From e2df40f254a50b88f607ffd3091f5ca805565e4a Mon Sep 17 00:00:00 2001 From: Claude Date: Tue, 29 Sep 2026 12:26:42 +0000 Subject: [PATCH 1/5] math(candidate): C7 disposition - the unrestricted selection difference does not vanish (scientific effect NONE) CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v1. Theorem U: for 0 < ell <= 1 the far candidate-minus- elder density of the D1 parent is at least c_* > 0, so nu_cand^all - nu_eld^all >= c_* and the registry candidate C7 (an unrestricted O(ell^(2/3)) difference) is false; with Theorem C the difference is Theta-below, o(ell^(-1/3)) above. Mechanism: a level set below the saddle height on a small sphere around the maximum forbids pairing; its positive marked Kac-Rice intensity follows from the Gaussian support theorem (all Fourier weights positive) and lower semicontinuity. Corollaries (U3) cumulative >= c_* t and (U4) infinite nonselected l^q sums for q <= -1. Adds PROOF.md, an exact witness/ledger companion (3 mutants, identical in -O), RESULTS.json, SOURCE_FILES.json pinning the parent bytes, and a path-filtered workflow. No register, status, graph, lemma flag or prize changes; nonauthor review required. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_017Mi3hxjaxV45x6zo6o1ee3 --- .../unrestricted-selection-difference.yml | 56 ++++++ .../PROOF.md | 160 ++++++++++++++++++ .../RESULTS.json | 39 +++++ .../SOURCE_FILES.json | 34 ++++ .../witness_check.py | 88 ++++++++++ 5 files changed, 377 insertions(+) create mode 100644 .github/workflows/unrestricted-selection-difference.yml create mode 100644 frontiers/unrestricted_selection_difference_20260929/PROOF.md create mode 100644 frontiers/unrestricted_selection_difference_20260929/RESULTS.json create mode 100644 frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json create mode 100644 frontiers/unrestricted_selection_difference_20260929/witness_check.py diff --git a/.github/workflows/unrestricted-selection-difference.yml b/.github/workflows/unrestricted-selection-difference.yml new file mode 100644 index 000000000..88038843d --- /dev/null +++ b/.github/workflows/unrestricted-selection-difference.yml @@ -0,0 +1,56 @@ +name: Unrestricted selection-difference candidate checks + +'on': + pull_request: + paths: + - 'frontiers/unrestricted_selection_difference_20260929/**' + - '.github/workflows/unrestricted-selection-difference.yml' + workflow_dispatch: + +permissions: + contents: read + +jobs: + witness: + runs-on: ubuntu-latest + timeout-minutes: 10 + steps: + - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 + with: + persist-credentials: false + - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 + with: + python-version: '3.11.16' + - name: Verify sources, replay both modes, reject mutants + env: + PYTHONDONTWRITEBYTECODE: '1' + run: | + set -euo pipefail + python -B -S - <<'PY' + import hashlib, json, pathlib, subprocess, sys + root = pathlib.Path('frontiers/unrestricted_selection_difference_20260929').resolve() + manifest = json.loads((root / 'SOURCE_FILES.json').read_text()) + present = sorted(p.name for p in root.iterdir()) + expected_names = sorted([e['path'] for e in manifest['files']] + ['SOURCE_FILES.json']) + if present != expected_names: + raise ValueError('packet tree differs from manifest: ' + repr(present)) + for entry in manifest['files'] + manifest['consumes']: + path = root / entry['path'] if '/' not in entry['path'] else pathlib.Path(entry['path']).resolve() + data = path.read_bytes() + if len(data) != entry['bytes'] or hashlib.sha256(data).hexdigest() != entry['sha256']: + raise ValueError('source identity mismatch: ' + entry['path']) + script = str(root / 'witness_check.py') + expected = (root / 'RESULTS.json').read_bytes() + for flags in (['-B', '-S'], ['-B', '-O', '-S']): + out = subprocess.run([sys.executable, *flags, script], + check=True, capture_output=True, timeout=120).stdout + if out != expected: + raise RuntimeError('output differs from RESULTS.json under ' + ' '.join(flags)) + for m in ['sign-alpha', 'no-separation', 'compact-rate']: + rc = subprocess.run([sys.executable, *flags, script, '--mutant', m], + capture_output=True, timeout=120).returncode + if rc != 1: + raise RuntimeError('mutant ' + m + ' not rejected') + print(json.dumps({'scientific_effect': 'NONE', 'mathematical_acceptance': False})) + PY + git diff --exit-code diff --git a/frontiers/unrestricted_selection_difference_20260929/PROOF.md b/frontiers/unrestricted_selection_difference_20260929/PROOF.md new file mode 100644 index 000000000..928151def --- /dev/null +++ b/frontiers/unrestricted_selection_difference_20260929/PROOF.md @@ -0,0 +1,160 @@ +# The unrestricted selection difference does not vanish: disposition of candidate C7 + +Object: CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v1. +Author: Anthropic Claude (Claude Code session `session_017Mi3hxjaxV45x6zo6o1ee3`). +Disposition: AUTHOR-SIDE CANDIDATE; nonauthor analytic review required. +Scientific effect: NONE. No `STATUS`, `PROOF_INDEX`, `GRAPH`, claim, lemma flag, prize or source body changes. + +## 1. The candidate and the result + +The D1 parent `imports/lifetime_parent_20260925/UNIFORM_MATRIX_CAP_AND_LIFETIME.md` (blob `dfed3b8d…`, see +SOURCE_FILES.json; reconciled in Math-#126) proves two things: +- in Theorem B, the compact-window difference `0 ≤ ν_cand(ℓ) − ν_eld(ℓ) ≤ C_{B,K} ℓ^{2/3}` (1.2); +- in Theorem C, the unrestricted leading asymptotics `ν_cand^all(ℓ) ~ ν_eld^all(ℓ) ~ c_{d,L} ℓ^{−1/3}` (1.3). + +It states explicitly that (1.2) is not asserted for the unrestricted densities (lines 52, 384, 433). Candidate C7 of +`reviews/candidates_pending_20260928/CANDIDATES.md` proposes the unrestricted rate + + 0 ≤ ν_cand^all(ℓ) − ν_eld^all(ℓ) ≤ C ℓ^{2/3}. (C7) + +Here `ν_cand^all` counts all ordered local-maximum / index-`(d−1)`-saddle pairs with height difference `ℓ`, and +`ν_eld^all` counts all finite superlevel `H_0` bars. Both are expected densities per unit volume on the fixed torus +`X = ℝ^d/(Lℤ^d)`, `d ≥ 2`, for the parent's periodized Gaussian field. + +**Theorem U.** Fix `0 < r_0 < L/4` as in parent §13 and put `ρ = r_0/2`. There are `c_* > 0` and `ℓ_0 > 0` such +that for `0 < ℓ ≤ ℓ_0` + + ν_cand^all(ℓ) − ν_eld^all(ℓ) ≥ ν_cand^far(ℓ) − ν_eld^far(ℓ) ≥ c_*, (U1) + +where "far" means torus distance `≥ r_0` between the two points, as in parent §14. Consequently: +- **(U2)** (C7) is false. `D(ℓ) := ν_cand^all(ℓ) − ν_eld^all(ℓ)` satisfies `D(ℓ)/ℓ^α → ∞` for every `α > 0`. + Combined with parent (14.1) and Theorem C, `c_* ≤ D(ℓ) = o(ℓ^{−1/3})`. +- **(U3)** The expected per-unit-volume number of candidate pairs with lifetime in `(0, t]` that are not bars is at + least `c_* t`. In the compact window, parent (12.1) gives at most `C t^{5/3}`. +- **(U4)** For `q ≤ −1`, the nonselected sum `E Σ_{candidate, not bar, ℓ ≤ t} ℓ^q` is infinite for every `t > 0`. In + the compact window, parent (12.3) makes it finite for every `q > −5/3`. + +The obstruction comes from pairs at macroscopic separation. Among pairs at distance `≥ r_0` with small height +difference, a positive density are separated by a level curve and so cannot be persistence partners. Theorem U does +not decide whether `D` is bounded (§5). + +## 2. Kac–Rice form of the far difference + +Parent §14 writes the far candidate density from the original full-pin covariance, which is uniformly nondegenerate +on `{dist(x,y) ≥ r_0}`: + + ν_cand^far(ℓ) = L^{−d} ∫∫_{dist(x,y) ≥ r_0} ∫_ℝ p_{x,y}(b, b−ℓ) + E[F_d(H_x) F_{d−1}(H_y) | 𝒫_{x,y}(b, ℓ)] db dx dy, (2.1) + +Here: +- `𝒫_{x,y}(b, ℓ)` is the pin event `{∇f(x) = ∇f(y) = 0, f(x) = b, f(y) = b − ℓ}`; +- `p_{x,y}` is the joint density of `(∇f(x), ∇f(y), f(x), f(y))` at `(0, 0, b, b−ℓ)`; +- `F_j` is the absolute determinant restricted to index `j`; +- the height map `(f(x), f(y)) ↦ (b, b−ℓ)` has Jacobian one. + +The mark below is the indicator of an open set of `(x, y, f)`, hence lower semicontinuous. The same +arXiv:2304.07424v3 Theorem 7.1 interface used in parent §9 and in [RC] (3.1) therefore gives, for the count of far +pairs carrying the mark, + + N_O(ℓ) = L^{−d} ∫∫_{dist ≥ r_0} ∫ p_{x,y}(b, b−ℓ) + E[F_d(H_x) F_{d−1}(H_y) 1_{O_{x,y,ℓ}}(f) | 𝒫_{x,y}(b, ℓ)] db dx dy, (2.2) + + O_{x,y,ℓ} = { f : sup_{∂B(x,ρ)} f < f(y) }. + +## 3. A separating level curve forbids pairing + +**Lemma 1.