KeffNet: Neural Neutron Transport — Monte Carlo Simulation Data
Description
This dataset contains simulation data and results for the paper "KeffNet: Neural Neutron Transport (KNNT) — A Physics-Informed Neural Network for Solving the Neutron Diffusion Eigenvalue Problem." Files: - ground_truth_fission_rates.csv: Ground truth fission rates for pure isotopes - predictions_final_debug.csv: Model predictions for diverse nuclear materials - analytic_keff_test.csv: Analytic keff values for validation - collocation_points.csv: Collocation points used for training - geometry_regions.csv: Geometry definitions for fuel and reflector regions - interface_points.csv: Points at the fuel-reflector interface - materials.csv: Material cross-section data - validation_flux.csv: Flux validation data All data was generated through Monte Carlo simulations and PINN training. Results are fully reproducible using the code available at: https://github.com/drgon19940-afk/monte-carlo-neutron-ml
Files
Steps to reproduce
Quick Setup bash # 1. Clone repository git clone https://github.com/drgon19940-afk/monte-carlo-neutron-ml.git cd monte-carlo-neutron-ml/results/last%20phase/last_phase_3D_Pinn # 2. Install dependencies pip install -r requirements.txt Requirements: Python 3.10+ PyTorch 2.0+ NumPy, Pandas, Matplotlib, SciPy Training bash # Generate training data (Monte Carlo simulations) python generate_data.py # Train PINN (Phases A, B, C) python train.py Training phases: Phase Laplacian Epochs A None 1-100 B Hutchinson 100-300 C Exact chunked 300-500 Evaluation bash # Evaluate on unseen isotopes (²³⁵U, ²³⁹Pu, Mix 50-50) python evaluate.py --model models/phaseC_model.pth Outputs: Flux and Laplacian slices evaluation_summary.csv (RMSE, R², keff error) Uncertainty Quantification bash # Input perturbation (1% domain size, 100 samples) python uncertainty.py --model models/phaseC_model.pth Outputs: keff_uncertainty.png — keff with ±1σ bars Flux uncertainty maps for all phases Speed Benchmark bash python benchmark.py Expected: Method Time Speedup Monte Carlo 180 s 1× PINN (CPU) 0.005 s 36,000× PINN (GPU) 0.0008 s 225,000× Key Results Summary Metric Phase A Phase B Phase C Boundary Error 0.000 0.000 0.000 Flux R² 0.999924 0.999947 0.999900 keff Error 4.65×10⁻⁴ 4.53×10⁻⁴ 2.82×10⁻⁴ Flux Uncertainty 1.250 0.621 0.106 Files File Description ground_truth_fission_rates.csv Ground truth fission rates predictions_final_debug.csv Model predictions for diverse materials collocation_points.csv Training collocation points geometry_regions.csv Domain geometry (Ω_f = [2,18]³ cm) validation_flux.csv Flux validation data Citation text Mukherjee, K. (2026). KeffNet: Neural Neutron Transport — Generated Data. Mendeley Data, v1. https://doi.org/10.xxxx/xxxxx Contact Kunal Mukherjee — drgon19940@gmail.com