Wiki
Wiki

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

Updated

Erdős 699: exact polynomial coverage of all index-9 position assignments


Complete follow-up. The index-nine proof closes all 917 residual root cases below and makes the entire index effective.

The certificate catalog and checker cover all 949^4 labeled assignments of the maximal positions of primes 2,3,5,72,3,5,7, including coinciding positions. For each assignment, they check all sixteen minimal cofactor vectors supplied by the retained prime-pair estimates.

Of the resulting 104,976 cases, 104,059 have a positive integer polynomial certificate. The remaining 917 labeled cases are recorded explicitly. Twelve have coinciding positions. These are finite configurations of cofactor exponents and positions, not a finite list of possible bad triples, and this calculation alone does not prove an effective bound at index 9.

The exact gain

Put H=n−8H=n-8, and choose rp∈{0,…,8}r_p\in\{0,\ldots,8\} maximizing vp(n−rp)v_p(n-r_p) for each p∈{2,3,5,7}p\in\{2,3,5,7\}. Write

n−rp=Ppup,Pp=pvp(n−rp).n-r_p=P_pu_p,\qquad P_p=p^{v_p(n-r_p)}.

The source graph combines the original symmetric BFT bounds with the retained asymmetric pair table. Its disjunctions imply, beyond an effectively computable common threshold, that one of the sixteen recorded vectors vv satisfies up≥Hvp/10000u_p\ge H^{v_p/10000} for every prime. The checker reconstructs these vectors from the canonical source tables; it does not read discovery outputs.

For each row tt, set

αt={1,no selected prime has position t,vp/10000,only prime p has position t,0,at least two selected primes have position t.\alpha_t=\begin{cases} 1,&\text{no selected prime has position }t,\\ v_p/10000,&\text{only prime }p\text{ has position }t,\\ 0,&\text{at least two selected primes have position }t. \end{cases}

The last case retains the bound ∏p:rp=tPp≤n−t\prod_{p:r_p=t}P_p\le n-t for the whole prime group. For nonnegative integer line exponents hah_a and row multiplicities mtm_t, a certificate of interpolation degree dd has total degree

D=2∑aha+max⁡{0,d+1}.D=2\sum_a h_a+\max\{0,d+1\}.

Its exact gain is

∑t=08αtmt−D.(1)\boxed{\sum_{t=0}^8\alpha_tm_t-D.} \tag{1}

The checker requires (1) to be strictly positive, with all comparisons performed at denominator 10000. The least first-use gain among the covered cases is 3/100003/10000. Prime labels are preserved throughout; there is no inference from sorted cofactor vectors to particular prime assignments.

Polynomial construction and ranks

Each record requires residual multiplicity

Rxy=max⁡{0,mx+y−hx−hy}R_{xy}=\max\{0,m_{x+y}-h_x-h_y\}

at every nonnegative integer point x+y≤8x+y\le8. Its symmetric jet matrix on polynomials of degree at most dd has full row rank modulo 251. All seventy records were checked by exact modular rank calculations. There are 95,014 jet rows in total; the largest matrix has 9,016 rows and 9,025 columns. Two records need no residual interpolation: their coordinate-line products already supply all multiplicities, and they use the convention d=−1d=-1.

The surjective interpolation lemma therefore produces a finite family of integer polynomials of total degree at most DD, all with the prescribed row multiplicities, and with at least one nonzero evaluation at each admissible (j,n−j)(j,n-j). The common coordinate-line factor is nonzero because both coordinates exceed 9.

The weighted divisibility argument then gives, for a suitable effectively computable constant CC,

n∑tαtmt−D≤Cn^{\sum_t\alpha_tm_t-D}\le C

once the source estimates apply. The conversion from H=n−8H=n-8 to nn only changes this constant. Strict positivity of (1) gives an effective stopping bound for every covered case. No numerical value for those bounds is needed in this component of the argument.

The twelve collision cases

Every remaining collision has r2=r5r_2=r_5, with the three positions r2,r3,r7r_2,r_3,r_7 forming a permutation of 0,1,20,1,2. There are two witness vectors, in prime order 2,3,5,72,3,5,7:

(3296,505,3931,0)/10000,(3296,502,3931,505)/10000.(3296,505,3931,0)/10000, \qquad(3296,502,3931,505)/10000.

A collision also supplies additional cofactor information. If kk primes share a position rr and ep=log⁡up/log⁡He_p=\log u_p/\log H, then

∏p:rp=rup=(n−r)k∏p:rp=rPp≥(n−r)k−1≥Hk−1,∑p:rp=rep≥k−1.(2)\prod_{p:r_p=r}u_p =\frac{(n-r)^k}{\prod_{p:r_p=r}P_p} \ge(n-r)^{k-1}\ge H^{k-1}, \qquad \boxed{\sum_{p:r_p=r}e_p\ge k-1.} \tag{2}

In particular these twelve cases satisfy e2+e5≥1e_2+e_5\ge1. A subsequent cofactor analysis may use (2); the polynomial coverage count above does not assume it.

Reproduction

The full checker recomputes all seventy modular ranks. The arithmetic mode checks the integer multiplicities and jet counts, source vectors, all 104,976 cases, the exact residual list, and the coverage hash, without repeating ranks (with the pinned dependencies named below installed in the repository's .venv/):

sh
.venv/bin/python \
  wiki/research/erdos_699/computations/erdos_699_index_9_baseline.py \
  --arithmetic-only

Omit the final option to repeat the rank calculations. NumPy and python-flint versions are recorded in the existing requirements file.