gpt-6.2-sol high: aht-repair-np-interface
Completed visible assessment, preserved as a readable check record. It is a model assessment, not a formal certificate. This document contains no private reasoning stream.
The assumed counter closes the specific component-counting bottleneck in all three tests. No weighted output is needed. The boundary-aware Euler calculation repairs the literal verifier's remaining arithmetic gap; the longitude step follows from the integral null-homology and orientability hypotheses, not from parity alone. No further component-counting interface obstruction was found.
R3
Checked statement: Assuming a correct unweighted orbit counter polynomial in k and log(N+1) for arbitrary signed interval pairings, its substitution supplies all three component tests in the 3-MANIFOLD KNOT GENUS NP verifier.
Returned assessment: repairable_gap
Yes for the component-counting bottleneck. The literal verifier also needs a boundary correction in its Euler-characteristic calculation, but that correction is computable directly from the same coordinates and needs no orbit weights.
For the compressed reductions, count the intersections on each triangulation edge from the local arc or disk coordinates and assign consecutive integer labels by edge, using prefix sums. On a triangular face the arcs joining a given pair of edges have their endpoints in contiguous blocks at the common corner of those edges. Their multiplicity and the chosen edge orientations determine two inclusive intervals and an increasing or decreasing pairing. For a normal surface, matching identifies the same face arc data from the incident tetrahedra, so use each face once. Omit empty families; a singleton family gives a valid singleton pairing. Thus a curve on s triangles gives at most 3s pairings, and a surface in t tetrahedra gives at most 12t. For a curve, the orbit graph is its arc graph. For a surface, its face arcs connect the corners around every elementary disk boundary. Attaching the disk interiors to these connected boundaries neither joins distinct components nor creates a component without an edge intersection. Hence the orbits count surface components as well. Building the pairings uses arithmetic on the compressed counts, not enumeration of the labels.
Require exactly one orbit for F and exactly one for its boundary curve on the peripheral torus. The boundary arc data come from the boundary faces, so the second test uses the curve reduction. An empty label set has zero components and fails either connectedness test. For the third test, the surface with coordinates 2v(F) is obtained by taking the two local parallel copies of each disk and gluing them by transport of the normal side, including along the boundary. By the normal-coordinate realization and uniqueness premise, this represents the normal-side double cover. For connected F it has two components if the side cover is trivial and one otherwise. Ambient orientability identifies two-sidedness with orientability of F. Require exactly two components in this test. All three calls return only an integer count; none needs orbit sizes, representatives, or weights.
With W the sum of disk coordinates, N is at most 4W, because each distinct edge intersection is represented by a disk corner. For curve arc-coordinate sum A, the analogous bound is N_curve ≤ 2A. In particular the boundary label set is a subset of the surface edge intersections. The quotient triangulation has only O(t) boundary faces. For the fundamental witness the supplied coordinate bound gives W ≤ 7t²·2^(7t+2) and N ≤ 28t²·2^(7t+2). Doubling the coordinates at most doubles the intersection bound without increasing the pairing bound. Thus the coordinates and pairing endpoints have O(t+log(t+1)) bits each on these witnesses; the verifier may reject coordinates exceeding the stated cap. The exterior triangulation has at most a constant times the original number of tetrahedra. The calls therefore take polynomial time in the binary input size, with no need for the excerpt's linear-time assertion.
The odd-intersection certificate checks essentiality, not the longitude slope by itself. A simple essential torus curve has primitive integral class, hence nonzero mod 2, and has odd intersection with some mod-2 cycle in the boundary 1-skeleton. That cycle has an O(t)-size encoding; parity is computed by summing the selected edges' intersection counts. The longitude conclusion uses the separate integral null-homology check on K. By the supplied spanning-surface existence premise, choose any oriented spanning surface and restrict it to the exterior using a collar of K. Its boundary supplies a longitude λ that vanishes in exterior homology. A meridian μ pushed into the exterior has signed intersection ±1 with that proper surface, so μ has infinite order in exterior homology. In the basis (μ, λ), the primitive oriented boundary class of an accepted F is pμ+qλ and vanishes in exterior homology. Thus p=0 and q=±1. It can be joined to K by a collar annulus in the solid torus without changing genus. This resolves the slope issue under the stated hypotheses; odd intersection alone would not resolve it.
Premises used
- A correct polynomial-time unweighted orbit counter on the compressed signed interval input is assumed, not established by this audit.
- Admissible nonnegative integral coordinates with the matching and quadrilateral constraints realize a properly embedded normal surface uniquely up to normal isotopy. The normal-side parallel construction respects the boundary and normal disk types.
- The stated input-validity checks and exterior construction apply: K is a simple knot in an orientable triangulated ambient manifold, its integral homology class is checked, and its regular neighborhood can be removed with the stated triangulation size bound.
- An integrally null-homologous knot has an orientable spanning surface. For a YES instance, the minimum genus is realized by an admissible fundamental normal witness in the exterior, with the supplied fundamental coordinate bound.
Missing inputs
- The correctness and complexity proof for the substituted counter and the omitted proofs of the listed topological and witness premises are not supplied as part of this obligation.
Scope changes
- The problem is the variable-ambient decision problem: a triangulated orientable 3-manifold M, a knot K in its 1-skeleton, and an integer threshold g, asking whether K has an orientable spanning surface of genus at most g. The supplied verifier starts with face-paired closed ambient input and works in the knot exterior with its peripheral torus.
- The result is a conditional NP verification argument, not a running-time result for the exact printed schedule or a hardness result for knots in fixed S³.
Located concern
AHT-verifier-excerpts, lines 1696–1710
The displayed Euler-characteristic calculation does not account for unglued boundary arcs.
Let D₃ and D₄ count triangle and quadrilateral disks. With A_int interior face arcs and A_bdy boundary face arcs counted in the quotient triangulation, 3D₃+4D₄=2A_int+A_bdy. An interior arc is counted twice by disk sides, but a boundary arc is counted only once. Using half the disk-side total alone therefore undercounts edges by A_bdy/2 and overstates Euler characteristic by that amount.
Compute V as the number of distinct surface-edge intersections, E=A_int+A_bdy from the face arc counts, and χ=V−E+D₃+D₄. After checking connectedness, orientability, and one boundary component, accept the genus test exactly when (1−χ)/2≤g.
The correction is polynomial-time compressed arithmetic. It is needed for a sound genus test but does not change or add any orbit-counter call.
Outside the audit
- Candidate R proof parts (1)–(3), including the modified schedule's correctness and running time.
- Omitted topological foundations and fundamental-witness proofs.
- The Section 3 NP-hardness reduction and any fixed-ambient hardness claim.
Additional supporting archive items are identified in the original report. View the manuscript hashes and repository snapshot.