Ni edge and screw dislocations: Atomsk -> GPU LAMMPS -> OVITO DXA

bash build.sh         four actual Atomsk geometries and checked fixed groups
bash run.sh           eight fresh Slurm minimizations, cross-GPU/FD job, analysis
bash analyze.sh work  analyze your completed native outputs again; no new solver
bash analyze.sh       analyze provided actual outputs; no new solver
bash plot.sh reference explicitly redraw immutable original source plot arrays

The full fresh comparison needs P100 and V10016 queue access. Each allocated
job uses one GPU, two CPU cores, one MPI process and one OMP thread. No direct
SSH to GPU nodes. The TC partition and V10032 nodes are excluded.
Atomsk preparation and OVITO/Python postprocessing run on one CPU thread.

Do not resubmit into existing work. If submission was interrupted, check the
job IDs in work/SUBMISSIONS.json and work/CROSS-SUBMISSION.json and squeue;
interrupting a terminal does not automatically cancel submitted jobs.

These are finite constrained Ni cells under EAM, not DFT core energies or
converged Peierls stress. Poisson0.31 is an input construction approximation.
The initial orientation-cell files are preserved. Only the free XY import-box
margins are enlarged to enclose displaced atoms; no atom or Z period is changed.
The outer5A XY shell is held, while Z is periodic. No finite-temperature MD.
DXA true Burgers vectors use local lattice coordinates; spatial vectors use A.
Native .ca networks, per-particle data and CSV tables are retained.

Compilers: Atomsk0.13.1 GFortran13.3; LAMMPS30Sep2026 GCC13.3,
CUDA12.8/OpenMPI5.0.10/Kokkos5.2.1; OVITO3.16.1 isolated Python environment.
Potential Ni_u3.eam from LAMMPS, Foiles etal PhysRevB33,7983(1986).

reference-networks/ holds immutable original OVITO .ca networks; work/dxa/ holds your newly analyzed networks.
