Repository navigation
P104: tiered ceilings for words meeting the native-class hypotheses (incl. the frontier words), Lean only - #7668
Open
DreamingOfClouds wants to merge 1 commit into
Conversation
…Lean, no kappa) T1 ceiling_native_J (t0 .0118103, junk count >= 4962), T2a ceiling_noHJ (t0 .0151027, Law A replaces the junk count), T2b ceiling_native_Z0 (t0 .0151869). Kernel-only; builds on CrocSwap#7656 and CrocSwap#7659. Hypotheses copied from the Lean in PROVED.md; word premises in NATIVES.md. DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Lean-only package, builds on #7656 and #7659 (their modules are included unchanged; new modules: the T1 pieces
CongestionNativeJ*,JunkLawA,JunkLawACeiling, the N96 and Z0 piecesCongestionNativeN*,CongestionNativeZ*). No κ claim and no word claim in Lean.Results. Each theorem proves
Finv (1 − t) (invoice (caller 6 7) n 17) < 0, the F-form of the 17-call invoice of the #7113 caller C(8|9), for every t with t0 ≤ t ≤ 1, under its full hypothesis list inPROVED.md(copied from the Lean text; the same hypotheses are also listed by script inREADME.md).ceiling_native_J: t0 = .0118103 (tJ0). Hypotheses include the junk counthJ : 4962 ≤ J.card.ceiling_noHJ: t0 = .0151027 (tN0). No junk-count hypothesis (96 ≤ |J|is derived in Lean from Law A,JunkLawA.lean), but it needsm : LawAModel K 104 D(hspan,hdiag,hodd,hpart ≤ 48),hD : D ≤ 56andhJdef : J = junkSet m.M.ceiling_native_Z0: t0 = .0151869 (tZ0). No junk-count hypothesis and no upper bound on owned mass (none of T1, T2a, T2b boundsks.sumfrom above; all keephmass : 56 ≤ ks.sum)."For the frontier words" means: the frontier words satisfy all the hypotheses of these theorems (computed,
NATIVES.md, a table of premise × word → result and script). Checked on all four ofst1(#7641), the #7637 word,x1t2aand the #7660 word: |J^arr| = 4962, rank M = 11, D = 24, the LawAModel premises, hmono, hd9, hmix = 5 (tight), hk, hmass, hmd, hmr, hY, hF, hEex, hst, the Cap class, and the legal-path and Cap-margin ingredients of hcong / hdist. They share one owned layer and one port set, so this is close to a single data point. The link from the Lean invoice to each word's pricing is the pricing script's. Nothing in Lean ties a word to the hypotheses.Law A as proved in Lean is
lawA_96: form : LawAModel K 104 DwithD ≤ 56(premiseshspan,hdiag,hodd,hpart ≤ 48, P = 104),96 ≤ m.J. Behind it,maisBound(Caro-Wei, needshdiag) gives Σ_t 1/(j_t+1) ≤ rank M, and rank M ≤ D byhspan. The general inequality J ≥ 2(P − D) is not stated in Lean. All theorems print only propext, Classical.choice, Quot.sound; nosorry, nonative_decide(CHECK.txt).Caveats, stated plainly.
hD,hJdef). Those are not checked for any such word, and hmix = 5 is tight on the known words. The same holds for a word with D > 56 and T2b.corrC ≤ 8 · cH,capC ≥ unLo / 48). Decoders with 9 or more rows are outside the model; rank 9, 10, 11 matrices that need that many rows are not covered.hJdef : J = junkSet m.Mis not a naming link only: it forceshcongandhdist(real-valued statements) to hold on exactly the off-diagonal support ofm.M. In T1,Jis any finset withhJ; in T2b,JanddistFare still quantified andhcong,hdistare assumed (even forJ = ∅). Lean does not tieJtoJ^arrof a word.hcong,hdist, the spF / cst brackets and the monotonicityhspMono,hcstMono(a paper theorem) are hypotheses in Lean, as are the Law A premises.NATIVES.mdcomputes their ingredients on the four words; the step from the computations to the Lean statements is not formalised.hDin T2a is the size of any spanning set of arrival vectors, so in effect rank M ≤ 56 (the words have rank 11, D = 24). Intermediate ceilings for 56 < D < 104 were not built.CHECK.txt: 65 of the 66 modules were compiled fresh.CongestionI3Cfg(unchanged from P104 Lean package 3: conditional ceilings (.0204 for nested Priced words, class K .00602), word structure and duality, spec-break lemmas (no κ) #7656, where it built fresh) was stopped by the 3.5 GB memory guard at about 3.44 GB, twice, so its prebuilt olean from the earlier build was used for the modules that import it.Premises under study are SYLforge's (#7616, #7620); the caller is rohanarun's (#7113). DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code.
🤖 Generated with Claude Code