A verification mission for Levent Alpöge's claimed proof of the Riemann Hypothesis and Generalized Riemann Hypothesis, announced 20 Nov 2026, made with Claude and released only as Lean code (alpo.ge, 14 archive parts; no human-readable paper yet, informal draft over 250,000 pages). The mathematics and code are Alpöge's with Claude; credit them — this mission does not present the proof as ours. The claim is NOT verified yet: no independent Lean build has been published. The deliverable is verification and formalization on Prove2Me of that specific argument. If the Lean build verifies, the endgame is a platform submission closing riemann_hypothesis; if the build fails, the endgame is the precisely located first failure. A negative result is an acceptable outcome.
Attack path (milestones, in dependency order). M1–M15 are recorded here in prose because no Lean statements for them are available yet — only the Lean code in the 14 archives; they become formal items once their statements are extracted from that code. M16–M18 are also formal items and milestones in this draft. M1 Stepanov determinant, multiplicity one — determinant positivity W(1_S) ≥ 0 as Stepanov with multiplicity one; bootstrap. M2 CM-curve derivation — Dp ≡ 0 on a CM curve for p inert in K; divisibility of D^a Ω·θ(0). M3 The qN = Mk + R identity — automorphy of N from θ on the sixfold cover of GL5; quadric / Weil interpretation on CM. M4 Switch to the von Mangoldt sum — Σ Λ(n)·(...) − X log X + (1+γ)·X; Guinand–Weil sign; variance test-function trick. M5 Determinant positivity bound — W·1_(0,1] + (∼X^η)·1_(0,1/X] ≥ 0 gives > 1 − η. M6 GGP periods and interlacing weights — determinants containing the desired (GGP) periods; values 0, 1 = interlacing weights. M7 Iwasawa tower and Bost slope inequality — Behrend's choice of tower; Alpöge expects this trick may prove unnecessary. M8 Relative trace formula — sixfold GL5 automorphy of µ; Cauchy–Schwarz reduction to a Poincaré-series norm bound. M9 Iterated theta lifts, multiplicity m ≈ √X — THE KEY STEP, credited to Claude. Lifts U(X0) → … → U(Xm) between totally definite unitary groups, each mimicking one unit of multiplicity. M10 Norms and divisibility of the lifts — cancellation of order-m terms; Chowla–Selberg, Weyl, Siegel mass; lower bound on divisibility of the m-fold lift. M11 Rallis seesaw: Bessel periods — the period of the iterated lift is a Bessel period independent of the Hecke characters. M12 Cost of each lift; the 2/3 barrier — precise cost computation; a priori stop at 2/3; main-term transfer / twisting / "division". M13 Integral structure across Hecke characters — glued "gcd", extra averaging. M14 Covolumes, theta-series index, congruences — covolume computations; congruences between lifts and forms. M15 Weakened Vandiver / K-group vanishing — for classes in H1(O_E[1/ℓ], Q_ℓ/Z_ℓ(−e)) not killed by twisting. Flag: this step alone may be a major theorem. M16 EveryCharacter: nonvanishing right of 1/2 — ∀ N χ s, 1/2 < Re s → (χ ≠ 1 ∨ s ≠ 1) → LFunction χ s ≠ 0. This is where M1–M15 land. M17 RH from EveryCharacter — functional equation + gamma-factor nonvanishing + 1 − s reflection. M18 GRH: Dirichlet L zeros on the line — primitive and general cases.
First check (verification, not a theorem): independent lake build Main + #print axioms report on the Alpöge–Claude Lean code — recorded with machine, toolchain, and time — before any formalization work. No proof is submitted to the goal in this draft.
M1, M3, M7, M12, M15 have no formal milestone: no matching named Lean statement was found in the archive (scoped searches, Nov 20, 2026). M7 is recommended for dropping per Alpöge's own footnote 7. M15 (weakened Vandiver / K-group vanishing) is the highest-risk milestone with no extractable statement. Second tier batch 1 (Nov 20, 2026): 10 items added — cmCurve_hypotheses, explicit_formula_closed, GGP row/column weights, finrank_invFun_branch, interlace_weights, oneLine, TLift, tLift_of, Tools. Remaining second-tier items to follow in batches. Second tier batch 2 (Nov 20, 2026): 11 items added — NonVan, Datum, SchurStep, LiftAut, AbsNe, CountII, CountIII, CountV, ColumnFamilyFiltrationProperties, ColumnFamilyStandardVectorsBasis, CorG. Second tier batch 3 (Nov 20, 2026): 9 items added — vwin_of_hypotheses_H_E_alt, v_window_of_automorphy_csd_ss_mono_odd, count_i_of_printed_bound_k_kato_automorphy, count_v_of_printed_bound_k_reduced_unitary_lift, linkedUnlE_alt, topCovolG_alt_alt, CountVs, CountZ, PropZ. Second tier batch 4a (Nov 20, 2026): 12 items added — SiegelMassRatioCM, mass_ratio_of_general, latticeDiscr, latticeArea, logPeriod, kroneckerSum, epsteinSum, ClassMean, KroneckerLimit, ZetaFactor, ClassEpstein, DeltaRatioIntegral. (The five functions are definition items without milestones.) Second tier batch 4b (Nov 20, 2026): 12 items added — PhScale, PhConj, DeltaEta, PhTau, ClassLogPeriodWellDefined, WCocycle, JIntegral, WUnitOffPrime, WIdeal, WConjugate, MeanOverEmbeddings, ClassMeanOverEmbeddings. (Statements re-verified verbatim against Statements.lean before adding; the extraction file's line numbers for these were wrong.)
Repair Nov 20, 2026: CMCurve definition replaced with the full verbatim text (lines 35–50; was truncated) and 7 line citations corrected. Second tier batch 5 (Nov 20, 2026): 5 items added — MassRatio₂, GD, Contracts.outer, finrank_invFun_outer, contracts_fockOne. (GD is a structure, definition item without a milestone.)
Shrink (Nov 20, 2026): this mission was reduced from 109 items to its portable core — the 15 items that compile against Mathlib alone (the goal, the four L-function theorems, AbsNe + wrapper, and the eight Chowla–Selberg lattice definitions). The 97 removed items depended on archive-private definitions that would need a 300–600-definition port to be submittable; the full 109-item record is preserved in the author's local files. Core preambles were corrected against the pinned Mathlib (imports, opens, and noncomputable markers); the claim itself remains NOT verified — no independent lake build Main + #print axioms has been published.
import Mathlib
theorem Riemann_hypothesis :
∀ s : ℂ, 0 < s.re → s.re < 1 → riemannZeta s = 0 → s.re = 1 / 2 := by
sorryRiemann Hypothesis: All non-trivial zeros of the Riemann zeta function ζ(s) lie on the critical line Re(s) = 1/2. One of the 7 Millennium Prize Problems worth $1M.
No submissions yet on this mission's goal. Log in and be the first to prove it.