Papers

30 papers across two research programs

The Hodge-Epsilon Program

ε² = 0

Four papers forming a loop: empirical measurement of byte streams reveals algebraic structure, which is formalized and proved in Lean 4, whose proof terms are measured for Kolmogorov complexity, and the complexity measurement validates the empirical starting point. Program overview.

Paper 1

The irreversibility depth of natural language

The Hodge decomposition of byte-level Markov fields defines D* — the depth at which the time arrow vanishes. D* = 7.5 bytes for English. A formality ladder from MIDI music (D* ~ 4) through programming languages to DNA (D* ~ 165). Cross-linguistic atlas of 49 languages.

Read PDF Full paper
Paper 2

The Kolmogorov complexity of machine-verified proofs

K(ε² = 0) = 5 DAG nodes exactly — the first exact Kolmogorov complexity of a mathematical proof, with matching upper and lower bounds proved in Lean. The richest theorem is the best axiom. The self-measurement tower stabilizes at Level 2–3. First machine-verified K-bound via formalized PPM-C compressor (2.16 bpb).

Read PDF Full paper
Paper 3

The ε²=0 tower: from axiom to self-measurement in Lean 4

One axiom forces a 17+ layer tower from elementary ring theory through homological algebra to operadic Koszul duality. The Krivine bridge connects abstract algebra to lambda-calculus reduction. 175 files, 28,000+ lines of Lean 4.

Read PDF Full paper
Paper 4

The Hodge language model

A three-layer architecture derived from the Hodge decomposition. Layer 0: zero-parameter vertex potential. Layer 1: harmonic correction (60K params). Layer 2: residual (200K params). The harmonic layer is 1.75× more parameter-efficient than the residual. Best: 2.26 bpb with 80K parameters.

Read PDF Full paper
Companion

Systematic byte-level compression with a machine-verified Kolmogorov bound

Twenty controlled experiments on PPM-C. From 3.68 to 2.16 bpb — a 41% reduction with zero learned parameters. First machine-verified K(x) upper bound via Lean 4 formalization with zero sorry's.

Read PDF Full paper
Companion

𝓕 — A unified theory of description

A single functional measuring model-data pairs across three system classes: deterministic, stochastic, and open. 𝓕(m, x) = Lmodel(m) + Lresidual(x | m). Three proven properties: monotonicity, encoding invariance, composition bound.

Read

The singular Hecke node at level 163

η² = 3η

Fourteen papers on one relation. The Hecke algebra at level 163 has a singular 3-adic factor — a split nodal curve where two branches of modular forms collide modulo 3. From this single quadratic identity: a trace inequality, a syndrome algebra, a complete module classification, a singularity category, a canonical Hamming code, a quantum stabilizer code, a finite-state automaton with confinement, a drift rate theorem, and a census across all prime levels.

Paper I

A trace inequality for Hecke operators at level 163

A bound on the trace Gram matrix of Heegner operators. Specific to the Heegner set, failing under enlargement. Case-by-case proof via Cauchy–Schwarz, Eichler–Selberg, orbit decomposition, and Hurwitz class-number sums.

Read PDF Full paper
Paper II

The Heegner syndrome algebra

The mod-3 Hecke action on the canonical 3-dimensional summand collapses to F₃ × F₃[ε]/(ε²). Projective unit group C₆. Orbit decomposition 1+1+2+3+6. The canonical collision T₄₃ = T₆₇ is exact and base-free.

Read PDF Full paper
Paper III

The singular 3-adic Hecke node

The complete local theory. Osing ≅ Z₃ × Z₃[η]/(η² − 3η). Three CM indecomposables. Singularity category with Σ² = Id and K₀ ≅ Z/2Z. Centre recovery: Z(End(M)) ≅ R. The node reconstructs itself.

Read PDF Full paper
Paper IV

Canonical Hamming geometry from the Hecke singularity

The Hecke module canonically produces PG(2,F₃) and the [13,10,3]₃ Hamming code. No choice at any step. CSS quantum code [[13,7,3]]₃. Census of singular 3-adic factors at 164 prime levels.

Read PDF Full paper
Paper V

The Hecke automaton at conductor 163

Six Hecke gates act on the 3-adic formal group of E = 163a1 as a finite-state automaton. Three theorems: monotonicity (conductor depth is a monotone counter), boundary symmetry (Z/2Z from T₇), confinement (the matter sector is one-way reachable).

Read PDF Full paper
Paper VI

A census of singular 3-adic Hecke factors at prime level

All primes ≤ 1000 classified as smooth, cusp, split node, or non-split node. Split nodes at 71, 127, 163, 167, 199, 269, 271, 307, 337, 359, ... Each is an instance of the η² = 3η singularity.

