Skip to content

Paid clones on skip-prefix strips: conditional κ = 4.529040672e-5 - #54

Merged
Mister-Meeseeks merged 6 commits into
CrocSwap:mainfrom
chafreaky:research/skip-prefix-clones
Oct 8, 2026
Merged

Mister-Meeseeks merged 6 commits into
CrocSwap:mainfrom
chafreaky:research/skip-prefix-clones

Conversation

@chafreaky

@chafreaky chafreaky commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

This improves the conditional exponent from PR #53 by 0.686876%, and from PR #36 by 17.769260%:

$$ T(n)=O\left(n(\log n)^{1-\kappa}\right),\qquad \kappa=\frac{141532521}{3125000000000}=4.529040672\times10^{-5}>2^{-15}. $$

The result remains conditional on OpenAI's base theorem and the inherited analytic and fixed finite-alphabet multitape interfaces.

Construction

Starting from PR #53's skip-prefix graphs, add 253 whole-chain clones at h=23 and 353 at h=25. Each added gate recomputes an existing scalar sum from disjoint sources and replaces one complete continuation chain. Two previously unused carrier capacities retain its inputs. All 606 added gates and 1,212 added carrier links are included in the cost.

Quantity PR #53 This PR
Roles at h=23 32,946 32,693
Roles at h=25 43,409 43,056
Physical wires W 160,799,739 159,592,676

Every clone retains its parent's original core/cover envelope. The common I+J bases, all 4,073,300 data pairs, copied centers, paid endpoint corrections, complex network and assembly interfaces remain unchanged. The exact bit saving is 1132311451/25000000000000.

This branch builds on #53 at 3ffd4021995c959ac02d12920e0279ae97dd03c7. The new implementation and proof are in research/skip-clones/.

Verification

Completed on the published construction:

  • Exact scalar identities, dense global supports and independent recount of every selected carrier use.
  • Fresh fixed-profile CRT replay and complete dirty-scratch basis checks in both orientations.
  • Exact rational moment bounds, next-grid rejection, 47 strict assembly constraints and seven positive margins.
  • 11 focused tests and 65 adversarial rejection cases, including malformed partitions, partial chains, provider collisions, invalid counts and stale source pins.

Latest main (d55bcd6) is integrated at 1d1bb12. Fresh make -j1 verify-research verify-skip verify-tests passed: 28 recovered research audits, complete PR #53/#54 producer replays, 20 + 11 focused tests, 450 integrated tests, and 20 historical patch checks. Regeneration was byte-clean. All 152 upstream-added files, all 239 pinned source entries, and the numerical certificate are unchanged. Both the new upstream research CI group and the existing skip CI group remain, across Python 3.11/3.13/3.14, with both formal jobs preserved. The integration receipt distinguishes these fresh scoped checks from prior complete runs.

The earlier full make verify passed after integrating main 64ed4cb, at tested merge 03aff8b: 277 integrated tests, all focused and complete producer replays, and 20 historical patch checks, with byte-identical regeneration. An independent review confirmed preserved main checks, credits, and numerical certificates. That full-run record remains intact; no fresh full-suite or local Lean result is claimed for the latest integration. All 23 GitHub CI jobs passed on 1d1bb12 (run): all seven Python groups on three versions, plus both inherited formal packages.

The original construction also passed its full suite at 7210de7 (218 integrated tests and 18 patch checks).

make skip-clones-verify
make verify

The local proof states the argument and the inherited obligations requiring expert review. This is not a Lean proof or a claim of global optimality.

Attribution

All inherited notices and credits remain, including icekylinx, Dominik Scholz, James Chang, Zhihao Chen, Paureel, Swapnil Jain, eumemic, Douglas Colkitt, OpenAI, David Harvey and Joris van der Hoeven.

ikeboy and others added 2 commits October 8, 2026 12:08
Only the scalar producer DAG (research/skip-strips/skip_graph.py) and its
pinned PR CrocSwap#44-weighted carrier matching change. Leave-one-out strips use
s_j = A_{j-2} + (v_{j-1} + B_{j+1}) instead of s_j = A_{j-1} + B_{j+1}, so
the two consumers of every prefix (and of every new suffix-side node) have
nested envelopes that the inherited carrier matching can join.
- Top level: the layout runs on the reversed value list.
- Deeper levels: values are sorted by (support size, support) (PR CrocSwap#48's
  idea), ascending at h=23 and descending support at h=25.
- Totals follow PR CrocSwap#48's rules.

Roles R23=32946, R25=43409 (PR CrocSwap#48: 36432, 48329); W=160799739.
Exact bit saving 4498347/10^11; kappa 4498144/10^11 (+9.21% over PR CrocSwap#48).
Everything else is PR CrocSwap#48's research/copied-fixed pipeline with recomputed
records and constants. Adds make skip-strips-verify, tests, README and
NOTICE entries.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013eMahk2wJVfY7JKgp81K6y
@chafreaky
chafreaky marked this pull request as ready for review October 8, 2026 16:49
@Mister-Meeseeks
Mister-Meeseeks merged commit 6ef3795 into CrocSwap:main Oct 8, 2026
23 checks passed
chafreaky pushed a commit to chafreaky/integer-mult-bounds that referenced this pull request Oct 8, 2026
…48671376e-5

Apply PR CrocSwap#51's backward-positive labels and paid whole-chain clone search to
PR CrocSwap#53's skip-prefix DAGs, with exact optimal carrier matching under the
positive-frame adjacency. Roles fall from 32,946 to 32,693 at h23 and from
43,409 to 43,009 at h25, and W to 159,509,439. Fixed I+J bases, so PR CrocSwap#53's
data corners and pipeline carry over unchanged.

kappa 4548671376/10^14: 0.4334% above PR CrocSwap#54, 1.1233% above PR CrocSwap#53. build.py
rebuilds everything from scratch and passes PR CrocSwap#51's independent compiler
check. The profiler reproduces PR CrocSwap#51's and PR CrocSwap#53's certified profiles
exactly. 47 constraints and seven margins pass, next grid points are
rejected, and the PR CrocSwap#53 and CrocSwap#54 networks are excluded.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011S35E2acyQDaVRjPRw4TDy
chafreaky pushed a commit to chafreaky/integer-mult-bounds that referenced this pull request Oct 8, 2026
Introduce the mathematical dual skip-suffix layout:
  s_j = (A_{j-1} + v_{j+1}) + B_{j+2}
which keeps core popcounts >= 2 and enables SharedPointCircuit to merge
530+ additions across groups. This lifts carrier matches to 12,894 (h=23)
and 17,051 (h=25), dropping roles to R_23 = 32,541 and R_25 = 43,049,
and slashing wire volume to W = 159,230,679 (-1,569,060 vs PR CrocSwap#53,
-361,997 vs PR CrocSwap#54) with zero paid clone tables.

Certified:
  - Bit saving: a_bit = 4541218/10^11 = 4.541218e-5 (strict gap 1.208e-13)
  - Exponent saving: kappa = 4541011/10^11 = 4.541011e-5 > 2^-15
  - All 47 strict assembly inequalities and 7 margins verified
  - Verified across CRT matrices (0 disagreements) and both-orientation timelines
  - OpenMP 32-core parallelization on data corners

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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.

3 participants