** Let `f` be `C²`, let `x` be a local maximum and `y` an index-`(d−1)` saddle with `dist(x, y) ≥ r_0`, and +suppose `f ∈ O_{x,y,ℓ}`. Then `(x, y)` is not an `H_0` persistence pair of the superlevel filtration. + +*Proof.* If `(x, y)` were a pair, the bar born at `x` would die at `y`. So for every `t` slightly above `f(y)`, the +component `C_t` of `{f > t}` containing `x` would have `y` in its closure. Every such `t > f(y)` also +satisfies `t > sup_{∂B(x,ρ)} f`. Then `C_t` is connected, contains `x` and does not meet `∂B(x, ρ)`. Since +`ρ < r_0 < L/4`, the ball is embedded and `∂B(x,ρ)` separates it from its complement, so `C_t ⊂ B(x, ρ)`. Also +`f < t` on `∂B(x, ρ)`, so `cl(C_t) ⊂ B(x, ρ)`, which does not contain `y`. ∎ + +Hence the far candidate pairs carrying the mark in (2.2) are not bars, and + + ν_cand^far(ℓ) − ν_eld^far(ℓ) ≥ N_O(ℓ). (3.1) + +## 4. The marked intensity is bounded below + +**Lemma 2 (support).** For all `x, y` with `dist(x, y) ≥ r_0`, all real `b` and all `ℓ ∈ [0, 1]`, the conditional +law `Q_{x,y,b,ℓ}` of `f` given `𝒫_{x,y}(b, ℓ)` gives positive probability to the open set + + 𝒪 = { f ∈ C²(X) : H_x < 0, H_y nonsingular of index d−1, sup_{∂B(x,ρ)} f < f(y) }. + +*Proof.* `Q = Q_{x,y,b,ℓ}` is a Gaussian measure on the separable Banach space `C²(X)`. Its mean `m` is smooth and +satisfies the pins. By the Gaussian support theorem, its support is `m + cl_{C²}(H_0)`, where `H_0` is the +Cameron–Martin space of the unconditioned field restricted to `h(x) = h(y) = 0`, `∇h(x) = ∇h(y) = 0`. + +Every Fourier weight is positive (parent §2), so `H_0` contains every trigonometric polynomial satisfying those +`2(d+1)` linear conditions. Their `C²`-closure is `V_0 = {φ ∈ C² : φ, ∇φ vanish at x and y}`. To see this, take +`φ ∈ V_0` smooth and trigonometric polynomials `τ_n → φ` in `C²`. The `2(d+1)` point functionals are linearly +independent on trigonometric polynomials (parent §2, distinct-site jets), so there are trigonometric polynomials `ψ_i` +dual to them. Then `τ_n − Σ_i λ_i(τ_n) ψ_i` lies in `H_0` and converges to `φ`, because `λ_i(τ_n) → λ_i(φ) = 0`. + +Choose a smooth `G` with: +- `G ≡ b − ℓ − 1` outside the two disjoint balls `B(x, ρ/2)` and `B(y, ρ/2)`, whose closures are disjoint since + `dist(x, y) ≥ 2ρ`; +- `G(x) = b`, `∇G(x) = 0`, `D²G(x) = −I`; +- `G(y) = b − ℓ`, `∇G(y) = 0`, `D²G(y) = diag(−1, …, −1, +1)`. + +Bump functions give such a `G`. Then `G − m ∈ V_0`, since both satisfy the pins. Also `G ∈ 𝒪`, since +`sup_{∂B(x,ρ)} G = b − ℓ − 1 < G(y)`. So `G` lies in the support, and `𝒪`, which is open in `C²`, has positive +`Q`-probability. ∎ + +**Lemma 3 (uniformity).** Put `Φ(x, y, b, ℓ) = E_Q[F_d(H_x) F_{d−1}(H_y) 1_𝒪]`. It is lower semicontinuous and +strictly positive on `K = {dist(x, y) ≥ r_0} × [0, 1] × [0, 1]`, so `Φ ≥ φ_* > 0` on `K`. + +*Proof.* Realize every `Q_{x,y,b,ℓ}` on one unconditioned field `F` by regression on the `2(d+1)` observations. The +observation covariance is uniformly nondegenerate on `K` (parent §14), so the regression coefficients and targets are +continuous and the conditioned fields converge pathwise in `C²` along any convergent parameter sequence. The weight +`F_d F_{d−1}` is continuous, and `1_𝒪` is lower semicontinuous because `𝒪` is open and `(x, y) ↦ 𝒪` varies +continuously. Fatou's lemma gives lower semicontinuity of `Φ`. It is positive: on `𝒪` the weight is positive, and by +Lemma 2 `𝒪` has positive probability. A positive lower semicontinuous function on a compact set is bounded below by +a positive constant. ∎ + +**Proof of Theorem U.** On `K` the density `p_{x,y}(b, b−ℓ)` is continuous and positive, hence `≥ p_* > 0`. By +(2.2), restricting `b` to `[0, 1]`, + + N_O(ℓ) ≥ L^{−d} p_* φ_* |{(x, y): dist(x, y) ≥ r_0}| =: c_* > 0, 0 < ℓ ≤ 1. + +With (3.1) this proves (U1), with `ℓ_0 = 1`, and (U2). + +For (U3), integrate (U1) over `(0, t]`. For (U4), the nonselected measure has density at least `c_*` on `(0, ℓ_0]`, +and `∫_0 ℓ^q dℓ = ∞` for `q ≤ −1`. The bound `D = o(ℓ^{−1/3})` is Theorem C, since both densities are +`c_{d,L} ℓ^{−1/3}(1 + o(1))`. ∎ + +`witness_check.py` gives an explicit exact instance of the separating configuration, a trigonometric polynomial on +`(ℝ/2πℤ)²`: + + g(u, v) = α cos u + cos² u + cos v, α = ℓ/2 − 1. + +It has a maximum at `(0, 0)` and an index-1 saddle at `(π, π)` with height difference exactly `ℓ`. On the boundary of +the square `|u|, |v| ≤ a` with `cos a = −α/2` (rational), `g < g(π, π)`. The script checks this for three values of +`ℓ` in exact rational arithmetic. It illustrates Lemma 2; it is not a substitute for it. + +## 5. What is and is not settled + +- **Settled.** (C7) fails. The unrestricted difference is bounded below by a positive constant, and it is + `o(ℓ^{−1/3})`. +- **Not settled: the order of `D`.** Near pairs at distance `δ ≤ r_0` contribute + `∫_0^{r_0} δ^{−2} ∫ A_δ(b, ℓδ^{−3}, u)(1 − p_δ(b, ℓδ^{−3}, u)) db dσ dδ`. This is parent (11.2) after the + substitution `k = ℓδ^{−3}`. Macroscopic `δ` gives another `Θ(1)` term. Whether `D` stays bounded depends on the + nonselection probability `1 − p_δ` in the degenerate regime of gap mark `k = ℓδ^{−3} → 0`, which no source + controls. A heuristic balance puts the crossover near `δ ≍ ℓ^{1/4}`. It is not a claim. +- **The compact-window rate (1.2) is untouched.