s}ms−1M_j=\prod_{s\in\{927,317,697,547\}}\max\{0,\rho^2+r_j^2+2\rho r_jc_{j,s}\}^{m_s}-1. By (6.1) and (A.1), Mj>0M_j>0 gives ∣qρ(rje2πiu)∣2>1|q_\rho(r_je^{2\pi iu})|^2>1 for all u∈Iju\in I_j, so Lemma 6.1 reduces to the 8192 rational inequalities Mj>0M_j>0.
  • Margins. Table 2 (p. 12) lists the least MjM_j in each block. The least of all is at (j,r)=(3318,1/300)(j,r)=(3318,1/300), in block 6, an exact rational with a numerator of 1289 digits and a denominator of 1295 digits; the paper states that the smallest MjM_j is at least 2.327036066010500056098356⋅10−62.327036066010500056098356\cdot10^{-6}. The other block minima run from about 2.6⋅10−62.6\cdot10^{-6} (block 8) to about 1.5⋅10−21.5\cdot10^{-2} (blocks 2 and 4). The paper calls its displayed decimals renderings of exact positive rationals.
  • Checker. Pages 12--14 print a Wolfram Language program that checks that the blocks cover j=0,…,8191j=0,\dots,8191, computes every MjM_j in exact rational arithmetic, stops if some Mj≤0M_j\le0, and prints the block minima with SHA-256 hashes of their exact values; floating point enters only the displayed decimals (p. 12). It was not run for this page.

Read depth. Claims checked: the construction, (6.1), Lemma 6.1, Theorem 6.2, Remark 6.4 and the definitions and Tables 1 and 2 of Appendix A were read clause by clause on the page images. Only the certificate's bookkeeping was recomputed here, in exact arithmetic and without the printed checker: the leading coefficient, the zero directions, which match Remark 6.4, and that the blocks of Table 1 cover j=0,…,8191j=0,\dots,8191 and give the radius 439/1000439/1000 to the intervals containing 67/96067/960 and 217/960217/960 and the radius 1/3001/300 to those containing 447/960447/960 and 797/960797/960. The derivation of (A.1) was read and not checked, none of the 8192 inequalities was checked, and the printed checker was not run. The paper is an unrefereed preprint, and nothing here is independently reviewed.

Proof pointer

Theorem 6.2, pp. 9--10. PρP_\rho is monic and admissible by construction. Since ∣Pρ∣=∣qρ∣|P_\rho|=|q_\rho|, Lemma 6.1 gives max⁡0≤r≤1∣Pρ(re2πiu)∣>1\max_{0\le r\le1}|P_\rho(re^{2\pi iu})|>1 for every uu, so no radius lies in {∣Pρ∣≤1}\{|P_\rho|\le1\}; then B(Pρ)>1\mathcal B(P_\rho)>1, and Proposition 2.2 gives Δ(Pρ)>0\Delta(P_\rho)>0. Lemma 6.1 (p. 9) follows from the certificate above, since the intervals IjI_j cover [0,1][0,1]. For (A.1) the paper uses that the cosine is symmetric about half-integers and decreasing on [0,π][0,\pi], that π0<π\pi_0<\pi, and Taylor's theorem for the cosine with remainder at most x21/21!x^{21}/21! (pp. 10--11).

Dependencies

Lemma 6.1 (p. 9) with Appendix A (pp. 10--14), and the harder half of Proposition 2.2 (p. 2), stated on the Theorem 1.1 page: zero detour forces a safe radius.

Bears on

  • Problem 1120: in the problem page's worst-case reading S(n)S(n), the theorem gives S(6)≥L(Pρ)>1S(6)\ge L(P_\rho)>1, so in degree six the trivial bound 11 is not attained. It gives no lower bound for L(Pρ)−1L(P_\rho)-1, does not say whether L(Pρ)L(P_\rho) is finite, and says nothing about other degrees.