RESEARCH BRIEF · FORMAL SYSTEMS · NOVEMBER 21, 2026
Architectural blueprint of the 17.37-million-line Lean 4 Riemann Hypothesis archive.
On November 20, 2026, Levent Alpoge released a 175,187-file Lean 4 formalization of the Riemann Hypothesis and Generalized Riemann Hypothesis (alpo.ge, Lean v4.33.1, Mathlib 0df444a3). Below is a verified structural map of the live proof path, its zero-axiom scan, and the 185-interface boundary separating classical literature from the core theorem chain.
01 / LIVE PROOF PATH
Five stages from foundational Galois representations to RH and GRH.
All five top-level exports in Main.lean pass through MainTheoremsAll, MainTheoremsCritical, MainTheorems, and a 14-line master switchboard in MainConclusion.lean. Although module imports also include an earlier pre-RecTExists shadow path (EveryCharacterE and ShapeWin) for shared definitions, the live proof executes through the ThirdPlaceSeries spine mapped below.
STAGE 05
Constructed Primitives
RowSrc combines definite unitary theta lifts (TPGHolds to TLiftK) with the holomorphic Lefschetz fixed-point formula (WeightOneGeometricRealization) to emit h : Row0b recT.
RecTExists builds Local Langlands on w.adicCompletion E across 12 laws (5 restricted to Good representations).
Firewall Closure
EveryCharacterOfLiteratureOfVWin.lean bundles 185 classical literature-interface wrappers into 9 typed hypothesis records. 177 wrappers are completely independent of OpenLiteratureStatements; its final 3 fields (2 distinct types: Row and RecTExists) are closed in MainConclusion.lean by 〈h, recTExists, h〉.
STAGE 04
Inputs & Machinery
Single top-level input h : RowT (entering LiteratureAutomorphyHypothesis.gal) plus unconditional rank-n isobaric base change (baseChangeN), weak descent (descWeak), quadratic base change (quadBC2), Ferrero-Washington, and LemmaK.
Output Exported
VWin (∀ ψ, VWinAt ψ 401): across cyclotomic-twisted fields built from prime pairs (q ≡ 3 mod 8, d ≡ 7 mod 16, q | d+1), bounds Σ log(l) Σ_(k < floor(x^401)) shaLen(ld l, k) ≤ exp(x^ε / 2) via five counting regimes (CountI Kato Selmer vanishing, CountII/III adjoint lifting & Sp4 rigidity, CountV unitary lift, countZ PropZ).
STAGE 03
Live Assembly
Synchronizes grid thresholds at max CF0 CF1, combining row rationality, Fourier-coefficient fit (ConstantFunctionFits), linked-bundle covolume bounds (CovolumeBoundOfX from 15 Arakelov, toroidal, and Chow inputs), and 2-and-3 prime bounds (BoundAtTwoThreeFamilyFonAlt).
Theta-Lift Tower & Output
Supporting FunctionModelRallisFromInputs proves archimedean Rallis descent, Bessel unfolding and seesaw propagation (stmt2c), and Rallis inner-product definite lift chains π_0 → … → π_m (stmt3c). Emits the 46-field container Nonempty (HypsAlt S0.NL DefiniteSideCondition).
STAGE 02
Unpacking & 9 Closed Links
forall_goal_of_shape_of_hyps unpacks all 46 fields of HypsAlt (splitting h.σ into 6 projections plus 2 external props) into the 53 parameters of TopShapeAlt. shape_closed_link_n_of_green forwards LinkN and builds 8 Closed links: LinkEta, LinkIRR (sole use of Green), SmallLevelRankEq, LinkDIVt (Cabs=2), ChosenE Fm 3 CH → NuFullF ∧ MuRank, LinkG, TuSize → LinkS Fm 3, and TuSize → DGamSize → LinkGLOBAL Fm 2 CH 3 2 0.