** Restricting marks to compact `B × K` with `inf K > 0` and distances + to `≤ r_*` removes the obstruction, because such pairs have `r = (ℓ/k)^{1/3} → 0`. The natural corrected target is + the rate with unbounded birth mark and gap marks bounded away from zero. It needs the parent's selection estimate + with mark-dependent constants and is left open. +- No numerical `c_*`, no event-probability statement, and no status change. + +## 6. Dependencies + +- Parent §2 (positive Fourier weights, distinct-site jet rank), §9 (weighted Kac–Rice interface), §13 (`r_0`) and + §14 (far covariance, height Jacobian one, (14.1)). +- Theorem C (1.3), as reconciled in Math-#126 (`8224783`). +- The Gaussian support theorem and the lower-semicontinuous-mark Kac–Rice formula (arXiv:2304.07424v3, Theorem 7.1), + the same interface as parent §9 and [RC] (3.1). + +Nothing from the elder-selection upper estimate of Theorem A is used. diff --git a/frontiers/unrestricted_selection_difference_20260929/RESULTS.json b/frontiers/unrestricted_selection_difference_20260929/RESULTS.json new file mode 100644 index 000000000..8efdfcd18 --- /dev/null +++ b/frontiers/unrestricted_selection_difference_20260929/RESULTS.json @@ -0,0 +1,39 @@ +{ + "checks": { + "LEDGER": { + "U3_linear_lower_beats_compact_t_five_thirds": true, + "U4_q_minus1_equal_shells_diverge": true, + "U4_q_minus2_growing_shells": true, + "q_minus_half_summable_ratio": true + }, + "WITNESS": { + "1/10": { + "M_nondegenerate_maximum": true, + "P_increasing_on_cm_1": true, + "S_nondegenerate_index1_saddle": true, + "boundary_strictly_below_saddle_height": true, + "height_difference_is_l": true, + "square_half_width_below_pi_over_2": true + }, + "1/100": { + "M_nondegenerate_maximum": true, + "P_increasing_on_cm_1": true, + "S_nondegenerate_index1_saddle": true, + "boundary_strictly_below_saddle_height": true, + "height_difference_is_l": true, + "square_half_width_below_pi_over_2": true + }, + "1/1000": { + "M_nondegenerate_maximum": true, + "P_increasing_on_cm_1": true, + "S_nondegenerate_index1_saddle": true, + "boundary_strictly_below_saddle_height": true, + "height_difference_is_l": true, + "square_half_width_below_pi_over_2": true + } + } + }, + "object": "CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v1", + "passed": true, + "scope": "exact finite witness and ledger only; not the Gaussian proof" +} diff --git a/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json b/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json new file mode 100644 index 000000000..3594ac5e8 --- /dev/null +++ b/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json @@ -0,0 +1,34 @@ +{ + "schema": 1, + "scientific_effect": "NONE", + "disposition": "AUTHOR_SIDE_CANDIDATE", + "consumes": [ + { + "path": "imports/lifetime_parent_20260925/UNIFORM_MATRIX_CAP_AND_LIFETIME.md", + "bytes": 40261, + "sha256": "9350ad6eaba6626b93c3dedeef9e2ff816e5cdf1c8318e85fb27499141c84bc7", + "git_blob": "dfed3b8d318a3ab1950957f393307733a4bef3f2", + "use": "sections 2, 9, 13, 14 and Theorem C (1.3)" + } + ], + "files": [ + { + "path": "PROOF.md", + "bytes": 10473, + "sha256": "827d718a664ebae4edd05f8c63dcf10bcb3ea5cad25f5625bbac328130b53058", + "git_blob": "928151def17915c641a1e503a9fb50d47cb94c1b" + }, + { + "path": "RESULTS.json", + "bytes": 1299, + "sha256": "1c95d69227d2dee9a3213e7682563e9ad9d0c09ecd2523372eb422845d2b5420", + "git_blob": "8efdfcd18cc144f90b679c7eb8614c27988d9799" + }, + { + "path": "witness_check.py", + "bytes": 4349, + "sha256": "6841618c69183bdf15fba4dfe286891f1c33e15ebad9c3bdb646f60fd0f6d5f1", + "git_blob": "854ff4e05467e4fb896c7e4ffc4c0d213c4aae17" + } + ] +} diff --git a/frontiers/unrestricted_selection_difference_20260929/witness_check.py b/frontiers/unrestricted_selection_difference_20260929/witness_check.py new file mode 100644 index 000000000..854ff4e05 --- /dev/null +++ b/frontiers/unrestricted_selection_difference_20260929/witness_check.py @@ -0,0 +1,88 @@ +"""Exact companion for CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v1 (PROOF.md). Standard library, exact rationals. + +WITNESS g(u, v) = alpha cos u + cos^2 u + gamma cos v on (R/2piZ)^2, alpha = l/2 - gamma, gamma = 1: + (0,0) is a nondegenerate maximum and (pi,pi) a nondegenerate index-1 saddle, g(0,0) - g(pi,pi) = l exactly; + on the boundary of the square |u|,|v| <= a with cos a = c_m = -alpha/2 (the interior minimum of + P(c) = alpha c + c^2), g < g(pi,pi), and (pi,pi) lies outside the square (a < pi/2). + Everything reduces to polynomial facts in c = cos u, cos v at rational points. +LEDGER (U3) c t versus the compact t^(5/3): ratio unbounded as t -> 0; (U4) dyadic shells of l^q with q = -1 carry + equal mass, q = -2 growing mass, q = -1/2 summable. +Mutants (each must fail): sign-alpha, no-separation, compact-rate. +Illustrates Lemma 2; the Gaussian support, Kac-Rice and uniformity arguments are in PROOF.md. +""" +import argparse +import json +import sys +from fractions import Fraction as F + +MUTANTS = ("sign-alpha", "no-separation", "compact-rate") +MUT = None + + +def witness(l): + gamma = F(1) + alpha = (gamma - l / 2) if MUT == "sign-alpha" else (l / 2 - gamma) + P = lambda c: alpha * c + c * c # p(u) = P(cos u) + dP = lambda c: alpha + 2 * c + q = lambda c: gamma * c # q(v) = gamma cos v + out = {} + # p'(u) = -sin u P'(cos u), q'(v) = -gamma sin v: both vanish at u, v in {0, pi}. + # p''(u) = sin^2 u P''(cos u) - cos u P'(cos u); at u = 0: -P'(1); at u = pi: P'(-1). q''(0) = -gamma, q''(pi) = gamma. + hess_M = (-dP(F(1)), -gamma) + hess_S = (dP(F(-1)), gamma) + out["M_nondegenerate_maximum"] = hess_M[0] < 0 and hess_M[1] < 0 + out["S_nondegenerate_index1_saddle"] = hess_S[0] < 0 < hess_S[1] + gM, gS = P(F(1)) + q(F(1)), P(F(-1)) + q(F(-1)) + out["height_difference_is_l"] = gM - gS == l + c_m = F(0) if MUT == "no-separation" else -alpha / 2 + out["square_half_width_below_pi_over_2"] = 0 < c_m < 1 # cos a in (0,1): 0 < a < pi/2 < pi + # sides |u| = a: g <= P(c_m) + gamma; sides |v| = a: for |u| <= a, cos u in [c_m, 1] where P' >= 0, so + # max p = P(1), and g <= P(1) + gamma c_m. + out["P_increasing_on_cm_1"] = dP(c_m) >= 0 and dP(F(1)) >= 0 # P' linear + sup_boundary = max(P(c_m) + gamma, P(F(1)) + gamma * c_m) + out["boundary_strictly_below_saddle_height"] = sup_boundary < gS + return out + + +def ledger(): + out = {} + # (U3): (c t) / t^(5/3) = c t^(-2/3) -> infinity; check on t = 8^-n where t^(2/3) = 4^-n is rational + c = F(1, 10) + ratios = [c * F(8) ** -n / (F(8) ** -n) * F(4) ** n for n in range(1, 8)] # c t / t^(5/3) + growing = all(ratios[i + 1] > ratios[i] for i in range(len(ratios) - 1)) and ratios[-1] > 1000 + out["U3_linear_lower_beats_compact_t_five_thirds"] = growing if MUT != "compact-rate" else not growing + # (U4): shells [2^-(n+1), 2^-n] of density >= c times l^q: lower bound c * 2^-(n+1) * (2^-n)^q + def shell(q, n): + return c * F(2) ** (-(n + 1)) * F(2) ** (-n * q) + s1 = [shell(-1, n) for n in range(30)] + s2 = [shell(-2, n) for n in range(30)] + out["U4_q_minus1_equal_shells_diverge"] = len(set(s1)) == 1 and sum(s1) == 30 * s1[0] > 0 + out["U4_q_minus2_growing_shells"] = all(s2[i + 1] == 2 * s2[i] for i in range(29)) + # control: q = -1/2 upper shells c*2^-n*(2^-(n+1))^(-1/2) form a geometric series of ratio 2^(-1/2) < 1 + out["q_minus_half_summable_ratio"] = F(1, 2) < 1 # (2^(-1/2))^2 = 1/2 < 1 + return out + + +def flatten(d): + for v in d.values(): + if isinstance(v, dict): + yield from flatten(v) + else: + yield v + + +def main(): + global MUT + ap = argparse.ArgumentParser() + ap.add_argument("--mutant", choices=MUTANTS) + MUT = ap.parse_args().mutant + checks = {"WITNESS": {str(l): witness(l) for l in (F(1, 10), F(1, 100), F(1, 1000))}, "LEDGER": ledger()} + passed = all(flatten(checks)) + print(json.dumps({"object": "CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v1", "checks": checks, "passed": passed, + "scope": "exact finite witness and ledger only; not the Gaussian proof"}, + indent=2, sort_keys=True)) + return 0 if passed else 1 + + +if __name__ == "__main__": + sys.exit(main()) From 70b0d0f03bf03f7ad6ba1ebcab8376fa6ac887cb Mon Sep 17 00:00:00 2001 From: Claude Date: Tue, 29 Sep 2026 12:30:56 +0000 Subject: [PATCH 2/5] review record: nonauthor review of #132 mixed inverse distance/height moments (scientific effect NONE) Anthropic/Claude nonauthor analytic review of OA-REMOTE-MIXED-INVERSE-20260929-v1 (PROOF.md blob 975d8211, head e75b5e1e). Verdict ACCEPT at the stated fixed-E, fixed-rho, compact-mark scope: Theorem M (finite iff q+3beta<2 at fixed small r; H ~ r^(5-q-3beta) D; weighted Beta(alpha,2) height law), fixed-r J_(r,beta) > 0 with cutoffs, beta>=1 comparison, and the residue (M22). No defect. N1: for beta>=1 even separated-position moments are infinite. Adds an exact-rational stdlib checker (6 groups, 6 mutants, identical in -O), RESULTS.json, SOURCE_FILES.json and a path-filtered workflow. Independent of the C7 candidate on this branch. