/- research/0025 — route 157: the exact capped-support certificate. The source's exact capped-numerator procedure (eprint 2026/1893, Lemma 4.20 and Appendix B) removes the cap by subtracting three correction regions `C_r, D_r, E_r` from the cap-free marginal. Two of them satisfy `R_r + V_r < u - d`; the paper then uses u - d = 2423 < ℓ = 2455 to conclude that the legal deletion regions already lie below the marginal base cutoff `ℓ`, so no extra base-cutoff condition is needed there. This file records that the premise is a *numerical accident of the source's k = 48 parameter set* and fails for the H₁ ≤ 216 candidate (k = 46): * candidate: u = A + ε_s = 0.2658, ℓ = A - ε_s = 0.2508, d = δ = 0.012, so `u - d = 0.2538 > ℓ`, i.e. `u - d - ℓ = 2 ε_s - δ = 3/1000 > 0`; * source: u = 0.2605, ℓ = 0.2455, d = 0.0182, so `u - d - ℓ = -32/10000 < 0`. Consequently, for the H₁ ≤ 216 candidate the domains `C_r` and `D_r` of Appendix B must be intersected with the base cutoff `R_r + V_r ≤ ℓ`; the paper's "automatically lie below the base cutoff" step does not transfer. Everything below is rational arithmetic; it is a statement about the source's procedure, not about the truth of any bounded-gap claim. -/ import Mathlib namespace TwinPrime0025 /-- H₁ ≤ 216 candidate radial base `A = 2583/10000`. -/ def A : ℚ := 2583 / 10000 /-- H₁ ≤ 216 candidate enlargement `ε_s = 3/400 = 0.0075`. -/ def eps_s : ℚ := 3 / 400 /-- H₁ ≤ 216 candidate rough-coordinate threshold `δ = 3/250 = 0.012`. -/ def delta : ℚ := 3 / 250 /-- H₁ ≤ 216 candidate outer radius `u = A + ε_s`. -/ def u : ℚ := A + eps_s /-- H₁ ≤ 216 candidate marginal base cutoff `ℓ = A - ε_s`. -/ def ell : ℚ := A - eps_s /-- The source's k = 48 outer radius (integer coordinates, 10⁴ t). -/ def uSrc : ℚ := 2605 / 10000 /-- The source's k = 48 marginal base cutoff. -/ def ellSrc : ℚ := 2455 / 10000 /-- The source's k = 48 rough threshold `d = 182`. -/ def dSrc : ℚ := 182 / 10000 theorem u_eq : u = 1329 / 5000 := by norm_num [u, A, eps_s] theorem ell_eq : ell = 627 / 2500 := by norm_num [ell, A, eps_s] /-- The candidate gap `u - δ - ℓ = 2 ε_s - δ = 3/1000` is positive. -/ theorem candidate_gap : u - delta - ell = 3 / 1000 := by norm_num [u, ell, A, eps_s, delta] theorem candidate_gap_pos : ell < u - delta := by have h : u - delta - ell = 3 / 1000 := candidate_gap linarith theorem candidate_of_eps : u - delta - ell = 2 * eps_s - delta := by norm_num [u, ell, A, eps_s, delta] /-- The candidate violates `u - d < ℓ` (`2 ε_s = 3/200 > 3/250 = δ`). -/ theorem candidate_two_eps_gt_delta : delta < 2 * eps_s := by norm_num [delta, eps_s] /-- The source's own parameter set satisfies `u - d < ℓ`. -/ theorem source_gap_neg : uSrc - dSrc - ellSrc = -32 / 10000 := by norm_num [uSrc, ellSrc, dSrc] theorem source_gap_lt : uSrc - dSrc < ellSrc := by have h : uSrc - dSrc - ellSrc = -32 / 10000 := source_gap_neg linarith /-- There is a base radial sum strictly between the candidate's base cutoff `ℓ` and `u - δ`: the interval on which the source's "automatic" argument has nothing to say. -/ theorem candidate_interval_nonempty : ∃ s : ℚ, ell < s ∧ s < u - delta := by refine ⟨(ell + (u - delta)) / 2, ?_, ?_⟩ <;> linarith [candidate_gap_pos] /-! ### Candidate cap constants in the engine's normalisation `u = t/A` `dc = δ/A = 40/861`, `c_1 = (3/20)/A = 500/861`, `c_m = (4/25)/A = 1600/2583` for `m ≥ 3`, and `2 ε_s/A = 50/861`. -/ theorem dc_eq : delta / A = 40 / 861 := by norm_num [delta, A] theorem c1_eq : (3 / 20 : ℚ) / A = 500 / 861 := by norm_num [A] theorem cm_eq : (4 / 25 : ℚ) / A = 1600 / 2583 := by norm_num [A] theorem two_eps_over_A : 2 * eps_s / A = 50 / 861 := by norm_num [eps_s, A] /-- The candidate threshold in the engine's normalisation is `1/A`. -/ theorem threshold_eq : 1 / A = 10000 / 2583 := by norm_num [A] end TwinPrime0025