Skip to content

IanD25/D_eff-Test-1.0

Repository files navigation

Fisher Diagnostic Suite

A multi-observable toolkit based on the Fisher Information Matrix that reads the information geometry of discrete systems. One construction, two regimes:

Geometric regime (exponential kernel): exact dimension estimation — integers on manifolds, Hausdorff dimension on fractals, DPI-monotone under coarse-graining.

Thermal regime (correlation kernel): structural probe — parameter-free detection and classification of phase transitions via spectral signatures in the FIM.

Author: Ian Darling | License: MIT


Quick Start

# Phase 1: Dimension estimation on a 2D torus (< 2 min)
cd PHASE1_GEOMETRIC/src/
python fisher_phase1.py

# Phase 2: Ising model phase transition detection (< 10 min)
cd PHASE2_THERMAL/src/
python ising_fisher_phase2.py

Results at a Glance

Geometric Regime (Phase 1)

System True d Fisher Rank Error
2D Torus 2 2.000 0%
3D Torus 3 3.000 0%
4D Torus 4 4.000 0%
Sierpinski Carpet 1.893 PR → 1.905 0.6%
ER Random Graph 1 (correct negative control)

Thermal Regime (Phase 2)

System Transition SV₂/SV₁ at T_c Swap?
2D Ising Continuous 1.000 YES
3D Ising Continuous 0.956 YES
Potts q=5 1st-order 0.691 NO
Potts q=10 1st-order 0.374 NO

Cross-Domain Transfer (Phase 3)

Domain Rank→1 at transition η extremum Systems
Lattice models 4 models
Financial networks 5/6 6/6 ~90 S&P 500 stocks, 2005–2024
EEG seizures 5/5 4/5 5 patients, 26 seizures

Repository Structure

D_eff-Test-1.0/
├── README.md                    ← You are here
├── PHASE1_GEOMETRIC/            ← Dimension estimation (Paper 1)
│   ├── README.md
│   ├── DS_Phase1_MANUSCRIPT_FINAL.pdf
│   ├── src/
│   └── results/
├── PHASE2_THERMAL/              ← Phase transitions + cross-domain transfer (Paper 2)
│   ├── README.md
│   ├── DS_Phase2_Thermal_Paper_FINAL_v2.pdf
│   ├── src/
│   └── results/
└── docs/                        ← Project documentation

Each phase is self-contained with its own README, source code, and results. All experiments reproduce on consumer hardware (M4 MacBook Air, 24GB RAM) in under 20 minutes per phase.


Papers

  1. Geometric regime: Effective Dimensionality from Fisher Information Rank: Operational Validation on Tori, Fractals, Random Graphs, and Coarse-Grained Lattices. PDF

  2. Thermal regime: Spectral Phase Transitions in the Fisher Information Matrix: Universality of the Degeneracy Swap and Classification of Transition Order. PDF


AI Disclosure

This research was conducted with AI-assisted development. Claude Opus (Anthropic) contributed to methodology design, analysis, and manuscript development. Claude Code (Anthropic) contributed to implementation. The author is responsible for all scientific claims, research direction, and editorial decisions.

About

DS Framework Phase 1: Validating effective dimensionality D_eff = rank(FIM) on benchmark torus lattices

Resources

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages