How mihapompe/pinn-nuclear-decay-equation works
This repository is a research implementation of physics-informed neural networks for nuclear decay and burnup equations. Its main workflow is to provide a system matrix and initial isotope composition, construct a decay or burnup PINN, train it over time, and obtain a predicted isotope evolution. The README also describes real- and complex-eigenvalue architectures and analysis of accuracy and stiffness. The documented example driver is absent from the supplied tree, and excerpts do not establish how solvers or analysis code are wired, so those relationships are left independent.
Main entry points: Bateman_PINN.py Decay and burnup PINNs · A_matrices Matrix data · SineCosineLayer.py Sine/cosine layers · CRAMsolve.py CRAM solver
Components 9
PINN models
- Decay and burnup PINNs
Bateman_PINN.py - PINN training
Bateman_PINN.py - Sine/cosine layers
SineCosineLayer.py
Numerical methods
- CRAM solver
CRAMsolve.py - RK45 solver
RKsolve.py
Analysis
- Experiment analysis
analysis/run_analysis.py
Other components
- Researcher
- Matrix data
data/A_matrices - Saved model
burnup_model.pt
Connections 10
- Researcher → Matrix data: selects inputs
- Researcher → Decay and burnup PINNs: configures model
- Matrix data → Decay and burnup PINNs: supplies matrix
- Decay and burnup PINNs → PINN training: trains network
- PINN training → Decay and burnup PINNs: updates weights
- Decay and burnup PINNs → Sine/cosine layers: uses for burnup
- PINN training → Saved model: saves model
- Researcher → CRAM solver: requests solve
- Researcher → RK45 solver: requests solve
- Researcher → Experiment analysis: runs experiments