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_017Mi3hxjaxV45x6zo6o1ee3 --- .../workflows/remote-mixed-inverse-review.yml | 60 +++++++ .../RESULTS.json | 38 ++++ .../REVIEW.md | 103 +++++++++++ .../SOURCE_FILES.json | 37 ++++ .../mixed_review_check.py | 164 ++++++++++++++++++ 5 files changed, 402 insertions(+) create mode 100644 .github/workflows/remote-mixed-inverse-review.yml create mode 100644 reviews/remote_mixed_inverse_claude_20260929/RESULTS.json create mode 100644 reviews/remote_mixed_inverse_claude_20260929/REVIEW.md create mode 100644 reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json create mode 100644 reviews/remote_mixed_inverse_claude_20260929/mixed_review_check.py diff --git a/.github/workflows/remote-mixed-inverse-review.yml b/.github/workflows/remote-mixed-inverse-review.yml new file mode 100644 index 000000000..b5f1b7fe8 --- /dev/null +++ b/.github/workflows/remote-mixed-inverse-review.yml @@ -0,0 +1,60 @@ +name: Remote mixed-inverse nonauthor review checks + +'on': + pull_request: + paths: + - 'reviews/remote_mixed_inverse_claude_20260929/**' + - '.github/workflows/remote-mixed-inverse-review.yml' + workflow_dispatch: + +permissions: + contents: read + +jobs: + exact-and-probe: + runs-on: ubuntu-latest + timeout-minutes: 10 + steps: + - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 + with: + persist-credentials: false + - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 + with: + python-version: '3.11.16' + - name: Verify sources, replay both modes, reject mutants + env: + PYTHONDONTWRITEBYTECODE: '1' + run: | + set -euo pipefail + python -B -S - <<'PY' + import hashlib, json, pathlib, subprocess, sys + root = pathlib.Path('reviews/remote_mixed_inverse_claude_20260929').resolve() + manifest = json.loads((root / 'SOURCE_FILES.json').read_text())['files'] + present = sorted(p.name for p in root.iterdir()) + expected_names = sorted([e['path'] for e in manifest] + ['SOURCE_FILES.json']) + if present != expected_names: + raise ValueError('packet tree differs from manifest: ' + repr(present)) + for entry in manifest: + path = root / entry['path'] + if pathlib.PurePosixPath(entry['path']).name != entry['path'] or path.is_symlink(): + raise ValueError('flat regular source required: ' + entry['path']) + data = path.read_bytes() + if len(data) != entry['bytes'] or hashlib.sha256(data).hexdigest() != entry['sha256']: + raise ValueError('source identity mismatch: ' + entry['path']) + script = str(root / 'mixed_review_check.py') + expected = (root / 'RESULTS.json').read_bytes() + mutants = ['beta-in-T', 'k-exponent', 'far-envelope', 'triangle-half', 'radial-ledger', + 'residue-denominator'] + for flags in (['-B', '-S'], ['-B', '-O', '-S']): + out = subprocess.run([sys.executable, *flags, script], + check=True, capture_output=True, timeout=300).stdout + if out != expected: + raise RuntimeError('output differs from RESULTS.json under ' + ' '.join(flags)) + for m in mutants: + rc = subprocess.run([sys.executable, *flags, script, '--mutant', m], + capture_output=True, timeout=300).returncode + if rc != 1: + raise RuntimeError('mutant ' + m + ' not rejected') + print(json.dumps({'scientific_effect': 'NONE', 'mathematical_acceptance': False})) + PY + git diff --exit-code diff --git a/reviews/remote_mixed_inverse_claude_20260929/RESULTS.json b/reviews/remote_mixed_inverse_claude_20260929/RESULTS.json new file mode 100644 index 000000000..a2d3ec47f --- /dev/null +++ b/reviews/remote_mixed_inverse_claude_20260929/RESULTS.json @@ -0,0 +1,38 @@ +{ + "checks": { + "COEFF": { + "M14_T_exponent_beta_free": true, + "M15_constant": true, + "M15_k_exponent": true, + "M4_ratio": true + }, + "ENVELOPE": { + "M11_far_exponent_is_minus2_minus_q": true, + "domain_exactly_lambda_lt_2": true + }, + "FIXEDR": { + "M12_separated_order": true, + "M18_radial_power_1_minus_3beta_all_d": true, + "M20_monotone_comparison": true, + "integrable_iff_lambda_lt_2": true + }, + "HEIGHT": { + "M16_normalizer": true, + "M17_correlation": true, + "M5_joint_and_marginal_mass_one": true, + "M6_variance": true, + "RP_alpha_two_thirds_and_q1_alpha_one_third": true, + "boundary_mean_gap_to_zero": true, + "pushforward_exponent": true + }, + "OVERLAP": { + "M13_exact": true + }, + "RESIDUE": { + "M22_matches_M21": true + } + }, + "object": "CLAUDE-REVIEW-REMOTE-MIXED-INVERSE-20260929-v1", + "passed": true, + "scope": "exact finite identities only; not an analytic proof" +} diff --git a/reviews/remote_mixed_inverse_claude_20260929/REVIEW.md b/reviews/remote_mixed_inverse_claude_20260929/REVIEW.md new file mode 100644 index 000000000..d7cf206d9 --- /dev/null +++ b/reviews/remote_mixed_inverse_claude_20260929/REVIEW.md @@ -0,0 +1,103 @@ +# Nonauthor analytic review: mixed inverse distance and height-gap moments (Math-#132) + +Scientific effect: **NONE**. This record changes no register, status, graph node, lemma flag, prize or source. +Integration is a separate act. + +## Object and exposure + +| Field | Value | +|---|---| +| Object | OA-REMOTE-MIXED-INVERSE-20260929-v1 (OpenAI / ChatGPT) | +| Head | `e75b5e1e715dacc0ec7fba9cf807ba0501593c85` ([Math-#132](https://github.com/d6g8k5htny-coder/Math-/pull/132)) | +| `PROOF.md` | Git blob `975d8211a30be05cb337c81036a0c8de8a9f4264`, 19335 B, SHA256 `f0cfa581e93ffbca8e167ccfaeddf222c27d0625dc81c492a1f3da1e8de123eb` | +| Request / pickup | [5889283745](https://github.com/d6g8k5htny-coder/Math-/pull/132#issuecomment-5889283745) / [5890297228](https://github.com/d6g8k5htny-coder/Math-/pull/132#issuecomment-5890297228). Lane split: xAI/Grok [5890256629](https://github.com/d6g8k5htny-coder/Math-/pull/132#issuecomment-5890256629). | +| Reviewer | Anthropic Claude, session `session_017Mi3hxjaxV45x6zo6o1ee3`. Different provider from the author; same GitHub account, so no organizational independence is claimed. | +| Exposure | **I authored RC** and reviewed RP (C3) and RI (#128, ACCEPT). This review checks how #132 uses them; it does not revalidate them independently. | +| Author code | `verify.py` replayed on an exact archive of the head: 18 tests and 8 intended mutant failures per mode, 9 identical output pairs. It is a finite companion only. | +| Own checks | `mixed_review_check.py`: exact rationals, 6 groups, 6 mutants, identical under `-O`. | + +## Verdicts + +| Interface | Verdict | +|---|---| +| (M1)–(M2), with truncation at zero gaps | **ACCEPT** | +| **Theorem M**: finite iff `q + 3β < 2` at fixed small `r`; (M3) `H ~ r^{5−q−3β}D`; (M4)–(M6) | **ACCEPT** at the stated fixed-`E`, fixed-`ρ`, compact-mark, existential scope | +| §3 (M9)–(M12): weight, envelope, both integrated ends, fixed-Borel truncation, separated two-height integral | **ACCEPT** | +| §4 (M13)–(M15): coefficient; `β` cancels from the `T` exponent but not from `k^{−β}` | **ACCEPT** | +| §5 (M16)–(M17): weighted `Beta(α, 2)` law, triangle halves, TV on the weighted measure, boundary weak diagonal law | **ACCEPT** | +| §6 (M7), (M8), (M18)–(M20): fixed-`r` coefficient `J_{r,β} > 0`, necessity, cutoffs, `β ≥ 1` comparison | **ACCEPT** | +| §7 (M21)–(M22): contact matching and residue | **ACCEPT** | + +No defect was found. + +## Hand re-derivations + +**Weight and envelope.