Read PDF Full paper
Paper VII

The drift rate of the Hecke automaton

Under uniform random gating, the mean shell increment is exactly 2/5. After n gates the increment is Bin(n, 2/5)-distributed. Hoeffding concentration. The constant depends only on the multiset {v₃(ap)}.

Read PDF Full paper
Paper VIII

Gate signatures of class-number-one Heegner-prime newforms

Five Heegner-prime levels, five distinct gate signatures, five distinct drift rates. Level 67 is frozen (zero drift). Level 43 has the highest drift — conductor 163 is not extremal.

Read PDF Full paper
Paper IX

Confinement from the conductor

The defect spectrum of the singular order R = Z₃[η]/(η² − 3η). Three sectors (one vacuum, two matter). K₀(Dsg(R)) ≅ Z/2Z. Singularity category with Σ² = Id.

Read PDF Full paper
Paper X

Hecke trace geometry and the cotangent Laplacian

The Hecke distance matrix determines the centered Gram matrix via double-centering. The cotangent Laplacian on K₇ is near-simultaneously diagonalizable with B (commutator norm 0.030, overlaps 0.931–0.997). Strongest at level 163 among {43, 67, 163} but not globally extremal.

Read PDF Full paper
Paper XI

The trace inequality does not extend uniformly

A negative result. The trace inequality from Paper I holds at N ∈ {11, 43, 163} but fails at N ∈ {19, 67}. Worst ratio 14/10 at N = 67. The inequality is not a Heegner-family phenomenon.

Read PDF Full paper
Paper XII

The quantum stabilizer code from the Hecke singularity

The CSS construction on the self-orthogonal pair (C, C) yields a quantum stabilizer code [[13, 7, 3]]₃. The Hecke operators act as code automorphisms. No choices at any step.

Read
Paper XIII

Ginzburg-Landau on the Hecke lattice

A negative result. The dissipative GL quench does not reproduce the eikonal survivor hierarchy (Spearman 0.071). The hierarchy is dynamical, not a trivial gradient-flow consequence of the Hecke distance quartic.

Read
Paper XIV

Local ring refinement beyond mod 3

T₂ is primitive over Q but the monogenic order has index 3³. The local factor Rloc has Loewy length k+1, with T₁₁ as generator. The canonical collision T₄₃ ≡ T₆₇ splits modulo 9.

Read

Research notes

Self-contained results from the research program. Each note has one claim, one proof path, and a link to the full writeup.

Note

The NilSquare hub: 17 theorems from one axiom

ε ≠ 0, ε² = 0 implies nilpotency, automatic differentiation, periodic resolution, Hamming codes, Koszul duality, Church-Rosser, and 11 more. Lean-verified with zero sorry's in the hub file.

Read PDF Full paper
Note

Church-Rosser in three lines

Confluence for deterministic evaluators follows from the partial-function property in three lines of Lean. No diamond lemma, no parallel reduction.

Read PDF Full paper
Note

Actor-critic RL decomposes as Hodge

The advantage function in actor-critic reinforcement learning is the harmonic component of the Hodge decomposition on the state-transition graph. The value function is the exact (potential) component.

Read PDF Full paper
Note

PPM escapes are Bayes-optimal

The PPM-C escape probability u/(t+u) is the maximum-likelihood estimator for the probability that the next symbol is novel given the current context. Empirically: 1.66 bpb with 0 learned parameters.

Read PDF Full paper
Note

The universal beta function

β(g) ≈ −0.7g across all 49 languages, independent of word order, morphology, or script. Every language is asymptotically free. The running coupling g(D) = g₀ exp(−0.7D) has one language-specific parameter.

Read PDF Full paper
Note

Spectral phylogeny via Ihara zeta

The Ihara zeta function of the byte-level de Bruijn graph produces a linguistic phylogeny from pure spectral data. Welsh ≡ Tagalog at opacity r = 0.12.

Read PDF Full paper
Note

ε² = 0: the axiom

Why this axiom and not another. The dual-number ring k[ε]/(ε²) is the simplest non-trivial quotient — the minimal perturbation theory, the boundary between reversible and irreversible.

Read PDF Full paper
Note

The history space

Computation is not a sequence of states but arithmetic of admissible histories between boundaries. The Platonic object is the space of histories 𝔥.

Read PDF Full paper
Note

The Conway system: compression as game theory

40 Lean 4 files, ~14,500 lines. Conway's star game connects NilSquare algebra to encoder/decoder duality. Layers from game core through byte mechanics, scattering, Hamiltonians, to guarded MCTS search.

Read