Skip to content

eluckydog/flt-from-scratch

Repository files navigation

FLT-from-Scratch

A knowledge-graph-driven walkthrough of Fermat's Last Theorem — from the Frey curve through modularity, Serre, Ribet, to the contradiction. Inspired by arXiv:2508.10362. Built on mathlib4 with reference to the Imperial College FLT formalization project.

This is a study project. Not peer-reviewed. Not authoritative. Just a structured attempt to trace the proof chain of FLT from first principles.


What's Here

Python verification (pip install -e .):

from flt_verification import run_all
run_all()
# → 43/43 tests pass:
#   Eisenstein λ-descent (n=3)  ██████████ 10/10
#   FLT n=4 (infinite descent)  ██████████  2/2
#   Frey curve properties       ██████████  6/6
#   Elliptic curve group law    ██████████  3/3
#   Hasse bound                 ██████████ 11/11
#   Modular forms               ██████████  4/4
#   Hecke operators             ██████████  3/3
#   Galois reps (structure)     ██████████  4/4

Lean 4 references — proof sketches and concept mappings (see status note below):

File Content Status
lean/ch01_elliptic_curve.lean Elliptic curve definitions norm_num-computable + stub deep theorems
lean/ch02_frey_curve.lean Frey curve construction ✅ ring/norm_num verifiable + stub theorems
lean/ch03_modular_curves.lean Modular forms, Hecke operators ✅ framework only
lean/ch03_S2_zero_proof.lean dim S₂(Γ₀(2)) = 0 — critical fact ⚠️ native_decide-verifiable + stubs
lean/ch04_galois.lean Galois representations ⚠️ structural stubs
lean/ch05_deformation.lean Deformation rings, R = T 🔴 deep stub (R=T is ~100 pages)
lean/ch06_serre_ribet_wiles.lean Serre, Ribet, Wiles assembly 🔴 deep stub (Fields Medal level)
lean/flt_base_cases.lean n=3, n=4 wrappers ✅ imports mathlib4 FLT theorems
lean/flt_n3_lambda_descent.lean Self-contained Eisenstein descent ⚠️ key lemmas stubbed
lean/generated/*.lean 487-concept index → mathlib4 📋 mapping reference only

⚠️ Lean status note: These Lean files are not a complete, compilable formalization. They are proof sketches, theorem declarations, and concept mappings intended as reference material. Deep theorems (Serre, Ribet, Wiles R=T) are declared as theorem ... := by sorry — their complete proofs are the subject of the Imperial College FLT formalization project. Only the Python verification layer (flt_verification/) produces executable, pass/fail results.

Knowledge graph — structured concept map of the entire FLT proof chain:

  • flt_teaching.html — interactive chapter-by-chapter documentation
  • kg/mathlib4_concept_index.csv — concept-to-mathlib4 mapping

Coverage

Domain Concepts Verified
Abstract Algebra 105 ✅ depth theorems marked
Elliptic Curves 67 ✅ Frey curve, Hasse, group law
Modular Forms 107 ✅ dim S₂(Γ₀(2)) = 0
Complex Analysis 38 ✅ structural
Galois Representations 71 ✅ structural
FLT Proof Chain 88 ✅ numerical consistency
Topology 11 ✅ structural

Deep theorems (Serre conjecture, Ribet level lowering, Wiles R=T) are documented as stubs — their full proofs span hundreds of pages and are the subject of the Imperial College FLT formalization project.


How It's Made

This project was built through human–AI collaboration, using natural language alone. The human participant has no formal training in number theory or programming. All code was generated through dialogue with an AI agent using open-source knowledge (arXiv:2508.10362, mathlib4, Diamond-Shurman, and standard references).

No LLM fine-tuning or proprietary models were used. Only publicly available mathematical knowledge and tools.


References

  • arXiv:2508.10362 — Fermat's Last Theorem: A Proof from Scratch
  • Khare, Wintenberger (2004) — Serre's modularity conjecture
  • Wiles (1995) — Modular elliptic curves and Fermat's Last Theorem
  • Taylor, Wiles (1995) — Ring-theoretic properties of certain Hecke algebras
  • Diamond, Shurman — A First Course in Modular Forms

License

MIT

About

A study project tracing Fermat’s Last Theorem from first principles — Frey curve, modular forms, Galois representations, Serre-Ribet-Wiles contradiction — via a 487-node knowledge graph, 43/43 Python verification tests, and Lean 4 concept references. Built through human–AI collaboration using natural language alone. Not peer-reviewed.

Topics

Resources

License

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors