Repository navigation
Construct rounded C0/C2 calls with all-width precision contracts (no new kappa) - #5485
Open
JoyboyBrian wants to merge 3 commits into
Open
JoyboyBrian wants to merge 3 commits into
JoyboyBrian wants to merge 3 commits into
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A concrete fixed-grid nearest-even butterfly backend now supplies the completed-call tolerance assumed by the actual PR4743 C0/C2 port analysis. For rho=2^-q, g=q+ceil(f/2)+1 grid bits suffice uniformly. The proof includes internal numerator ranges, actual stored inputs, adjoints, every two/three-call forward/reverse schedule and w=H+g+2f+ceil(f/2)+9 signed component bits. Deterministic nonzero rounding is executed rather than injected.
The exact source pins, all-width proof, executable controls and open next questions are in
research/native-rounded-butterfly-contract/. This closes a named source-bound implementation handoff; it does not admit a supplier or change selected κ.Validation: 20,003 signed half-rounding cases;36 completed-call cases at widths1/2/3 across C0/C2 and adjoints;12 actual forward/reverse port cases with nonzero rounding and off-grid input quantization; every actual completed-call input/output; overflow/axis/adjoint controls;625 one-butterfly residue fixtures. Registered Python3.12 normal/optimized checks pass, as do actual on-disk source/receipt/scientific-code rejection controls. Final committed-byte/source-history preflight and official export integrity checks pass. Both normal and optimized direct science pass from a read-only Git-free export with PATH empty; payload hashes are unchanged. Separate AI peer review found no blocking issue.
Limits: The direct backend costs O(fN), N=8^f, and does not inherit an accelerated recursive rank-one tariff. The source-ordered external channel bounds are reused with attribution, not claimed as a new norm calculation. Concrete fault patterns obey additional parity constraints; no attainability of the earlier independent-ball sharp constant is claimed. Accelerated recursion, fixed tapes, native supplier/outer admission and selected-frontier replay remain open.
Prepared by JoyboyBrian with substantial OpenAI Codex assistance; Apache-2.0. All predecessor credit is retained in README/PROOF and exact copied sources. Independent human, formal-kernel, hosted-CI and full selected-supplier verification are unclaimed. Canonical investigation: #5475.