Skip to content

📐 GJ §17.5 Thm 17.5.1 lsc half — random-current OZ two-point lower bound (thread) — subissue of #4386 #4418

Description

@phasetr

PR Checklist — SL-A to SL-E Lemma 5.1 Ingredient Thread

2026-07-12 GOVERNANCE NOTE (dev-issue-manager, OFF-BOOK): This entire thread targets GJ's literal "everywhere continuity" closing sentence of Thm 17.5.1, which GJ's own printed proof (pp. 311–312) does not itself establish (Lipschitz m⁻ + Lemma 17.5.2 sandwich only, constant C>1 never shown to collapse). docs/index.md:1764–1771 labels every OZ Wall-#2 brick produced by this thread "off-book optional", each row stating primary GJ Thm 17.5.1 rigorous content (gj_theorem_17_5_1_rigorous) is already done. No "Ornstein-Zernike"/"Simon-Lieb" theorem exists in the GJ book body (bibliography-only hit). This thread is OFF-BOOK, not an in-scope GJ-book proposition; see #4405 "GOVERNANCE RECONSTRUCTION" and #4259 Next-concrete-step for the STOP-and-ask finding (do not resume M2/B4 without a fresh explicit user authorization decision).

Purpose: Track the staged build of Lemma 5.1 (GJ eq. (17.5.1), random-current weight edge-partition factorization) toward the lsc half of Theorem 17.5.1.

Thread: Part of Group 1a (random-current / Simon-Lieb D-family explicit tracking for hLogLip gate relocation).

Status: SL-D₁ COMPLETE (SL-A through SL-D₁b all MERGED, axiom-free); SL-D₂ CORRECTED + SUPERSEDED (2026-07-12, .self-local/tex/rc-oz-lemma51-SLD2-inequality.tex) — user provided Aizenman 1982 (Commun. Math. Phys. 86, 1-48) PDF at ~/junk/0-pdf-for-translation/Aizenman.1982.p48.*.pdf and directed engagement (user's own concrete provisioning act, not agent self-authorization). TRACKING CORRECTION: "Aizenman Lemma 4.1" does NOT exist in the source (verified pdftotext full-text search: zero hits); the subgraph-conditioning content is Aizenman §9, Lemma 9.2 eq.(9.11) p.25 + Lemma 9.3(iii)/(iv) eq.(9.13)-(9.14) pp.25-26 (both Griffiths/GKS-monotonicity, inequality-driven, not an exact switching identity) and §3 Lemma 3.2 p.7 (combinatorial switching identity, used elsewhere in the paper, not for this ingredient). math-before-code re-derivation (rc-oz-lemma51-SLD2-inequality.tex) further found: (a) hLogLip is a pure upper bound (log-derivative = nonnegative excess, already proven by merged Current.hasDerivAt_log_correlation_beta), so SL-D₂ only needs the one-sided inequality version, which closes on existing merged repo infra (correlation_deleteEdges_le = Aizenman Lemma 9.3; correlation_inducedGraph_eq_weightSum_ratio re-instantiated = Aizenman Lemma 9.2) — it is not an irreducible research core; (b) the entire SL-A/B/C/D pivotal-fiber route is in fact bypassable for hLogLip via two already-merged bricks (Current.doubledSourcefree_edgeExcess_eq_truncated4pt + correlation_beta_deriv_le_lebowitz); the genuine remaining research core is the OZ two-point-ratio summability bound (1/⟨σxσy⟩)Σₑ(⟨σxσu⟩⟨σyσv⟩+⟨σxσv⟩⟨σyσu⟩) + |E|/⟨σxσy⟩ ≤ (K/J)·d(x,y) — a convolution/decay estimate, independent of SL-D₂. SL-E gated on this OZ ratio bound, not on SL-D₂.


✅ Completed Ingredients

✅ SL-A/B/C/D₁/M1 = mainline infra (not wasted work)

SL-A (edge-partition), SL-B (component extraction), SL-C (avoiding-set/bridge-uniqueness), SL-D₁ (range-independence, weight factorization Σ_C=(βJ)·Ξ_int·Ξ_ext) and M1 (Lebowitz convolution upper bound) are standing, axiom-free, reusable infra for the random-current / OZ route; they are not superseded/deleted, only advanced on the critical path. M1 discharges the false "crude Lebowitz blow-up" diagnosis and reduces hLogLip to the genuine remaining core.

