Solar Physics Questions to answer
Open questions to resolve during revision or project work.
NF2 / dissertation questions
- Can the NF2 workflow for AR 11158 be reproduced locally from HMI SHARP data with a documented setup?
- How does NF2 compare with a potential-field baseline?
- How well does the trained field satisfy force-free and divergence-free diagnostics across the volume?
- How sensitive are the results to model size, loss weights, sampling, crop size, and training time?
- Does a vector-potential representation improve solenoidal behaviour enough to justify extra derivative/training cost?
- Which diagnostics improve with training, and which just make the output look better without improving the physics?
Literature / context questions
- ✅ Resolved — How do classical optimisation NLFFF methods define their residual functionals? See Classical NLFFF extrapolation methods §2 and Classical optimisation and the NF2 loss.
- ✅ Resolved — What metrics are standard for comparing extrapolated fields in Schrijver-style NLFFF comparisons? See NLFFF quality metrics (Schrijver et al. 2008 figures of merit).
- What exactly did Jarolim et al. report for runtime and AR 11158 reproduction quality?
- What does Green et al. say about helicity/evolution in AR 11158 that could motivate the case study?