** The weight is `δ^{−q}Δ^{−β} = r^{−λ}s^{−λ}|t|^{−β}`, with `λ = q + 3β` and no factor of `k`. It +multiplies RP's envelope `Cs(1+|t|)^{−N}1_{(M9)}`. +- For `s ≤ 1`, the `t`-integral is finite because `β < 1`, which (M2) forces since `β < 2/3`. +- For `s ≥ 1`, (M9) confines `|t| < k/s³`, and `∫_{|t| −1` (at infinity). `ENVELOPE` decides this +exactly on rational grids with dyadic shells. + +**Separated pairs.** `∫∫_{I_r²}|y−y'|^{−β} = 2ℓ^{2−β}/((1−β)(2−β))` with `ℓ = kr³`. The contribution is +`O(r^{6−3β})`, which is `O(r^{1+q})` after division by `r^{5−λ}`. Using the actual two-height integral here, rather +than an unweighted pair bound, is necessary. + +**Coefficient.** +- (M13) is checked exactly (`OVERLAP`). +- Multiplying by `36|t|^{2−β}` gives the `T` exponent `2 − β − (2−λ)/3 = (4+q)/3`, independent of `β`. +- Converting with `t = −T/12` uses `108·12^{−(4+q)/3} = (3/4)12^{(2−q)/3}`. +- The `k`-power is `(5−λ)/3 = (5−q)/3 − β`. +- Against RI's `A_{−q}`, which RI establishes for `p = −q ∈ (−2, 0]`, this gives (M4)'s ratio + `k^{−β}(2−q)(5−q)/((2−λ)(5−λ))`. + +All of these are checked (`COEFF`). Mutants that let `β` into the `T` exponent or drop it from the `k` exponent are +rejected. + +**Height law.** Under `g = s³|t|/k`, `s^{1−λ}ds = (1/3)(k/|t|)^α g^{α−1}dg` with `α = (2−λ)/3`. The window leaves +the smaller deficit free on `(0, 1−g)`. Swapping the ordered pair preserves `δ` and `Δ`, so both triangles carry equal +mass; this is simpler than RP's `e ↦ −e` and equally valid. `HEIGHT` checks, for every grid point: +- the normalizer (M16); +- the moments (M6); +- the correlation (M17), derived independently from the conditional-uniform smaller mark; +- unit mass of the joint density and of the marginal; +- the special values `1/4, 9/176, 2/7` at `α = 2/3` (RP) and `1/7, 7/11` at `α = 1/3`. + +TV convergence comes from weighted `L¹` convergence of the blown-up densities, not from RP's unweighted TV. At the +boundary `α ↓ 0`, weak convergence to the diagonal holds with TV distance 1, and it is not a joint limit. + +**Fixed `r`.** The radial power (M18) is `(d−1) − (d+3) + 3 + 2 − 3β = 1 − 3β` for every d. With `T = −12t`, the +height disintegration gives (M19), which reduces to #128's (I15) at `β = 0`. It is positive by the distinct-site +support argument (`|T|^{2−β} > 0` when `T ≠ 0`). Necessity comes from `J_{r,β} > 0`, not from a failed envelope. The +comparison (M20) holds because `Δ < kr³` and `δ ≤ D_T`. + +**Residue.** At `q* = 2 − 3β`, the quantity `(2−λ)D` has prefactor `3/(4·3) = 1/4`, 12-power `β`, `k`-power 1 and +`T`-power `2 − β`. That is exactly (M21). It is the `β`-analogue of the Abelian cross-check in my #128 review +(`RESIDUE`). + +## Notes + +- **N1 (`β ≥ 1` is infinite even at separated positions).** At `δ ≥ η_0`, the joint height density given zero + gradients is positive and continuous on the diagonal `y = y'`. So `∫∫_{I_r²}|y−y'|^{−β} = ∞` for `β ≥ 1`, and every + distance-cutoff moment is infinite there. This strengthens the note's remark that a distance cutoff "need not" + regularize `β ≥ 1`. It gives a second proof of the `β ≥ 1` part of (M20). +- **N2.** (M20) in fact covers every `β ≥ 2/3` and every `q ≥ 0`. For `2/3 ≤ β < 1` the positive diagonal (M7) + already gives this. +- **N3.** As in #128, the `o(1)` in (M7) has no rate, so `λ = 2` gives an equivalence only, with no bounded remainder. + The note says so. +- **N4.** The mixed domain is genuinely coupled: `q < 2` and `β < 2/3` separately do not suffice (`GRID` includes + `q = 3/2`, `β = 1/6`, where `λ = 2`). This matches the note's warning that separate finiteness of the two factors + does not imply mixed finiteness. + +## Not reviewed or accepted + +This review does not cover C4's crossover, a same-index threshold, an event-conditioned law, arbitrary real +two-exponent domains, coupled `ε(r)`, `q(r)` or `β(r)` limits, `ρ → 0`, numerical constants, or status. + +## Reproduce + + python -B -S mixed_review_check.py # prints RESULTS.json byte for byte + python -B -O -S mixed_review_check.py # identical + python -B -S mixed_review_check.py --mutant M # exit 1 for each of the 6 mutants diff --git a/reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json b/reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json new file mode 100644 index 000000000..c1d5e4307 --- /dev/null +++ b/reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json @@ -0,0 +1,37 @@ +{ + "schema": 1, + "scientific_effect": "NONE", + "review_status": "NONAUTHOR_REVIEW", + "reviewed_object": { + "pr": "d6g8k5htny-coder/Math-#132", + "head": "e75b5e1e715dacc0ec7fba9cf807ba0501593c85", + "paths": [ + { + "path": "frontiers/remote_mixed_inverse_20260929/PROOF.md", + "git_blob": "975d8211a30be05cb337c81036a0c8de8a9f4264", + "bytes": 19335, + "sha256": "f0cfa581e93ffbca8e167ccfaeddf222c27d0625dc81c492a1f3da1e8de123eb" + } + ] + }, + "files": [ + { + "path": "REVIEW.md", + "bytes": 6991, + "sha256": "44a4b89a1bb367e3e9d7c80002d0aac8ec9ceee256e0d6316b1ce5534bba8a21", + "git_blob": "d7cf206d9808400e84a81d94de368b17f22f5fef" + }, + { + "path": "RESULTS.json", + "bytes": 1009, + "sha256": "92d669341cd1408a11b366636c035708505e43d2da7866f7f7d40ebb3fb4dad0", + "git_blob": "a2d3ec47f8a7fa71b80f406e229a6de85aa5de11" + }, + { + "path": "mixed_review_check.py", + "bytes": 7872, + "sha256": "4ae5ae51d50662adef47989eab18130bc1d557a37b2deb251bff6938e2d2b461", + "git_blob": "4691c98eb3a08b5e655bde45444b5bc83cae2c81" + } + ] +} diff --git a/reviews/remote_mixed_inverse_claude_20260929/mixed_review_check.py b/reviews/remote_mixed_inverse_claude_20260929/mixed_review_check.py new file mode 100644 index 000000000..4691c98eb --- /dev/null +++ b/reviews/remote_mixed_inverse_claude_20260929/mixed_review_check.py @@ -0,0 +1,164 @@ +"""Independent nonauthor checks for OA-REMOTE-MIXED-INVERSE-20260929-v1 (Math-#132, PROOF.md). Stdlib, exact rationals. + +ENVELOPE (M10)-(M11): for s >= 1 the t-window |t| < k/s^3 gives s^(1-q-3b) * s^(-3(1-b)) = s^(-2-q); dyadic-shell + ratios decide the domain exactly: finite iff q + 3b < 2 (at 0) and q > -1 (at infinity). +OVERLAP (M13) on exact instances (s_max a perfect power), lambda = q + 3b rational in [0, 2). +COEFF (M14) T-exponent (4+q)/3 independent of b; (M15) 108 * 12^(-(4+q)/3) = (3/4) 12^((2-q)/3); k-exponent + (5-q)/3 - b; (M4) D/A_{-q} = k^-b (2-q)(5-q)/((2-lambda)(5-lambda)). +HEIGHT pushforward exponent alpha - 1, (M16), Beta(alpha, 2) mean/variance (M6), marginal mass, the correlation + (M17) from the conditional-uniform smaller mark, and the RP values at alpha = 2/3 (1/4, 9/176, 2/7) and + alpha = 1/3 (1/7, 7/11); the joint density alpha(alpha+1)|a-b|^(alpha-1)/2 has mass one. +FIXEDR (M18) radial power 1 - 3b for every d; integrability iff q + 3b < 2; (M12) separated order r^(6-3b) and + r^(1+q) after division; (M20) exponent bookkeeping for b >= b0 = 2/3. +RESIDUE (M22): at q* = 2 - 3b the prefactor is 1/4, the 12-power b, the k-power 1, the T-power 2 - b, matching (M21). +Mutants (each must fail): beta-in-T, k-exponent, far-envelope, triangle-half, radial-ledger, residue-denominator. +""" +import argparse +import json +import sys +from fractions import Fraction as F + +MUTANTS = ("beta-in-T", "k-exponent", "far-envelope", "triangle-half", "radial-ledger", "residue-denominator") +MUT = None +GRID = [(F(q), F(b)) for q in (0, F(1, 2), 1, F(3, 2)) for b in (0, F(1, 9), F(1, 6), F(1, 3), F(1, 2))] + + +def check_envelope(): + ok_far = ok_dom = True + for q, b in GRID: + far = (1 - q - 3 * b) + (-3 * (1 - b)) if MUT != "far-envelope" else (1 - q - 3 * b) - 3 + ok_far &= far == -2 - q + lam = q + 3 * b + n = lam.denominator * q.denominator + near_ratio = F(2) ** (-int(n * (2 - lam))) if n * (2 - lam) >= 0 else F(2) ** int(-n * (2 - lam)) + finite = near_ratio < 1 and (-2 - q) < -1 + ok_dom &= finite == (lam < 2) + return {"M11_far_exponent_is_minus2_minus_q": ok_far, "domain_exactly_lambda_lt_2": ok_dom} + + +def check_overlap(): + ok = True + for q, b in GRID: + lam = q + 3 * b + if lam >= 2: + continue + den = lam.denominator + for base in (2, 3): + for tt in (F(1), F(3, 7)): + m = F(base) ** den + k = m ** 3 * tt + m_2l = F(base) ** int(den * (2 - lam)) # s_max^(2-lambda) + lhs = k * m_2l / (2 - lam) - tt * m_2l * m ** 3 / (5 - lam) + ok &= lhs == 3 * k * m_2l / ((2 - lam) * (5 - lam)) + return {"M13_exact": ok} + + +def check_coeff(): + out = {"M14_T_exponent_beta_free": all( + (2 - b - (2 - q - 3 * b) / 3) == ((4 + q) / 3 - (b if MUT == "beta-in-T" else 0)) for q, b in GRID)} + out["M14_T_exponent_beta_free"] &= all((2 - b - (2 - q - 3 * b) / 3) == (4 + q) / 3 for q, b in GRID) + out["M15_constant"] = F(108) == F(3, 4) * 144 and all((2 - q) / 3 + (4 + q) / 3 == 2 for q, _ in GRID) + kexp = lambda q, b: (5 - q - 3 * b) / 3 + out["M15_k_exponent"] = all(kexp(q, b) == (5 - q) / 3 - (0 if MUT == "k-exponent" else b) for q, b in GRID) + # (M4): D's rational prefactor 3/(4(2-lam)(5-lam)) over RI's A_(-q) prefactor 3/(4(2-q)(5-q)), and k-exponents + pref_D = lambda q, b: F(3) / (4 * (2 - q - 3 * b) * (5 - q - 3 * b)) + pref_A = lambda q: F(3) / (4 * (2 - q) * (5 - q)) + out["M4_ratio"] = all(pref_D(q, b) / pref_A(q) == (2 - q) * (5 - q) / ((2 - q - 3 * b) * (5 - q - 3 * b)) + and kexp(q, b) - (5 - q) / 3 == -b for q, b in GRID if q + 3 * b < 2) + return out + + +def beta_moments(a): + m1 = a / (a + 2) + m2 = a * (a + 1) / ((a + 2) * (a + 3)) + return m1, m2 + + +def check_height(): + out = {} + ok_push = ok_norm = ok_mom = ok_corr = ok_joint = True + for q, b in GRID: + lam = q + 3 * b + if lam >= 2: + continue + a = (2 - lam) / 3 + ok_push &= (1 - lam) / 3 - F(2, 3) == a - 1 # s^(1-lam) ds -> g^(alpha-1) dg + ok_norm &= F(1) / a - F(1) / (a + 1) == F(1) / (a * (a + 1)) # (M16) + m1, m2 = beta_moments(a) + ok_mom &= m2 - m1 ** 2 == 2 * a / ((a + 2) ** 2 * (a + 3)) + E2 = F(1, 3) - m1 / 6 + m2 / 3 + Ex = F(1, 3) - m1 / 6 - m2 / 6 + corr = (Ex - F(1, 4)) / (E2 - F(1, 4)) + ok_corr &= corr == (2 - a - a * a) / (2 + a + a * a) + # joint density alpha(alpha+1)|a-b|^(alpha-1)/2: mass = alpha(alpha+1)/2 * 2 * int_0^1 int_0^y u^(alpha-1) du dy + half = F(1) if MUT == "triangle-half" else F(1, 2) + ok_joint &= a * (a + 1) * half * 2 * (F(1) / (a * (a + 1))) == 1 + # marginal (alpha+1)(a^alpha+(1-a)^alpha)/2 has mass (alpha+1)/2 * 2/(alpha+1) = 1 + ok_joint &= (a + 1) / 2 * 2 / (a + 1) == 1 + m1, m2 = beta_moments(F(2, 3)) + rp = (m1 == F(1, 4) and m2 - m1 ** 2 == F(9, 176) + and (F(1, 3) - m1 / 6 - m2 / 6 - F(1, 4)) / (F(1, 3) - m1 / 6 + m2 / 3 - F(1, 4)) == F(2, 7)) + m1b, m2b = beta_moments(F(1, 3)) + q1 = m1b == F(1, 7) and (F(1, 3) - m1b / 6 - m2b / 6 - F(1, 4)) / (F(1, 3) - m1b / 6 + m2b / 3 - F(1, 4)) == F(7, 11) + out.update({"pushforward_exponent": ok_push, "M16_normalizer": ok_norm, "M6_variance": ok_mom, + "M17_correlation": ok_corr, "M5_joint_and_marginal_mass_one": ok_joint, + "RP_alpha_two_thirds_and_q1_alpha_one_third": rp and q1}) + # boundary: E G = alpha/(alpha+2) decreases to 0 as alpha -> 0 + seq = [beta_moments(F(1, 10 ** n))[0] for n in range(1, 6)] + out["boundary_mean_gap_to_zero"] = all(seq[i + 1] < seq[i] for i in range(4)) and seq[-1] < F(1, 10 ** 5) + return out + + +def check_fixedr(): + out = {} + ok = True + for d in range(2, 11): + for _, b in GRID: + extra = 0 if MUT == "radial-ledger" else -3 * b + ok &= (d - 1) - (d + 3) + 3 + 2 + extra == 1 - 3 * b + out["M18_radial_power_1_minus_3beta_all_d"] = ok + out["integrable_iff_lambda_lt_2"] = all(((1 - 3 * b - q) > -1) == (q + 3 * b < 2) for q, b in GRID) + out["M12_separated_order"] = all((6 - 3 * b) - (5 - q - 3 * b) == 1 + q for q, b in GRID) + # (M20): for b >= b0 = 2/3 and 0 < Delta < l, Delta^-(b-b0) >= l^-(b-b0) (monotone, exponent >= 0); + # checked on integer exponents b - b0 in {0, 1, 2} + ok = True + for e in (0, 1, 2): + for Dl, l in ((F(1, 4), F(1, 2)), (F(1, 100), F(1, 3))): + ok &= Dl < l and Dl ** -e >= l ** -e + out["M20_monotone_comparison"] = ok + return out + + +def check_residue(): + ok = True + den = 5 if MUT == "residue-denominator" else 3 + for b in (F(0), F(1, 9), F(1, 6), F(1, 3), F(1, 2)): + qs = 2 - 3 * b + ok &= (5 - qs - 3 * b) == den and F(3) / (4 * den) == F(1, 4) + ok &= (2 - qs) / 3 == b and (5 - qs) / 3 - b == 1 and (4 + qs) / 3 == 2 - b + return {"M22_matches_M21": ok} + + +def flatten(d): + for v in d.values(): + if isinstance(v, dict): + yield from flatten(v) + else: + yield v + + +def main(): + global MUT + ap = argparse.ArgumentParser() + ap.add_argument("--mutant", choices=MUTANTS) + MUT = ap.parse_args().mutant + checks = {"ENVELOPE": check_envelope(), "OVERLAP": check_overlap(), "COEFF": check_coeff(), + "HEIGHT": check_height(), "FIXEDR": check_fixedr(), "RESIDUE": check_residue()} + passed = all(flatten(checks)) + print(json.dumps({"object": "CLAUDE-REVIEW-REMOTE-MIXED-INVERSE-20260929-v1", "checks": checks, "passed": passed, + "scope": "exact finite identities only; not an analytic proof"}, indent=2, sort_keys=True)) + return 0 if passed else 1 + + +if __name__ == "__main__": + sys.exit(main()) From e8fb985adf7b11e42ad5f5f307c0e85bfab2280e Mon Sep 17 00:00:00 2001 From: Claude Date: Tue, 29 Sep 2026 12:32:54 +0000 Subject: [PATCH 3/5] C7 candidate v2: repair Lemma 1 at the closed critical superlevel (scientific effect NONE) As found in 5890369316: v1 claimed y lies in the closure of x's component of {f>t} for t>f(y), which is impossible since that closure lies in {f>=t}. v2 works at {f>=s}, s=f(y): on the generic Morse/distinct-value locus the saddle killing x's bar lies in the closure of x's component of {f>s}, so x and y share a component of {f>=s}; the shell barrier below s traps it in B(x,rho). ell=0 is only a compactness endpoint. Theorem U, Lemmas 2-3, witness and corollaries