⏳ Pending / On-going — TRUE REMAINING CORE = OZ two-point-ratio summability (NOT SL-D₂)

  • SL-D₂ (subgraph-conditioning, exterior collapse): re-framed 2026-07-12 as a one-sided inequality Ξ_ext ≤ ⟨σbσy⟩_Λ·weightSum^{C^c}(∅), which closes on already-merged repo infra (correlation_deleteEdges_le + correlation_inducedGraph_eq_weightSum_ratio re-instantiated) modulo bounded ↑Λ-vs-V subtype plumbing (SL-D₂(a), moderate, not a research core). SL-D₂ is off the hLogLip critical path (see below) but remains available as an alternative route if needed.
    • SL-D₂.0 math-before-code CORRECTED (2026-07-12): .self-local/tex/rc-oz-lemma51-SLD2-inequality.tex — sourced against the actual Aizenman 1982 PDF (§9 Lemma 9.2/9.3, not the non-existent "Lemma 4.1"); supersedes the equality-framed rc-oz-lemma51-SLD-collapse.tex verdict.
    • SL-D₂(a) plumbing (optional, only if pivotal route retained): Current.exteriorWeightSum_le_correlation_mul_weightSum_empty, bounded subtype/edge-set identification.
  • ▶ TRUE REMAINING CORE (genuine research residual, on the critical path): the OZ two-point-ratio summability bound
    (1/⟨σxσy⟩)·Σₑ(⟨σxσu⟩⟨σyσv⟩+⟨σxσv⟩⟨σyσu⟩) + |E|/⟨σxσy⟩ ≤ (K/J)·d(x,y)
    (GJ p.312 OZ content: two-leg decay convolution is O(d(x,y)) uniformly). This is NOT SL-D₂ and NOT the pivotal fiber; it is reached directly from two already-merged bricks (Current.hasDerivAt_log_correlation_beta + correlation_beta_deriv_le_lebowitz).
    • ✅ M1 MERGED (PR feat(gj-17.5-oz-M1): Lebowitz convolution upper bound on log-derivative #4495, main cf1909a0): Route-B Lebowitz convolution upper bound on log-derivative∂_β log⟨σxσy⟩ ≤ (J/⟨σxσy⟩)·∑_e L_uvxy (OZ ratio reduction, outer-shell brick, axiom-free via merged infra). Discharges false "crude Lebowitz blow-up" diagnosis from earlier design, reduces target to OZ summability. New file IsingModel/Inequalities/SourcefreeConnectionLebowitzConvolutionBound.lean. M1 does NOT stand alone; M2 (sharp lower-bound two-point decay, multi-session research core) + SL-E capstone required to close.
    • ▶ M2 IN QUEUE: Sharp matching two-point lower bound (OZ 壁; genuine multi-session research) = the OZ two-point-ratio summability bound itself. Convolution/decay estimate (GJ p.312, FFS Ch.12 backbone-tail content; not in Aizenman 1982). Bypasses SL-D₂ entirely. Upon M2 discharge + SL-E capstone assembly, Lemma 5.1/hLogLip closes.
      • M2 sub-checklist (sharp matching two-point lower bound = OZ summability; genuine multi-session research):
        • B1 (degenerate-edge Lebowitz bound via GKS-II + tanh, axiom-free, PR feat(gj-17.5-oz-M2-B1): degenerate-edge Lebowitz bound via GKS-II (axiom-free) #4496, main ec186b46): degenerate edges (u or v ∈ {x,y}) contribution S_deg ≤ (deg(x)+deg(y))·J/tanh(βJ)·d(x,y) via existing infra (gks_second + single-edge tanh bound); axiom-free, hedge is hypothesis (not axiom); does NOT close M2 (B2/B3/B4 remain).
        • B2 — RETRACTED/FALSE for d≥2 (design report .self-local/reports/design-oz-M2-nondegenerate.md, 2026-07-12): off-axis crossed-edge geometric tube sum. The claimed poly(0)·D tube bound is FALSE: latticeDistance is ℓ¹ (non-unique fat geodesics); excess-0 shell alone has ∏_i(|Δ_i|+1) points (e.g. d=2 diag: (n+1)²), so ∑_a ρ^{d(a,x)+d(a,y)} = ρ^D·∏_i(|Δ_i|+1+c) is O(D^d), not O(D). Do NOT commit as designed.
        • B3 — RETRACTED (design report, 2026-07-12): dimension-dependent non-degenerate reduction does NOT close even granting H_low as hypothesis: substituting H_low only removes the (ρ_+/ρ)^D factor; the residual ∏_i(|Δ_i|+1+c) is still O(D^d) (AM-GM), so S_nd ≤ K·D is unreachable via this route. B3 is not H_low-gated-closable; it depends on the (false) B2 tube bound.
        • B4 (matching two-point lower bound H_low, genuine OZ 壁): the irreducible research core — two-point decay matching ⟨σxσy⟩ ≥ c·ρ^{d(x,y)} (rate ρ ≥ 2d·tanh(βJ)) uniformly at the summation rate. GJ p.312 / OZ bubble / FFS Ch.12 backbone-tail content; binding-pair pseudo-mass deriv infra structurally incapable. From-scratch multi-session research (3-6 PR, Aizenman §9 / FFS 12 / repo-absent bottleneck).
        • ⚠️ hedge/B0 (single-edge tanh lower bound ⟨σ_u σ_v⟩ ≥ tanh(βJ)): Unconditional discharge requires gated refactor (private-helper relocation; no inline axiom/sorry); do NOT attempt direct axiomatization. B1 uses hedge as hypothesis (present route, no obstacle).
    • Then: OZ summability design: fresh dev-design for (E1) geometric two-point decay input, (E2) matching lower-leg bound (twoPointFunction_ge_tanh_betaJ_pow_dist, already present), (E3) convolution/summability count — (E3) is the genuine multi-session OZ core (M2 B1–B4 component assembly).
    • Then: Staged implementation PRs per design, each axiom-free / Tier-1-clean / reviewed, scoped strictly to Group 1a Lemma 5.1 / hLogLip.
  • SL-E (capstone assembly): binding the OZ ratio bound → hLogLip → §17.5.1 lsc. Gated on the OZ summability core above, not on SL-D₂.

