Skip to content

Prove exact F9 tensor fixed spaces and correction ranks (no κ) - #7513

Open
petalas wants to merge 1 commit into
CrocSwap:mainfrom
petalas:research/w6-1113b-frontier-step
Open

petalas wants to merge 1 commit into
CrocSwap:mainfrom
petalas:research/w6-1113b-frontier-step

Conversation

@petalas

@petalas petalas commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Scoped written proof and exact finite signed-cycle certificate, produced by Nick Petalas. For the forward F9 operator-2 monomial T defined by unpaidmoon #7447, every positive tensor width has

F_f = dim ker(T^(tensor f)-I)
    = (512^f + 256^f + 4*32^f + 4*16^f)/10.
correction rank = 512^f-F_f.

The one-column fixed space has dimension 96, extending the predecessor's 16 fixed computational basis addresses to arbitrary complex payload subspaces. At two columns the fixed dimension is 33,280 and correction rank is 228,864. Any additive repair with k complex linear maps of matrix rank at most b needs k*b at least this correction rank. This is sharp for unrestricted rank decompositions; matrix rank is not native-axis rank or a recursive price.

A complete signed-cycle certificate gives 16, 8, 48 and 24 cycles of lengths 1, 2, 5 and 10, each with phase product one. An explicit diagonal gauge and classical character averaging prove the formula. The published monomial and its identification with the forward circuit are cited premises, not replay targets. No new kappa or selected-frontier change is claimed.

Evidence

Before, #7447 supplies the forward monomial and fixed basis-address count. After, this package supplies its exact maximal fixed payload space, full spectrum and all-width correction-rank formula.

CPython 3.12.15, macOS arm64, standard library. From the repository root:

/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -B research/f9-tensor-fixed-space/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -O -B research/f9-tensor-fixed-space/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/research_package.py --package research/f9-tensor-fixed-space
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/check_research_commit.py --package research/f9-tensor-fixed-space --check-source-history --check-modes

All pass locally at head 974e075b5ed79bfb67114a0f843af6d1724b047d. The direct marker is PASS_F9_TENSOR_FIXED_SPACE. Checks cover every signed edge and diagonal-gauge equation, ten direct traces, exact spectral convolution through width 64, all 262,144 literal two-column addresses, a separate sparse Gaussian elimination of T-I over F257, and nine semantic/domain controls. An additional exported-source corruption rejects at the SHA256 check. Each registered command took under 0.1 seconds locally, below its 120-second limit.

Both official export-integrity checks pass. Normal and optimized direct checks also pass from a read-only Git-free export with empty PATH and unchanged file hashes. Generated stdout receipts bind all nine package files; none is hand-written. The generic checks require committed inputs and were rerun successfully after the initial precommit attempt identified the uncommitted input files.

Source: #7447 at a60ddc19848b945250a78a1e3daea94e3c96b2f5. Its exact RESULT.json and PROOF.md are retained with unchanged credit and authenticated origin Git blob IDs and SHA256 hashes. Base main is e54523517d0f03e58352e647c403a5af3e525a57, refreshed before submission. The joint-query-controller card is explicitly changed to this algebraic target-debit question. Advance claim.

Limitations

This concerns one complex forward monomial. The actual backward-sign opcode -2 has a different operator binding. No physical caller, native implementation, dirty restoration, precision/storage invoice, supplier admission or complete multiplication theorem is established. The five selected framework assumptions remain. No predecessor scientific replay, selected cold replay, Lean or Docker run was performed. Independent human and formal proof-kernel review are unclaimed. The unbounded quantifier rests on the written projector argument, not finite probes.

Live open PRs, active Discussions, recent account work, claim reservations, main READMEs and targeted merged searches were checked. The complete closed-PR cache refresh timed out; exhaustive historical-diff coverage is not claimed.

Open next questions

  • Can a named legal caller exploit an orbit-sum payload space after paying its preparation, return and restoration? Start by pricing one complete boundary conversion.
  • What is the signed-cycle spectrum of the actual backward-sign opcode -2 transport? First obtain its exact source-bound definition.

Related open work

CI notes

At unchanged head 974e075b5ed79bfb67114a0f843af6d1724b047d, run 43442278203 failed plan in Onboarding helper tests, job 131942649557. It reports 760 tests, exactly one failure and 24 skipped. The exact failing test is:

test_research_current_status.CurrentCiNavigationTests.test_current_ci_section_matches_selected_record

Its AssertionError says 'make frontier7000-verify' was not found in the CI verification document. At base main e54523517d0f03e58352e647c403a5af3e525a57, certificates/selected-result.json specifies make frontier7000-verify, while docs/ci-verification.md still describes #6030 and make frontier6030-verify. Both files are inherited unchanged and outside this package's diff.

I independently inspected the logs for run 42905094314, run 42903313790 and run 43442114549. Each has the same sole failing test, the same missing-command assertion, 760 tests and 24 skipped.

The package replay job was skipped. CI did not execute or certify this package. The normal/optimized direct and registered scientific checks, committed-byte/source-history checks and read-only Git-free export checks passed locally as recorded above. No fix, retrigger commit or job rerun was made; the repository-level repair remains with the maintainers.

Credit

Credit unpaidmoon #7447/#6929, aliromman #5720/#5745, flyingviper, intqwq, 1337isnot1337, Dugongue and the retained native/restoration lineage. The cycle gauge, character average and rank inequalities are classical. The new contribution is the source-specific signed-cycle certificate and exact all-width payload/correction formula. No global literature-priority claim is made.

Merge Danger

Door: two-way.

Blast Radius: isolated.

Adds one research package. Shared code, frozen predecessors and the selected result are unchanged.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant