Wiki
Wiki

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

Updated

Erdős 699: a numerical stopping bound for index seven


This is a retained earlier research note. The current reading guide gives the latest status and distinguishes completed proofs from remaining cases.

P699 research · Erdős problem 699 · Original effective proof

Theorem. Every bad triple (n,7,j)(n,7,j) for the original threshold p≥7p\ge7 satisfies

n<103 906 174 477 003.\boxed{n<10^{3\,906\,174\,477\,003}.}

All constants are numerical. The aggregate checker and certificate combine the sources and polynomial coverage. A separate finite exclusion covers the lower range. The subsequent interval covers join these ranges and complete the case.

Complete result. The combined proof excludes every admissible bad triple at seven. The source and polynomial estimates below are retained as components of that proof.

Initial pair alternatives

Put h=n−6h=n-6. Choose rp∈{0,…,6}r_p\in\{0,\ldots,6\} maximizing vp(n−rp)v_p(n-r_p) for each p∈{2,3,5}p\in\{2,3,5\}, and write n−rp=Ppupn-r_p=P_pu_p, ep=log⁡up/log⁡he_p=\log u_p/\log h. The initial alternatives have the form ep>αe_p>\alpha or eq>βe_q>\beta. Their endpoints are smaller by 1/100001/10000 than in the original proof.

(p,q)(p,q)(α,β)(\alpha,\beta)Sufficient exponent KK for h≥10Kh\ge10^K
(2,3)(2,3)(.2920,.2920)(.2920,.2920)1674735
(2,3)(2,3)(.3295,.2915)(.3295,.2915)3990256
(2,5)(2,5)(.2675,.2710)(.2675,.2710)922680314061
(2,5)(2,5)(.2678,.2678)(.2678,.2678)2790743305738
(3,5)(3,5)(.2561,.3773)(.2561,.3773)19729939

The initial-source checker retains a rational upper bound for each positive-part denominator cost, a uniform coefficient-gcd estimate, and all finite Padé constants. Completed finite gcd certificates improve three of the starting bounds. The symmetric (2,5)(2,5) source uses seed 128−125=3128-125=3 and directions (79,18,79)(79,18,79). Its bound is the largest initial-source contribution.

The 258 joint sources

The original clauses are

∑p(gp+bpep)+≥R.\sum_p(g_p+b_pe_p)_+\ge R.

Their intercepts, sign regions and block parameters tt are retained; the right sides are replaced by R−ηR-\eta. Usually η=1/10000\eta=1/10000. The five exceptions are stored exactly in the relaxed coverage certificate. For each of the 210 distinct direction triples, choose a rational gcd rate ℓ\ell at least the maximum of

ℓold−3η4t\ell_{\mathrm{old}}-\frac{3\eta}{4t}

over its sources. This preserves at least η/4\eta/4 of additional analytic margin in every clause. The uniform gcd certificates use the sharper Chebyshev estimates and have numerical starting indices m0≤1012m_0\le10^{12}.

The explicit finite-constant argument then evaluates every joint source. The 258 numerical records include the seed, directions, weakened target, gcd starting index, integral prefactors and directed decay comparisons. Before adjusting the witness block, their largest height exponent was

Koriginal=20219858230830.K_{\rm original}=20219858230830.

It comes from source_361, with directions (10,5,11)(10,5,11), t=6689537/311451090t=6689537/311451090, m0=1012m_0=10^{12}, and η=6367/109\eta=6367/10^9. Pair gaps are bounded by six and degree-two product gaps by 36. Actual gap zero is included by the same integer argument.

Adjusting the witness blocks

The block adjustment lemma reduces the residual denominator cost when a smaller Padé block is chosen. The original intercepts and weakened thresholds stay fixed. The adjusted source certificates and checker recompute all 210 uniform gcd bounds and all 258 numerical source bounds. Seventy-five analytic margins and 180 individual stopping bounds improve.

The new common joint-source exponent is

Kjoint=3906174477002.K_{\rm joint}=3906174477002.

The maximum now comes from source_039, with directions (20,9,20)(20,9,20), block parameter 7705356/6930427677705356/693042767, and m0=1011m_0=10^{11}. The former maximum, source_361, falls to exponent 40476804424047680442.

Polynomial coverage and the final bound

The exact relaxed forest covers all 540 original position and minimal-cofactor cases, using 4336 nodes and 3388 rational Farkas leaves. It verifies the weaker initial endpoints and all joint weakenings together. Its weighted coefficient requirement is

40482625000000404863<99990923.\frac{40482625000000}{404863}<99990923.

The Newton-basis coefficient bounds and arithmetic nonvanishing therefore give the polynomial starting height h≥1099990923h\ge10^{99990923}. Coincident positions and the central column are covered by the original reductions. Existing interpolation ranks are reused.

The joint-source exponent exceeds the initial-source and polynomial exponents. Consequently badness implies n−6<103906174477002n-6<10^{3906174477002}. Adding one decimal power absorbs the shift and proves the theorem.

The aggregate names every input by basename and loads it, matches all source identifiers and targets, checks every exact Farkas leaf, and lists its inputs by basename in its own output as a run record, not as a condition for running. Run

sh
.venv/bin/python wiki/research/erdos_699/computations/erdos_699_index_7_numerical_tail.py

The additional --sources option regenerates all 210 analytic gcd bounds, all 258 original and adjusted joint thresholds, and the five initial source thresholds. It does not repeat interpolation ranks or enumerate binomial rows.