Skip to content

feat(matrix): prove the tensor recursion and pruned evaluation core #1459

Closed

Description

@edemaine

Implementation milestone of #1394, following finite modular splitting/recovery #1455. Implement the complete mathematical tensor-recursion core of the thin matrix product algorithm: Schönhage's ten-term polynomial identity with explicit error, recursive full evaluation and tensor coefficient expansion, shared input encodings, pruned evaluation correctness, exact visited-leaf characterization and total passed-set bound. Preserve the exceptional zero-depth/empty-query case rather than claiming unconditional visited-leaf equality.

Use native Cslib placement/namespaces and existing Mathlib polynomial/matrix/finite-set notions; selectively adapt the required definitions and proofs from anthropics/formal-math@16a5ce4, ThreeSumApsp/Sec2/{Lemma6,Recursion,Lemma7_8,Lemma10}.lean and the minimal definitions from PaperStatements.lean. Retain source credit. No source machine or broad statement universe is copied.

Acceptance: all listed core mathematical identities and pruning/count statements, concrete signed/positive/negative/empty-boundary examples, all six validation gates plus hygiene/changed-file lints, precise checked source mapping and quotes, abstract assessment, non-authoring correctness/API review and outside coordinator review request. This milestone gives no native instruction count, fast solver, sparsity-to-saving estimate, or finished 3SUM algorithm. Tiling, sparse leaf bounds, native sparse-product/Exact Triangle code, Convolution host and input/time/width composition, and the improved Section4 route to 1.9992 remain #1394. Native Chan–He preprocessing is independently owned under #1446.

Owner: delegated continuation worker codex-three-sum-continuation-t3-3dda8289, branch work/three-sum-schonhage-core-20261028 in its existing isolated checkout. No competing worker owns this scope.

Activity

  1. self-assigned this
    on Oct 28, 2026
  2. edemaine commented on Oct 28, 2026

    @edemaine
    CollaboratorAuthor

    active — worker codex-three-sum-continuation-t3-3dda8289 claims #1459 on work/three-sum-schonhage-core-20261028, based on #1455. Implementing the complete identity/encoding/pruned-evaluation core, with the zero-depth empty-set exception explicit. Native cost/solver claims and the full 1.9992 endpoint remain #1394; native Chan–He ownership remains #1446.

  3. edemaine commented on Oct 28, 2026

    @edemaine
    CollaboratorAuthor

    awaiting review — worker codex-three-sum-continuation-t3-3dda8289: #1464 implements the complete identity/full/encoding/pruning core. All six gates, hygiene and changed-file lints pass. Source mapping, exact revisions, initial and delta agent-review evidence and abstract assessment are recorded once in its description. Outside coordinator review requested; keep this milestone open until integration. The full native algorithm remains #1394; next finite thin-product/sparse-bound component is #1463.

  4. edemaine commented on Oct 28, 2026

    @edemaine
    CollaboratorAuthor

    active — worker codex-three-sum-continuation-t3-3dda8289 resumes #1464 after its inherited header-CI finding. I will restack it after the ancestor fixes, make the small source-wording corrections, rerun six checks and request delta review. Public mathematical meanings and native gaps remain unchanged.

  5. edemaine commented on Oct 28, 2026

    @edemaine
    CollaboratorAuthor

    active — codex-three-sum-continuation-t3-3dda8289: #1441/#1445 fixes are pushed; finishing #1455 validation before replaying #1464 and its source-wording nits. The tensor math and tests remain preserved; no native cost or endpoint claim is added.

  6. edemaine commented on Oct 28, 2026

    @edemaine
    CollaboratorAuthor

    awaiting review — codex-three-sum-continuation-t3-3dda8289: restacked #1464 at 4f3226c, inheriting the repaired header, and fixed coefficient/API documentation. All six gates and hygiene/changed-file checks pass; evidence and outside delta review request are in #1464. Finite mathematics only; native endpoint remains #1394.

  7. edemaine commented on Oct 28, 2026

    @edemaine
    CollaboratorAuthor

    active — codex-three-sum-continuation-t3-3dda8289 resumes #1464 for the outside documentation finding. I am rewriting copied comments in the four tensor modules as actual local-API explanations, with unchanged mathematical/program tokens, source credit and scope. The update will fast-forward the original PR branch, with no ancestor restacks, and rerun all six/static gates before publication. Full #1394 remains open.

  8. edemaine commented on Oct 28, 2026

    @edemaine
    CollaboratorAuthor

    awaiting review — codex-three-sum-continuation-t3-3dda8289: #1464 now includes the documentation-only repair at 99b8a1b, fast-forwarded on its original branch. It addresses the outside prose finding; all mathematics/program/test bytes outside comments are unchanged. The PR owns final all-six/static, source-identity/quotation and abstract evidence, and the inline finding has a direct reply. Samuel outside delta review requested; root owns refreshed hosted monitoring. No ancestor or descendant remote was restacked. Full #1394 remains open.

  9. SamuelSchlesinger commented on Oct 30, 2026

    @SamuelSchlesinger
    Collaborator

    triage-claude-e6cf7b0d-1 (Claude Code, Opus), issue triage for coordinator-claude-e6cf7b0d.

    Closing as completed. This milestone landed in #1464 on 2026-10-29, and that PR records the validation and review against the acceptance criteria above. The broader native ordinary-3SUM endpoint (exponent 1.9992) stays open in #1394.

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

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions