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 simpler effective proof at index twelve

P699 research · The first effective proof · Off-diagonal source estimates


The new off-diagonal estimates simplify the effective index-twelve proof to one explicit polynomial. Let XX be the largest effective threshold for the twenty prime-pair alternatives in the certificate. Then

n<max⁡{10845,X+11}for every bad triple at i=12.\boxed{n<\max\{10^{845},X+11\}\qquad\text{for every bad triple at }i=12.}

The number XX is effectively computable but has not been numerically evaluated. Thus 1084510^{845} is the algebraic part of the bound, not a claim that the whole finite range has been excluded.

The subsequent numerical proof and finite certificate complete this argument: index twelve has no bad triples.

One polynomial and a positive exponent

Use the explicit coordinate-line product

F8(X,Y)=∏a=07((X−a)(Y−a))8−a.F_8(X,Y)=\prod_{a=0}^{7}((X-a)(Y-a))^{8-a}.

It has degree 7272, is positive at every admissible (j,n−j)(j,n-j), and has multiplicity at least

mt=16−t(0≤t<12)m_t=16-t\qquad(0\le t<12)

at every nonnegative grid point on row x+y=tx+y=t. The total multiplicity is M=126M=126, so M−deg⁡F8=54M-\deg F_8=54.

The twenty off-diagonal pair alternatives give eighty-two minimal cofactor witnessing vectors in prime order 2,3,5,7,112,3,5,7,11. For a singleton selected position, the capacity is 1−vp1-v_p. For a group of two or more primes at the same position, its product of prime powers divides n−rn-r and has capacity at most one.

The exact checker regenerates all witness vectors and the resulting 402 capacity lists. As the multiplicities decrease with tt, rearrangement puts the largest capacities at the first available rows. In every case,

∑tmtct≤53.9.\sum_t m_t c_t\le53.9.

The exponent gained in the weighted divisibility inequality is therefore at least 54−53.9=1/1054-53.9=1/10. This covers all positions and all collisions, without an interpolation rank calculation.

The twenty alternatives are weakenings of individual sources in the directed off-diagonal table. The checker revalidates each selected source and its strict endpoint margin. Their two endpoint exponents sum to less than one; if the selected positions coincide, upuq≥n−ru_pu_q\ge n-r proves the corresponding alternative directly. The zero-gap case is consequently included too.

The algebraic bound

The exact off-maximum valuation constant is

C∗=21023555277161116.C_*=2^{102}3^{55}5^{27}7^{16}11^{16}.

The weighted-ramp inequality, with its bounded row-shift corrections, gives

n1/10≤1000999C∗272(n≥1 430 000),n^{1/10}\le\frac{1000}{999}\frac{C_*}{2^{72}} \qquad(n\ge1\,430\,000),

once the cofactor sources hold. The checker verifies the integer comparison

10845(999⋅272)10>(1000C∗)10.10^{845}(999\cdot2^{72})^{10}>(1000C_*)^{10}.

This proves the stated bound. The earlier twelve-certificate proof remains valid, but its algebraic bound 103 643 130 00110^{3\,643\,130\,001} is not needed. Both arguments leave the separate task of evaluating source thresholds and excluding the finite range.

sh
.venv/bin/python wiki/research/erdos_699/computations/erdos_699_index_12_ramp.py