Export to Analytic Stage
DesWin.topDesWin intersects all 9 asymptotic links with DesignExists to prove FamilyClassConnected (ChainTied). SupplyChi converts VWinAt ψ 401 into HostExW (3/2) (GChi ψ).req.
DivinNOfHypV.lean): LinkDIVt cannot fire from LinkN alone because LinkN carries the prime-weighted shaLen error sum. Stage 4's VWin supplies HypV, bounding that sum by √b ≤ b (with (log x)² ≤ x^ε eventually). divinN_of_hypV absorbs both error terms into DIVinN, unlocking LinkDIVt, deducing LinkMu inside DesWin.topDesWin, and firing LinkGLOBAL.STAGE 01
Half-Plane to Critical Line
Uses once-integrated von Mangoldt sums (An.G, An.gChi) and weighted cyclic levels Level.Gm with weights 1/(3o+1) (psK := none on Level.cyc ψ) to obtain ZeroFreeAll on Re(s) > 1/2. Imprimitive lifts to all characters via Euler-factor nonvanishing; RhOfRight (uncompleted functional equation) and CritLine / ClaimStrip (completed functional equations) reflect zeros from 0 < Re(s) < 1/2.
Five Final Exports (Main.lean)
riemannHypothesis · riemannZeta_eq_zero_re · LFunction_ne_zero_of_half_lt_re · LFunction_eq_zero_re (full GRH) · LFunction_eq_zero_re_of_isPrimitive (primitive GRH). Definitions in Statements.lean import standard Mathlib only.
02 / SCALE ANATOMY
Why the repository is 17.37 million lines.
Nearly half of the entire archive sits in Development/AbelianVarietyTheory (8.14 million lines across 99,187 files). A source scan across all 175,187 files found zero custom axiom, sorry, or admit tokens; decades of classical algebraic geometry, representation theory, and automorphic forms appear in source alongside the core RH argument.
Top 10 Subsystems (within MainTheorems, 175,179 files)
Line volume and share of the 17,374,215-line repository
Apex Import Funnel (BFS depth 0 to 42)
Modules by shortest import distance from Supplements.Kernels.MainTheorems
The full dependency graph funnels through just 12 modules at distance 0 to 2 before expanding to 423 modules at distance 3 and peaking at 16,736 modules at distance 9.
- Apex bottleneck (dist 0 to 2)
- 12 modules
- First expansion (dist 3)
- 423 modules
- Peak layer (dist 9 / max 42)
- 16,736 modules
03 / CLASSICAL BILL OF MATERIALS
The 185-wrapper literature firewall.
EveryCharacterOfLiteratureOfVWin.lean declares 195 theorems (185 individual literature-interface wrappers, 9 composite bundle builders, and 1 master assembly call, plus 1 structure OpenLiteratureStatements). All 185 wrappers are transitively used; 177 are completely independent of OpenLiteratureStatements. Raw wrapper count includes 1 no-op slot (UnusedSlot72 := trivial), cross-namespace re-exports, and internal combiners.
