Patrice Portemann Independent Researcher in Applied Mathematics & Physics
My work focuses on breaking the paradigm of statistical induction in Artificial Intelligence — and deriving physical structure from geometric first principles.
Web ecosystem: histoire-des-sciences.eu · index.portemann.eu · portemann.eu · noeticindustries.com
A reproducible computational environment for testing field-theoretic structures. Given a candidate model (ansatz, Lagrangian, symmetry-breaking pattern) and measured anchoring data, it returns a verdict — existence, spectrum, quantisation, regime — with documented invariants, frozen protocols, and SHA-256 hashed artefacts. Zéro paramètre ajusté · levier discriminant · échecs publiés (B3-FAIL).
| Repository | Description | Status |
|---|---|---|
noetic-machine |
Cœur — banc fondateur SU(2) Georgi–Glashow, 5 predictions confirmed (BPS calibration, Bohr spectrum, Dirac charge, flux tube, phase diagram) | Public, v1.0 |
noetic-machine-complete |
Archive canonique — corpus P0–P45 + série A (méthode) + méta-chantiers M1/M1b : ~360 fichiers figés SHA-256, registre des frontières (18 entrées mesurées, falsifieurs exécutables), release de données Allen Cell Types | Public, actif |
noetic-applications |
Vitrine — 32 case studies on experimental data (P7–P31): atomic, nuclear, particle, condensed-matter, molecular, quantum chemistry | Public, v1.0 |
noetic-ash |
Acquisition — Analyseur spectral à géométrie harmonique : invariants topologiques O(1) par fenêtre, sans apprentissage, embarqué STM32/ESP32 ; v1.1.0 audité par la machine elle-même | Public, v1.1.0 |
noetic-ash-corpus |
Mémoire falsifiable — registre de signatures et verdicts ASH (statuts T > C > E > H, falsifieurs obligatoires) | Public, actif |
Fondations (P0–P4) : BPS calibration C(ρ=1) = 1.3098 · Coulomb pur à 10⁻⁴ · quantification de Dirac e·g = 2π exacte · tube de flux émergent · diagramme de phases à 2 régimes.
Applications (P7–P31) : 24 succès / 9 partiels-négatifs — déplacements isotopiques, effet EMC, Hückel, Geiger-Nuttall, Regge, Hall fractionnaire (Jain), isolants topologiques, moments de Schmidt, cusp de Kato…
Suite du corpus (août 2026) — la machine s'éprouve elle-même :
- 🧠 P34–P38 : le verdict porté aux architectures neuronales — biais constitutif, σ réfutée comme modèle biologique (B3-FAIL), profondeur constitutive dès que la tâche itère, position constitutive de l'attention (8/262144 sans, 262144/262144 avec), neurone fractionnaire éprouvé comme mémoire t^(−α)
- ⚛️ P39–P42 : frontière r₁₂ fermée (couple zéro paramètre en tout Z) · Z_max = 180 recomputé exact (α = 2⁻¹⁰ gelé) · neurones corticaux réels (Allen) : type I dominant, HH non représentatif · pont arithmétique 120 ↔ E₈ exécuté (quiver de McKay calculé)
- 🔧 Série A (méthode) : batterie de perturbation (le verdict devient un couple (V, Σ)), moteur de leviers, tamis de Jastrow, registre des frontières (18 entrées, 7 fermées / 5 ouvertes / 6 partielles) — dont F9 : réparation forensique du corpus (p31 corrigé, verdict historique 3/5 → 5/5)
- 📉 M1 / M1b (méta) : le postulat central du corpus éprouvé sur 23 chantiers — réfuté avec inversion (robustesse Kolmogorov-gzip, Spearman 0,980) ; ce qui s'effondre à r₁₂, c'est le taux de succès des confrontations externes, pas le ratio informationnel
- 🎛️ P43 : l'instrument sous le verdict — l'ASH auditée (7/7) : O(1)/fenêtre confirmé, ReN réfuté comme invariant physique (pente −0,996) → errata publiés, noetic-ash v1.1.0
- 🧬 P44 (B3-FAIL) : EEG d'imagerie motrice réelle (BCICIV-2a) — les règles zéro paramètre sont réfutées en essai unique ; l'illettrisme BCI est mesuré, la voie de fermeture déclarée
- 📊 P45 : les benchmarks tiennent sans ReN — classification officielle par invariants normalisés (10/10 paires, invariants d'amplitude à 1e-9)
A research programme constructing falsifiable physical predictions from non-commutative spectral triples. Four theorems proved, a new arithmetic law established, and numerical benchmarks validated on the Gross–Pitaevskii solver.
| Repository | Description | Status |
|---|---|---|
spectral-triple-minimality |
Article foundations — 4 theorems (dimension, k-bound, margin-3, non-uniqueness) + KO-6 arithmetic law | arXiv-ready, v1.0 tagged |
ko6-spectral-solver |
Spectral solver NUE — Strang splitting, benchmarks B1–B3 (Taylor-Green, KdV, Ising 2D) | Public, CI green |
koilon-scale-e8 |
Échelle Koilon-E8 — mapping du vide hydrodynamique à la géométrie E8/Lie ; confronté aux masses mesurées (P40 : Z_max = 180 exact, P-FISSION réfutée sur JEFF-3.1.1) et au pont 120 ↔ E₈ (P42) | Public, v1.0 |
gauge-non-abelian (private) |
Non-abelian gauge model — Georgi–Glashow SU(2)+Higgs functional C(ρ), Bogomolny bound, Dirac theorem | Active research |
Key results:
- 🧮 Theorem T1: dim H_F ≥ 2R+1 (exact, saturated) — Standard Model emerges at R=3, dim=7
- 🧮 Theorem T2: k ≥ 2 (no field without at least 2 vertices)
- 🧮 Theorem T3: max(m_ij) ≥ 3 for 3 pairs + chiral zero (proved via odd-margin lemma, 622,560 cases, 0 violations)
- 🧮 Theorem T4: Non-uniqueness of the minimal class at dim 7 (33,148 solutions)
- 📐 KO-6 Arithmetic Law: At m ≤ 2, all margins are even or odd perfect squares; ranks 5 and 7 forbidden; 9 = 3² is the largest exact bound
For the past two decades, I have been developing a Topological Fractional AI framework. The core hypothesis is that complex systems (biological signals, fluid dynamics, materials) should not be approximated by massive flat tensors (Deep Learning), but solved by embedding strict geometric priors directly into the state-space equations.
My engine replaces standard gradient descent with:
- Orthogonal Spectral Decomposition: Forcing the system into strict independent subspaces.
- Causal Coupling Matrices (M_ij): Solving transduction rules via algebraic inversion instead of backpropagation.
- Atangana-Baleanu Fractional Calculus: Embedding non-local memory (Mittag-Leffler kernel) to respect the true thermodynamics of the modeled system.
A "Gray-Box" solver capable of mapping complex non-Markovian dynamics with only 28 parameters, achieving comparable AUC to 10,000+ parameter Graph Neural Networks.
Relève dans le corpus : la dynamique fractionnaire est désormais exécutable et mesurée — P37 (neurone fractionnaire : mémoire t^(−α) éprouvée, spike réfuté), P43 (audit de l'instrument). Le script du modèle 84,1 % AUC reste à publier (registre F10, ouvert).
For institutional evaluation, technology transfer, or licensing inquiries: patrice@portemann.eu