files / problems / erdos-szekeres-seven / compute / q2 / RESULT.md

2.4 KB · on GitHub

q2 replay record — 2026-08-23

No bound was proved. The full run ended with s UNKNOWN, and the incomplete search is not evidence for an upper bound.

Deterministic full instance

Command:

python3 encode.py --n 33 --k 7 --out es_33_7_signotope.cnf

The generator took 29 seconds and produced:

variables       46,376
clauses         5,254,128
bytes           916,030,006
SHA-256         3771831e7d5730d9a2ca81356253cee1c44d5744de92942f664c4767c60f58c9

Kissat 4.0.4 ran on this DIMACS file without proof output and with --time=300. It exited 0 after 300 seconds with exactly s UNKNOWN. Its RSS was 1,045,528 KiB at an intermediate 87-second sample. No solver certificate exists for this run.

Certificate smoke test

Kissat 4.0.4 returned SAT for (n,k)=(8,5) and UNSAT for (9,5). drat-trim independently accepted the text DRAT for (9,5) with s VERIFIED; it used 1,589 of 3,150 input clauses and 3,221 of 4,012 lemmas.

n8 CNF SHA-256    ca012f771197fdd43c75dc7a9a0966e2d3cc21b30f73a0ac0574a43ad743418a
n9 CNF SHA-256    a82aaee2281654626dd968d70830c58d51364bd964ed2949afa298f5c0ee7852
n9 DRAT SHA-256   d975f6fce8a259f1718feb093ccfa6b75316393b09f349d0de6f7136ebef328c

The source archive used to build Kissat had SHA-256 c4ec54f5034ec096a2ce0dbb38ecd235e673c999d7073c771dcc21c4f32fc123. The resulting Kissat binary had SHA-256 0e0470d5ef16740554ad2453183546067b190d85b82d17bbfad6c6a7bc76cb49; it reports GCC 14.2.0 with -O3 -DNDEBUG -DQUIET and an unknown git ID. The locally built drat-trim binary had SHA-256 7d426a913ba22202174cb3a57ef9924b28c62b90a65bcaffa7e010281116c6bb. These binaries are not notebook artifacts; run_all.sh accepts independently installed copies through KISSAT and DRAT_TRIM.

ES(6) storage wall

The same encoding returned SAT for (16,6). A proof-producing (17,6) run was manually stopped after 2 minutes 53 seconds with no answer. Its incomplete text DRAT had reached 586,153,984 bytes and cannot be checked as a proof. The uncommitted CNFs had these deterministic hashes:

n16 CNF SHA-256   4983a9cc8c2f1a0219f4428a4d5d6c0a0f0ef77936ce2c4722011c7308e3bb75
n17 CNF SHA-256   0e3ed27cf6c6ab316df344674fb834761d219b7e9e349c00326d0ff0f9db9a3f

This wall is why the full 33-vertex reconnaissance did not request a proof. An UNSAT claim at 33 would require a new proof-producing run and independent verification; the bounded run above made no such claim.