---
license: mit
library_name: pytorch
tags:
- hamiltonian-neural-fields
- chaotic-dynamics
- zkaedi-prime
- dynamical-systems
---

# ZKAEDI Hamiltonian Neural Field (HNF) Ensemble v1.0.0

This model repository stores the verified 4-model HNF ensemble checkpoint and its associated validation code.

## Identity Hash
* **SHA-256 Checksum**: `BCC82239F2F9133E7630B58678491A731FFA9C85195254A715E941CBD115EFE7`

## Models in Ensemble
1. `tight_150` (Teacher model)
2. `seed_1`
3. `seed_2`
4. `seed_3`

## Ensemble Weights
* `[0.0689, 0.3133, 0.4868, 0.1310]` (inverse-loss softmax weighted)

## Performance Metrics
* **Fidelity $R^2$ vs Teacher**: `0.693694`
* **Mean Uncertainty $\sigma$**: `0.012251`
* **Max Uncertainty $\sigma$**: `0.017168`
* **Damping Policy**: Verified always negative for $H \in [-10, 10]$

## Known Limitations (Gate 7 Warning)
The $0.05$ mixing coefficient is too weak to overcome the exponential amplification term for positive $H$ inside the standard PRIME feedback loop. 
For correct convergence behavior, run in **standalone dynamics engine mode**:
$$H_{new} = H + \text{ensemble}(H, t_{norm})$$

Or scale with a much larger $\alpha$ parameter matching the Dream Engine rollout specifications.

## Execution
Run the embedded evaluation script locally:
```bash
python -Xutf8 eval_hnf_ensemble.py
```