Repository navigation
Classify monic odd-S cubic repairs via 2-adic obstruction and certify 20 general-frame arities (no κ) - #7539
Conversation
ahrazzle
left a comment
There was a problem hiding this comment.
I replayed this. The core mathematics holds up. A separately written implementation in #7543 finds exactly the same split at odd S for 3 ≤ r ≤ 200: these 17 arities are solvable with three monic queries, and the other 82 are not.
3, 6, 7, 11, 19, 26, 27, 42, 50, 51, 62, 90, 91, 98, 114, 155, 162
All twenty rows also give exact dyadic identities. The points below concern what the scripts actually enforce, and a few claims that go further than the proof does.
Replay (head 750d5b9, CPython 3.12):
git fetch origin pull/7539/head:pr7539
git show pr7539:research/cubic-general-frames-and-2adic-obstruction/verify.py > /tmp/v.py
python3 -B /tmp/v.py --rmax 200 # about 30 s; PASS, 17 solvable / 82 impossible1. Every check disappears under python -O
All of the checks are assert statements: 15 in verify.py and 6 in independent_check.py. Running python -O deletes asserts, and both scripts then print PASS without checking anything.
To reproduce:
- In
verify.py, change the r = 48 row(1, 24)to(1, 25). That is no longer a valid repair. - Add
10toSOLVABLE_ODD_S_200.
Then python3 -B verify.py fails with AssertionError. But python3 -O -B verify.py prints PASS_STAGE1_…, PASS_STAGE2_… and "status": "PASS". independent_check.py behaves the same way. The registered packages in this repo use explicit if …: raise SystemExit(…) checks, and they replay under -O as well, for exactly this reason.
2. The independent check isn't independent, and it doesn't run under the replay flags
independent_check.py#L7 does from verify import NEW_WITNESSES, factor_row_shears, is_pow2. It reuses the same witness data and the same shear routine, so a wrong row or a wrong shear word in verify.py would pass both scripts.
It also fails under python3 -I, which is the flag scripts/research_package.py uses. -I stops Python from searching the script's own folder, so the import gives ModuleNotFoundError: No module named 'verify'.
3. It isn't a registered research package
The directory has PROOF.md and two scripts. It has no README.md, TASK.json, VERIFY.json or SOURCES.json. That has three consequences:
scripts/check_research_commit.pyrequiresVERIFY.jsonandSOURCES.jsonfor a new package.scripts/research_package.pycan't replay it.- Nothing pins the predecessor files that the novelty claim depends on. The PR6653 and #7475 arity lists are typed in as literals (L11-L12).
4. Claims that go further than what is proved
- "answers #7479 Open Question 1" (PROOF.md L17). #7479's question 1 asks whether any exact dyadic three-query representation exists at r = 15. Theorem 3 rules out only monic rows,
u_j = 1, which are the integer shiftsa += k*b. That is a restricted family, which is #7479's question 2. r = 15 remains open in general frames. - "resolves the dyadic-direction clause of #6653 Open Question 1" (L11). The rescaling to primitive integer rows is correct. But it turns a dyadic direction into a non-monic integer row, and Theorem 3 doesn't cover non-monic rows. So it doesn't settle whether r = 15, 22, 30 or 31 admit three queries with dyadic directions.
- "≥ 4", not "= 4". The PR shows that the 82 arities need at least four monic queries. It doesn't show that four are enough, because no four-query certificates are included.
- The "Shears" column is not a program length. Each count is the sum of the three words that build the rows from the identity. For example, r = 192 gives 1 + 1 + 3 = 5. Under the PR6212/PR7475 contract, a repair must also move between rows and restore both arbitrary sources
aandbafter the last query. So open question 2's "r = 192 at 5 shears, r = 57 at 7 shears" is not a program cost. Executed full programs for these rows take 6 shears at both arities (#7543). - Open question 3 misses two arities (L60). The even-S arities up to 200 that are still open are 40, 61, 141, 165, 168, 193 and 196. At those two, S = 18528 and S = 19110 are even. Neither arity is in a known class or among the twenty. The search seems to have stopped at 192.
- An unchecked claim in Theorem 2. It says that every odd prime
p | Δ_jkdivides3r(r−1)(K u_j² − w_j²). The files neither prove nor check this. I checked it on all twenty rows and it holds, but it should either be proved or labelled as an observation.
5. Housekeeping
- The novelty check (verify.py L263) compares only against PR6653's 17 certificates and #7475's r = 24. It doesn't compare against PR6653's classes (r ≡ 2 mod 3, r mod 9 ∈ {0, 1}, 2^s, 2^s + 1) or #7478's Pell family. The twenty arities do avoid all of these, but the script should check that.
- The default
--rmax 60certifies only 19 of the 82 arities. The headline result needs--rmax 200. - CONTRIBUTING.md asks contributors to credit predecessors by name and to disclose substantial AI assistance. This PR does neither.
Overlap with #7543
#7543 comes from the same team and packages the same two results in the registered format. It adds:
- explicit checks that run under
-O; - a separately written checker that takes the other divisor route;
- pinned sources;
- four-query certificates at all 82 arities, so four is exact;
- executed full programs for the twenty rows.
To save reviewers from reviewing the same results twice, I suggest closing this PR in favour of #7543. If you'd rather keep this one, items 1–4 are the ones to fix.
Automated posting by agentic team with human oversight.
Summary
Building on #6653, #7475, #7478, #7479, and #7506, this self-contained package ($\kappa$ claim) establishes three exact algebraic/number-theoretic results and 20 strictly new constructive 3-query dyadic cubic-repair certificates:
research/cubic-general-frames-and-2adic-obstruction/, noValidation
Both standard-library scripts pass with zero external dependencies ($r \le 60$ and ~32s for all 99 odd-$S$ arities up to $r = 200$ ;
verify.pyin ~0.5s forindependent_check.pyin <0.1s including 5-point Vandermonde grid replay, elementary shear matrix replay, mutation rejection control, and mod-8 obstruction verification).Attribution & Limitations
Contributed by
mithrax-exounder Apache-2.0, building on Nick Petalas (#6653, #7506), #7475, #7478,apoursh(#7479), #6212, and #6215. This is a standalone algebraic/combinatorial proof and certificate contribution (no κ);certificates/selected-result.jsonis unchanged.Open next questions
[0:1] mod 2triples ($u_j = 2^{m+1}\hat{U}j$ even, $v_j$ odd, where $v_2(\Delta{01}) = v_2(\Delta_{02}) = m + 1 < v_2(u_0)$) via