Roadmap Note

Scope: Group 1a (Lemma 5.1 SL-A→E) only; no spillover to Group 1b/1c or untracked research. Each ingredient lands axiom-free/Tier-1-clean before next.

Canonical status: #4259 (Current status + Next concrete step); #4405 (Group 1a section).

2026-07-12 CORRECTION (issue-manager governance — B2/B3 design finding, .self-local/reports/design-oz-M2-nondegenerate.md): Independent design verification of M2's non-degenerate decomposition found the prior "B2/B3 no new hypothesis / YES-graded" checklist entries and the tex .self-local/tex/rc-oz-lemma51-M2-convolution-estimate.tex §sec:verdict grading (B2="YES", B3="YES given H_low") to be WRONG (overclaim), retracted:

  • B2 is FALSE for d≥2 (not a difficulty-graded-low brick, the claimed proposition is mathematically false): latticeDistance is ℓ¹, whose geodesics are non-unique/"fat" (multinomial count); the excess-0 shell alone between x,y has ∏_i(|Δ_i|+1) points (e.g. d=2, y=(n,n): (n+1)² points), so the two-focus vertex-convolution generating function ∑_a ρ^{d(a,x)+d(a,y)} = ρ^{D}·∏_i(|Δ_i|+1+c) is O(D^d), not the claimed poly(0)·D tube bound.
  • B3 does not close even granting H_low as hypothesis: substituting H_low into the ratio only kills the (ρ_+/ρ)^D factor; the residual ∏_i(|Δ_i|+1+c) term is still O(D^d) (AM-GM), not O(D), so S_nd ≤ K·D is NOT obtained. B3 is not "H_low-gated axiom-free"; it is unreachable by this route regardless of hypothesis.
  • H_low itself is suspect as stated (rate ρ≥ρ_+=2d·tanh(βJ), ℓ¹ metric): the true on-axis decay rate is ~tanh(βJ) (directed-path/SAW dominated), far below ρ_+; H_low as ⟨σxσy⟩ ≥ c₀·ρ_+^{D} is false for large D under the true asymptotics. The log(2d) gap between the merged upper rate (2d·tanh, ℓ¹ SAW-entropy overcount) and the merged easy lower rate (tanh) is an artifact of upper-bound looseness (isotropic ℓ¹ overcounting), not a missing matching lower bound; the correct fix direction is a sharper anisotropic/Euclidean-rate upper bound, not H_low.
  • Genuine core: the non-degenerate S_nd bound requires the true OZ mechanism (Euclidean rate + diffusive √-width tube/prefactor) — i.e., §18 OZ machinery itself, not an elementary H_low-gated reduction. This is consistent with, and further specifies, the pre-existing MEMORY finding that "hLogLip = §18 walk-sum content, multi-session, no clean 1-PR entry."
  • B1 (degenerate part, PR feat(gj-17.5-oz-M2-B1): degenerate-edge Lebowitz bound via GKS-II (axiom-free) #4496 merged) and M1 (PR feat(gj-17.5-oz-M1): Lebowitz convolution upper bound on log-derivative #4495 merged) are unaffected and remain correct — GKS-II route, no H_low dependency, no ℓ¹ tube issue.
  • Overreach boundary: B2/B3 MUST NOT be committed as designed (would formalize a false/broken proposition). The tex rc-oz-lemma51-M2-convolution-estimate.tex §sec:bricks/§sec:verdict grading of B2/B3 as "YES" and the H_low framing require correction by lean-math-scribe/dev-docs-sync (not performed here; issue-manager records the fact only). Next step for M2 is re-scoping the non-degenerate core directly onto the genuine §18 OZ body (anisotropic sharp upper bound / true OZ convolution) — this is multi-session from-scratch research with no clean 1-PR entry, requiring further explicit user authorization/scoping, not autonomous continuation.

