heat-surrogate

Work with the case23 family of reduced-order surrogate models for 1D copper heat transport, **and the upstream LAMMPS MD pipeline that generates their training data** — adding new methods (FNO, DeepONet, other materials), running comparisons against structural descriptors, training or rolling out the existing JAX models in `heat/methods/0?_*/`, extending `comparison/result_case23.ipynb`, or porting the MD in `heat/md/` (LAMMPS inputs + custom `eam/alloy/mesh + g1mesh + g2mesh` pair_style + slurm drivers) to another metal. Trigger this skill when the user mentions case23, the heat archive, the GENERIC framework with α/β/γ/ICNN, the `heat/md/` LAMMPS workflow, or asks for benchmarks against FNO/DeepONet/order parameters in this context.