unchanged; manifest refreshed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_017Mi3hxjaxV45x6zo6o1ee3 --- .../PROOF.md | 36 ++++++++++++++----- .../SOURCE_FILES.json | 6 ++-- 2 files changed, 31 insertions(+), 11 deletions(-) diff --git a/frontiers/unrestricted_selection_difference_20260929/PROOF.md b/frontiers/unrestricted_selection_difference_20260929/PROOF.md index 928151def..7c7f9cbf6 100644 --- a/frontiers/unrestricted_selection_difference_20260929/PROOF.md +++ b/frontiers/unrestricted_selection_difference_20260929/PROOF.md @@ -1,6 +1,6 @@ # The unrestricted selection difference does not vanish: disposition of candidate C7 -Object: CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v1. +Object: CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v2. Author: Anthropic Claude (Claude Code session `session_017Mi3hxjaxV45x6zo6o1ee3`). Disposition: AUTHOR-SIDE CANDIDATE; nonauthor analytic review required. Scientific effect: NONE. No `STATUS`, `PROOF_INDEX`, `GRAPH`, claim, lemma flag, prize or source body changes. @@ -63,14 +63,25 @@ pairs carrying the mark, ## 3. A separating level curve forbids pairing -**Lemma 1.** Let `f` be `C²`, let `x` be a local maximum and `y` an index-`(d−1)` saddle with `dist(x, y) ≥ r_0`, and -suppose `f ∈ O_{x,y,ℓ}`. Then `(x, y)` is not an `H_0` persistence pair of the superlevel filtration. +**Lemma 1.** Let `f` be a Morse function with distinct critical values (the almost-sure generic locus of parent §8). +Let `x` be a local maximum and `y` an index-`(d−1)` saddle with `dist(x, y) ≥ r_0`, and suppose `f ∈ O_{x,y,ℓ}`, that +is, `sup_{∂B(x,ρ)} f < s := f(y)`. Then `(x, y)` is not a finite `H_0` persistence pair of the superlevel +filtration. -*Proof.* If `(x, y)` were a pair, the bar born at `x` would die at `y`. So for every `t` slightly above `f(y)`, the -component `C_t` of `{f > t}` containing `x` would have `y` in its closure. Every such `t > f(y)` also -satisfies `t > sup_{∂B(x,ρ)} f`. Then `C_t` is connected, contains `x` and does not meet `∂B(x, ρ)`. Since -`ρ < r_0 < L/4`, the ball is embedded and `∂B(x,ρ)` separates it from its complement, so `C_t ⊂ B(x, ρ)`. Also -`f < t` on `∂B(x, ρ)`, so `cl(C_t) ⊂ B(x, ρ)`, which does not contain `y`. ∎ +*Proof.* We work at the closed critical superlevel `{f ≥ s}`, not above `s`. + +We use one incidence fact on the generic locus. If a finite `H_0` bar is born at `x` and killed at `y`, then `y` lies +in the closure of the component `C` of `{f > s}` that contains `x`. The reason is the local picture at a Morse saddle +of index `d−1` at level `s`. Near `y`, `{f > s}` consists of two local cones. The merge at `y` joins the two global +components of `{f > s}` that contain them, and the bar killed at `y` is the younger of those two components, which is +`C`. Since `cl(C) ⊂ {f ≥ s}` is connected, `x` and `y` lie in one connected component `K` of `{f ≥ s}`. + +Now `f < s` on `∂B(x, ρ)`, so `K` does not meet `∂B(x, ρ)`. Since `ρ < r_0 < L/4`, the ball is embedded and its +boundary separates it from its complement. As `K` is connected and contains `x`, `K ⊂ B(x, ρ)`. But +`dist(x, y) ≥ r_0 > ρ`, so `y ∉ B(x, ρ)`, a contradiction. ∎ + +The height difference `ℓ = f(x) − f(y)` is positive for every actual pair. The value `ℓ = 0` enters below only as an +endpoint of the compact parameter set in Lemmas 2–3, never as a paired event. Hence the far candidate pairs carrying the mark in (2.2) are not bars, and @@ -158,3 +169,12 @@ the square `|u|, |v| ≤ a` with `cos a = −α/2` (rational), `g < g(π, π)`. the same interface as parent §9 and [RC] (3.1). Nothing from the elder-selection upper estimate of Theorem A is used. + +## Revisions + +- **v2.** Lemma 1 is repaired as requested in + [5890369316](https://github.com/d6g8k5htny-coder/Math-/pull/133#issuecomment-5890369316). The v1 proof asserted that + `y` lies in the closure of `x`'s component of `{f > t}` for `t > f(y)`. That is false, because such a closure lies + in `{f ≥ t}` and `f(y) < t`. v2 argues at the closed critical superlevel `{f ≥ f(y)}`, using the generic-locus merge + incidence stated explicitly, and treats `ℓ = 0` only as a compactness endpoint. Theorem U, Lemmas 2–3, the witness + and the corollaries are unchanged. diff --git a/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json b/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json index 3594ac5e8..97559c96e 100644 --- a/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json +++ b/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json @@ -14,9 +14,9 @@ "files": [ { "path": "PROOF.md", - "bytes": 10473, - "sha256": "827d718a664ebae4edd05f8c63dcf10bcb3ea5cad25f5625bbac328130b53058", - "git_blob": "928151def17915c641a1e503a9fb50d47cb94c1b" + "bytes": 11811, + "sha256": "0e7b0728048856b67e6e0fca08b3c4cc1f05c7e130acd47ba2eff103643720f2", + "git_blob": "7c7f9cbf6c161a353b725dabe4ef0df901549db2" }, { "path": "RESULTS.json", From f2dc8b5be4aa7e2ee96275881e8e72e70ac7c9c7 Mon Sep 17 00:00:00 2001 From: Claude Date: Tue, 29 Sep 2026 12:36:21 +0000 Subject: [PATCH 4/5] C7 candidate v3: (U3) as c_* min(t, ell_0), explicit density-version convention, V_0 closure note (scientific effect NONE) Qualifications requested in 5890428429. Theorem U, Lemmas 1-3, witness and corollary U4 unchanged; manifest refreshed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_017Mi3hxjaxV45x6zo6o1ee3 --- .../PROOF.md | 20 +++++++++++++++---- .../SOURCE_FILES.json | 6 +++--- 2 files changed, 19 insertions(+), 7 deletions(-) diff --git a/frontiers/unrestricted_selection_difference_20260929/PROOF.md b/frontiers/unrestricted_selection_difference_20260929/PROOF.md index 7c7f9cbf6..5a55b179a 100644 --- a/frontiers/unrestricted_selection_difference_20260929/PROOF.md +++ b/frontiers/unrestricted_selection_difference_20260929/PROOF.md @@ -1,6 +1,6 @@ # The unrestricted selection difference does not vanish: disposition of candidate C7 -Object: CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v2. +Object: CL-D1-UNRESTRICTED-DIFFERENCE-20260929-v3. Author: Anthropic Claude (Claude Code session `session_017Mi3hxjaxV45x6zo6o1ee3`). Disposition: AUTHOR-SIDE CANDIDATE; nonauthor analytic review required. Scientific effect: NONE. No `STATUS`, `PROOF_INDEX`, `GRAPH`, claim, lemma flag, prize or source body changes. @@ -30,7 +30,7 @@ where "far" means torus distance `≥ r_0` between the two points, as in parent - **(U2)** (C7) is false. `D(ℓ) := ν_cand^all(ℓ) − ν_eld^all(ℓ)` satisfies `D(ℓ)/ℓ^α → ∞` for every `α > 0`. Combined with parent (14.1) and Theorem C, `c_* ≤ D(ℓ) = o(ℓ^{−1/3})`. - **(U3)** The expected per-unit-volume number of candidate pairs with lifetime in `(0, t]` that are not bars is at - least `c_* t`. In the compact window, parent (12.1) gives at most `C t^{5/3}`. + least `c_* min(t, ℓ_0)` for every `t > 0`, and so at least `c_* t` for `0 < t ≤ ℓ_0 = 1`. In the compact window, parent (12.1) gives at most `C t^{5/3}`. - **(U4)** For `q ≤ −1`, the nonselected sum `E Σ_{candidate, not bar, ℓ ≤ t} ℓ^q` is infinite for every `t > 0`. In the compact window, parent (12.3) makes it finite for every `q > −5/3`. @@ -100,7 +100,8 @@ Cameron–Martin space of the unconditioned field restricted to `h(x) = h(y) = 0 Every Fourier weight is positive (parent §2), so `H_0` contains every trigonometric polynomial satisfying those `2(d+1)` linear conditions. Their `C²`-closure is `V_0 = {φ ∈ C² : φ, ∇φ vanish at x and y}`. To see this, take -`φ ∈ V_0` smooth and trigonometric polynomials `τ_n → φ` in `C²`. The `2(d+1)` point functionals are linearly +`φ ∈ V_0` smooth (a general `φ ∈ V_0` is first smoothed in `C²` and then corrected by the same finite constraints; +only the smooth `G − m` below is needed) and trigonometric polynomials `τ_n → φ` in `C²`. The `2(d+1)` point functionals are linearly independent on trigonometric polynomials (parent §2, distinct-site jets), so there are trigonometric polynomials `ψ_i` dual to them. Then `τ_n − Σ_i λ_i(τ_n) ψ_i` lies in `H_0` and converges to `φ`, because `λ_i(τ_n) → λ_i(φ) = 0`. @@ -132,7 +133,14 @@ a positive constant. ∎ With (3.1) this proves (U1), with `ℓ_0 = 1`, and (U2). -For (U3), integrate (U1) over `(0, t]`. For (U4), the nonselected measure has density at least `c_*` on `(0, ℓ_0]`, +*Density convention.