| Mathematical Domain | Wrappers | Representative Formalized Interfaces |
|---|---|---|
| Automorphic Forms, Theta Correspondence, Siegel-Weil, Rallis, Doubling & Base Change | 44 | Siegel-Weil (unitary, convergent, Schrodinger, standard), Rallis inner product, Doubling method, Howe duality (real and GL), Weil/metaplectic representation, Stone-von Neumann, Kudla splitting, Flath factorization, Jacquet-Shalika rigidity, Gelfand-Kazhdan, Clozel purity, Base change (includes 1 no-op UnusedSlot72) |
| Galois Cohomology, Class Field Theory, Hermitian Forms & Automorphic Galois Reps | 36 | Poitou-Tate, Greenberg-Wiles, Local/Global Tate duality and Euler characteristic, Bloch-Kato local, Global Artin reciprocity, Ray class fields, Chebotarev, Hasse norm theorem, Landherr, Hermitian Hasse principle, Caraiani local-global, Polarized automorphic Galois representations |
| Local Langlands, p-adic & Finite GL_n Representation Theory | 35 | Local Langlands for GL_n, Local Converse, Bernstein-Zelevinsky finiteness and geometric lemma, Zelevinsky polynomial ring, Langlands classification/multisegments/exhaustion, Moy-Prasad, Depth-zero induction, Borel-Casselman, Green characters, Weyl branching |
| Abelian Varieties, CM, Deformations, Toroidal Compactifications & p-adic Hodge Theory | 23 | Elliptic Main Theorem of CM, Grothendieck-Messing and Grothendieck deformation, Semistable reduction, Toroidal compactifications and Siegel toroidal integrals, Dual abelian schemes, Zarhin, Fontaine B_cris, de Rham implies potentially semistable, Bhatt-Scholze integral/crystalline lattices, Prismatic Dieudonne-Hodge, Fargues exactness |
| Arakelov Geometry, Heights, Resolution of Singularities & Complex Geometry | 21 | Resolution of singularities and principalization in char 0, Arithmetic Hilbert-Samuel, Kawamata-Viehweg, Nadel vanishing, L^2-extension, Poincare-Lelong, Arakelov cycle heights and Chow forms, Formal algebraization, Serre holomorphic sections (plus 2 unwrapped Moriwaki/Hilbert-Samuel proofs) |
| Real Groups, Harish-Chandra Modules & Discrete Series | 17 | Harish-Chandra discrete series existence, classification, lowest K-type, analyticity, and finiteness; Casselman-Wallach; Dixmier-Malliavin factorization; Schur orthogonality; Formal degree compact dual |
| Analytic Number Theory, Hecke L-Functions & Deligne-Ribet | 9 | Hecke-Tate functional equation, Partial zeta functional equation, Deligne-Ribet, Siegel-Klingen-Deligne-Ribet, Linnik's theorem, LogDerOne, join_out_of_literature |
| No domains match this filter. Clear the filter to restore all 185 wrappers. | ||
| Total matching individual wrappers | 185 | 177 O-independent · 8 O-dependent (closed by RowSrc and RecTExists) |
04 / VERIFICATION BOUNDARY
What this study verifies, and what remains below the spine.
Formal audits are only trustworthy when their exact boundary is stated alongside the findings. Source archive: alpo.ge (released November 20, 2026).
Verified directly in source
Full-archive scan + 86-file live spine audit
- Zero custom
axiom,sorry,sorryAx,admit,unsafe,implemented_by, orexterntokens across all 175,187 files. - Exact live proof-term call chain from
Main.leanand standard MathlibStatements.leanthroughThirdPlaceSeries,DesWin,VwinOfRow,RowSrc, andRecTExists. - Exact formal signatures for key bridge predicates (
VWin,VVle,Row0b,TLiftK,RecTExists,HypsAlt,LinkN,LinkDIV,BesselId,LemmaK,KatoTwoW).
Below this blueprint's depth
Deeper leaf proofs and local builds not run in this top-level map
- Local Lean 4 compilation and
#print axiomskernel verification of the full 17.37M-line build (reported in the archive release onalpo.geagainst standard Mathlib axioms; not re-executed locally in this structural source audit). - Internal proof bodies of
zeroFreeOn_of_tiedWin_cyc,galoisLocalGlobalStrong_of_geometry_lift,sp4Main_alt,descWeak, andbaseChangeN_of. - Disjointness proofs of
Host.CaseI/II/III/V/Zand the internal call site wiringFormalism.BehrendSetsinto host construction. - Individual statement-strength audits for each of the 185 wrapped
GlobalTheory.Published.*interfaces.
FORMAL SYSTEMS & CODEBASE AUDITS
Need an auditable specification map for a large formal or regulated codebase?
Dark Vector Cognition delivers fixed-scope architecture audits, specification-boundary reviews, and accessible compliance systems for federal, state, local, and regulated engineering teams.