Assessing the energy budget of long-lasted active regions: NLFFF methods using a physics informed neural network

This dataset contains YAML configuration files for NF², a Physics-Informed Neural Network (PINN), designed to model solar active regions (ARs) using SHARP (Space-weather HMI Active Region Patch) data. The files define the architecture, training parameters, and physical constraints for simulating magnetic vector potentials and related electromagnetic fields in solar physics. Key Features: File Naming Convention: Format: SHARP_{ID}_{DATE}_{TIME}_{TYPE}_{DESCRIPTION}.yaml SHARP_{ID}: Unique identifier for the solar active region (e.g., 8032, 8088). {DATE} & {TIME}: Timestamp of the simulation or data snapshot (e.g., 2022-02-24T08:00:00). {TYPE}: Configuration type, including: batch{XXX}_type_vector_potential_series_{N}: Batch training setups (e.g., batch100000) for vector potential (B_x, B_y, B_z) with series-specific parameters. series_{N} or series_short: Time-series configurations for shorter or full datasets. type_vector_potential_epoch{XX}: Epoch-based training (e.g., epoch20) for iterative optimization. Purpose: Configure PINN architectures to solve magnetostatic and electromagnetic equations (e.g., ∇·B = 0, Faraday’s law). Define training hyperparameters (batch size, epochs, optimizers) and physics-informed loss functions (balancing data and physics constraints). Specify input data (e.g., SHARP magnetic field measurements: Bx, By, Bz) and boundary conditions. Applications: Solar Physics: Reconstruct magnetic fields and vector potentials in active regions. Data-Driven Modeling: Combine observational data (SDO/HMI) with physical laws to improve accuracy. Reproducibility: Standardized configurations for sharing and reusing PINN setups in solar research. Workflow Integration: Compatible with frameworks like TensorFlow, PyTorch, or DeepXDE for PINN implementation. Designed for scalable training (large batches, time-series, or epoch-based). Note: These files are part of a reproducible pipeline for studying solar magnetic fields using machine learning and physics-informed constraints. For full reproducibility, pair with the corresponding SHARP data files (e.g., .h5 or .fits).