2026-07-12 GOVERNANCE FINAL — PERMANENTLY BANNED (explicit user directive, supersedes all prior authorization)

User directive (this session, verbatim intent): "現時点で異常だとわかっている内容はもう絶対に触れない" — this entire thread (OZ / Ornstein-Zernike / Simon-Lieb / SL-D₂ / M2 B2-B3 / hLogLip-via-OZ) is PERMANENTLY BANNED, effective immediately, superseding the "SL-D₂ AUTHORIZED" / "M2 IN QUEUE" state recorded above. Do not resume SL-D₂, M2-B2/B3/B4, or any further ingredient in this thread under any future autonomous /goal reissuance — a fresh, explicit, unambiguous user re-authorization naming this thread specifically is required, not implied by general "continue" instructions.

Disposition: this issue stays OPEN (close/reopen is the user's decision, not issue-manager's), but is retired from the active work queue. SL-A/B/C/D₁/M1/M2-B1 remain merged, axiom-free, standing infra (not deleted/reverted) — only further work on this thread's remaining ingredients (SL-D₂, M2-B2/B3/B4, SL-E) is banned.

Replacement route: GJ Thm 17.5.1 everywhere-continuity is now pursued via the §18 cluster-expansion / window-analyticity route, tracked at #4386 (redefined 2026-07-12) — a route independent of everything in this thread. See #4405 item 1a′ and #4259 Next concrete step.


2026-07-12 (cont.): dev-issue-manager governance — PARKED at honest capstone (user directive), both candidate primary sources tested insufficient

Two user-provided candidate primary sources for closing the OZ wall have now been tested and found insufficient:

  • Aizenman 1982: BLACK (.self-local/reports/research-aizenman1982-oz-feasibility.md, .self-local/tex/aizenman1982-gapcloser-verdict.tex) — zero magnetic field, one-sided/wrong-rate bounds only, m(β) continuity merely cited to Simon, not proved.
  • CIV03 (Campanino–Ioffe–Velenik 2003): fixed-β sharp OZ asymptotic WHITE (over-delivers), but β-continuity of the mass BLACK (never stated/proved in the paper) and entirely h=0 (BLACK for ∂/∂h) (.self-local/reports/research-civ03-oz-feasibility.md, .self-local/tex/civ03-gapcloser-verdict.tex). Prerequisites (countable-alphabet RPF theory, Kato perturbation, Gibbs-Markov LLT) absent from mathlib/repo; realistic scale 13–23 PR-blocks even before addressing the missing β-continuity extension.

Per explicit user directive ("調査結果を明記の上で区切る"), this thread (#4418) is PARKED, not closed (close/reopen remains the user's decision). SL-A/B/C/D₁/M1/M2-B1 remain merged as standing axiom-free infra (not reverted, not further extended). SL-D₂/M2-B2/B3/B4/SL-E remain banned/parked, not resumed. Reopening requires a new constructive β-analyticity reference or explicit multi-month from-scratch authorization (per #4386/#4405 parking terms), not a general re-issuance of /goal. Full detail: #4386, #4405.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions