Happy Ending · Erdős–Szekeres & ordered rank-3 signotopes

Restricted capacities N(m,r,s) and planar thresholds

Path-avoiding signotope capacities across m = 7, 8, 9 and general m, combining replayed CNF+DRAT and native-trace integer dual certificates with structural working proofs.

Run: pre-MathRSI Trellis + ARIS (through Nov 3) and Trellis + ARIS + MathRSI v3 (Nov 5–17) → integrated in v6 main maintenance.

Objects and three-step assembly

On a complete ordered rank-3 signotope (every 4-tuple packet changes sign at most once), N(m,r,s) is the maximum number of elements avoiding blue r-vertex paths, red s-vertex paths and same-endpoint weak m-gons, while ES(m) is the planar general-position threshold forcing a convex m-gon. Write Cₘ = N(m,m−1,m−1), Qₘ = N(m,m−1,m), Fₘ = N(m,m,m), and let Dₘ denote the R7 residual capacity with junction exclusion (no element simultaneously ends a red (m−2)-path and starts a blue (m−3)-path; Dₘ cannot substitute for unconstrained N(m,m−2,m)). For m ≥ 6, disjoint H–R splitting (R7) gives Qₘ ≤ Cₘ + Dₘ, leftmost-pivot chart transport (R20) gives Fₘ = Qₘ + 1, and planar inclusion adds 1: ES(m) ≤ Fₘ + 1 ≤ Cₘ + Dₘ + 2.

Results by setting

SettingKey boundsVerification & manuscript scope
m = 7 (heptagons)ES(7) ≤ 77 (working)N(7,5,6) = 25 accepted (same-family/provisional) + 25-point E21 abstract witness & DRAT ≤4 lock blocks replayed; analytic N(7,5,7) ≤ 39, N(7,6,6) ≤ 49 and 75/76 → 77 working proofs (vs. Schroeder v1 planar ≤ 81; review pending).
m = 8 (octagons)N(8,6,6) ≤ 61 · N(8,8,8) ≤ 318milestone9 115-row CNF+DRAT replayed for N(8,6,6) ≤ 61 (exact matrix minimum 9). Schroeder’s 15-file omission trees replayed (193 / 124) + signotope packet bridge give working N(8,8,8) ≤ 318; planar ES(8) ≤ 319 is Schroeder’s published bound (autonomous fallback: 322 / 323).
m = 9 (nonagons)N(9,8,8) ≤ 688 · ES(9) ≤ 1148Native-trace integer duals replayed in stdlib Python (zero float) for C₉ ≤ 688 (≥236 holes) and D₉ ≤ 458 (≥124 holes, matching the 47,635-row depth-≤3 LP ceiling). R7/R20 assembly gives working N(9,9,9) ≤ 1147 and ES(9) ≤ 1148 (vs. 1508 printed / 1501* planar reference; review pending).
General m2b − 1 · n²/12 · n³/80Working coordinate-slice double-counting proofs for N(2p−3,p,p) ≤ 2b − 1 (p ≥ 4), quadratic 3-column deficits for N(m,5,m−1), and cubic 4-column deficits for N(m,6,m−2); no closed-form ES(m) claimed for m ≥ 10.

m = 8 · shared-label CEGAR

N(8,6,6) ≤ 61

In the 4 × 4 box of 70 states, 63 six-order box chains and 30 shared-pivot rows give 93 baseline rows (≤ 62). Adding 11 DRAT-verified shared-label CEGAR kernels (plus transposes = 115 rows) makes budget 8 UNSAT, forcing at least 9 holes.

A 9-hole hitting vector intersects all 115 rows, proving this matrix relaxation peaks at 9 (a matrix witness, not a 61-point signotope).

m = 9 · native boundary traces

ES(9) ≤ 1148

Self-contained integer duals (Λ = 10⁶) prove ≥ 236 holes in 924 states (C₉ = N(9,8,8) ≤ 688) and ≥ 124 holes in 582 restricted states (D₉ ≤ 458, where a rational vector of sum 123,539,833 / 10⁶ ≤ 124 satisfies all 47,635 depth-≤3 rows).

R7 and R20 assembly gives working bounds Q₉ ≤ 1146, F₉ ≤ 1147 and ES(9) ≤ 1148, below Mojarrad–Vlachos (2016) 1508 and the 1501* (= 924 + 575 + 2) planar reference.

Heptagon and octagon structure notes

Assurance boundaries & open targets
Five finite suites (native-trace 688 & 458 + 47,635-row ceiling, milestone9 N(8,6,6) ≤ 61 CNF+DRAT, Schroeder’s 15-file omission trees, m = 7 five-lock-block DRAT, and the 25-point E21 abstract witness) are replayed directly in-repo. All structural, cyclic-sign transfer, and R7/R20 assembly manuscripts remain working proofs pending independent mathematical review, and there is no Lean formalization. Unconditional N(7,6,7) ≤ 74, whether N(8,6,6) drops below 61, and depth-≥4 or shared-label improvements beyond D₉ ≤ 458 remain open.

Certificate architecture & workflow refinements

Parameterized tracesreconstruct rows from (q, p, inc, exc, weight)
Integer {0,1} dualcheck D − P > (H − 1)Λ with zero floating point
Ceiling checkhalt matrix tightening when dual meets feasible primal
  1. Stop when a relaxation hits its exact ceiling: When an integer/DRAT dual lower bound meets an exact relaxation-feasible witness at the same integer (the 9-hole hitting vector for the 115-row N(8,6,6) matrix, or the 47,635-row rational vector for depth-≤3 D₉), stop searching inside that relaxation and add shared-label or geometric constraints instead.
  2. Ship generative sparse integer duals: Store only nonzero-weight row parameters (q, p, inc, exc, weight) so a compact standard-library checker rebuilds every used trace row and integer overload penalty P = ∑ max(0, ℓ − Λ) without bundling multi-hundred-megabyte LP matrices.

Runnable checks & archive layout

The curated in-repo archive (plugins/happy-ending/archive/nmrs-20261117/, 955 files, 48.18 MiB, max single file 9.56 MiB, zero Git LFS) preserves all manuscripts, small helper libraries, acceptance receipts and all five replayable certificate suites byte-for-byte; duplicate pre-rendered HTMLs and exploratory ≥2 MiB CNF/DRAT artifacts stay in the external NMRS-results-20261117(1).zip and are listed with SHA-256 digests in INVENTORY.json.

Earlier m = 7 reader (ES(7) ≤ 77) ↗中文 · Case summary & verification table ↗ · Happy Ending plugin & checkers ↗ · v6 maintenance notes ↗

Back to evolution ↑ · Back to results ↑

MathRSI · GitHub ↗