* As in the parent, (U1) is a statement about compatible density versions. The marked +Kac–Rice identity (parent §9, a finite-measure identity for Borel marks) dominates the nonselected expected measure +on `ℓ ∈ (0, 1]` from below by the measure with density `N_O(ℓ)`. So the nonselected density is at least `N_O(ℓ)` for +almost every `ℓ`, and the versions in (U1) are chosen compatibly. The generic-locus assertion of Lemma 1 is used +only almost surely inside that identity, never pointwise at an exceptional conditioning value. The support +argument at `ℓ = 0` needs no genericity and is only a compactness endpoint. + +For (U3), integrate (U1) over `(0, min(t, ℓ_0)]`. For (U4), the nonselected measure has density at least `c_*` on `(0, ℓ_0]`, and `∫_0 ℓ^q dℓ = ∞` for `q ≤ −1`. The bound `D = o(ℓ^{−1/3})` is Theorem C, since both densities are `c_{d,L} ℓ^{−1/3}(1 + o(1))`. ∎ @@ -172,6 +180,10 @@ Nothing from the elder-selection upper estimate of Theorem A is used. ## Revisions +- **v3.** Qualifications requested in + [5890428429](https://github.com/d6g8k5htny-coder/Math-/pull/133#issuecomment-5890428429): (U3) holds as + `c_* min(t, ℓ_0)`; an explicit density-version convention (expected-measure domination, a.e. densities); a note + on the full `V_0` closure. No change to Theorem U or the witness. - **v2.** Lemma 1 is repaired as requested in [5890369316](https://github.com/d6g8k5htny-coder/Math-/pull/133#issuecomment-5890369316). The v1 proof asserted that `y` lies in the closure of `x`'s component of `{f > t}` for `t > f(y)`. That is false, because such a closure lies diff --git a/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json b/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json index 97559c96e..c0ec506ff 100644 --- a/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json +++ b/frontiers/unrestricted_selection_difference_20260929/SOURCE_FILES.json @@ -14,9 +14,9 @@ "files": [ { "path": "PROOF.md", - "bytes": 11811, - "sha256": "0e7b0728048856b67e6e0fca08b3c4cc1f05c7e130acd47ba2eff103643720f2", - "git_blob": "7c7f9cbf6c161a353b725dabe4ef0df901549db2" + "bytes": 13028, + "sha256": "93a63002d32fd5c35ae994fe28b1205d8098026d6ddde19c67d4977dcf5817a9", + "git_blob": "5a55b179a974a90c65d257f3f93765cc6fb30bb8" }, { "path": "RESULTS.json", From 03a03817492f9012493aa04c3faa949fa6475d7b Mon Sep 17 00:00:00 2001 From: Claude Date: Tue, 29 Sep 2026 12:41:20 +0000 Subject: [PATCH 5/5] review record #132: qualify N1 (positive-measure cutoff pair set; full weighted kernel) (scientific effect NONE) Requested in 5890507179. No verdict, proof reading or checker change; manifest refreshed. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_017Mi3hxjaxV45x6zo6o1ee3 --- .../REVIEW.md | 19 +++++++++++++++---- .../SOURCE_FILES.json | 6 +++--- 2 files changed, 18 insertions(+), 7 deletions(-) diff --git a/reviews/remote_mixed_inverse_claude_20260929/REVIEW.md b/reviews/remote_mixed_inverse_claude_20260929/REVIEW.md index d7cf206d9..00c565c33 100644 --- a/reviews/remote_mixed_inverse_claude_20260929/REVIEW.md +++ b/reviews/remote_mixed_inverse_claude_20260929/REVIEW.md @@ -79,10 +79,15 @@ comparison (M20) holds because `Δ < kr³` and `δ ≤ D_T`. ## Notes -- **N1 (`β ≥ 1` is infinite even at separated positions).** At `δ ≥ η_0`, the joint height density given zero - gradients is positive and continuous on the diagonal `y = y'`. So `∫∫_{I_r²}|y−y'|^{−β} = ∞` for `β ≥ 1`, and every - distance-cutoff moment is infinite there. This strengthens the note's remark that a distance cutoff "need not" - regularize `β ≥ 1`. It gives a second proof of the `β ≥ 1` part of (M20). +- **N1 (`β ≥ 1` at separated positions).** Consider a distance cutoff whose retained pair set in `E × E` has positive + measure; this holds for every sufficiently small fixed cutoff when `E` has positive volume. At separated sites + (`δ ≥ η_0`), the full weighted kernel is positive and continuous near the diagonal `y = y'`. That kernel is the + joint height density given zero gradients, times `E[W_r F_i(H_x) F_j(H_{x'}) | …]` with the actual endpoint weight + and witness determinants. Positivity comes from the distinct-jet support with the correct endpoint Hessian types and + nonzero witness determinants; a positive Gaussian height density alone would not control a weight that could vanish + on equal heights. Hence `∫∫_{I_r²}|y−y'|^{−β}(…) = ∞` for `β ≥ 1`, and every such cutoff moment is infinite. (A + cutoff larger than `diam E` retains no pairs and gives zero.) This extends the note's deliberately weaker "need not + regularize" remark; it is not a defect in Theorem M, and it gives a second route to the `β ≥ 1` part of (M20). - **N2.** (M20) in fact covers every `β ≥ 2/3` and every `q ≥ 0`. For `2/3 ≤ β < 1` the positive diagonal (M7) already gives this. - **N3.** As in #128, the `o(1)` in (M7) has no rate, so `λ = 2` gives an equivalence only, with no bounded remainder. @@ -101,3 +106,9 @@ two-exponent domains, coupled `ε(r)`, `q(r)` or `β(r)` limits, `ρ → 0`, num python -B -S mixed_review_check.py # prints RESULTS.json byte for byte python -B -O -S mixed_review_check.py # identical python -B -S mixed_review_check.py --mutant M # exit 1 for each of the 6 mutants + +## Revisions + +- **v2** ([5890507179](https://github.com/d6g8k5htny-coder/Math-/pull/133#issuecomment-5890507179)): N1 is qualified + to cutoffs whose retained `E × E` pair set has positive measure, and it now rests on the full weighted + determinant–height kernel rather than the height density alone. No verdict or checker change. diff --git a/reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json b/reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json index c1d5e4307..10653b8ee 100644 --- a/reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json +++ b/reviews/remote_mixed_inverse_claude_20260929/SOURCE_FILES.json @@ -17,9 +17,9 @@ "files": [ { "path": "REVIEW.md", - "bytes": 6991, - "sha256": "44a4b89a1bb367e3e9d7c80002d0aac8ec9ceee256e0d6316b1ce5534bba8a21", - "git_blob": "d7cf206d9808400e84a81d94de368b17f22f5fef" + "bytes": 7956, + "sha256": "f82f50ddbf983d4965575d7326c686b144a7d9dda0b858385b7ef33d65f041f0", + "git_blob": "00c565c3379c53cd8940de019ae24221f54cbbc3" }, { "path": "RESULTS.json",