ATAT calcbulk + native atat-vasp CPU MPI example: FCC Al, 5 volumes.
Run from this directory as an ordinary account:
  bash run.sh
  bash plot.sh

The submitted Slurm job uses the existing CPU VASP6.4.2 wrapper, two MPI
processes. The tested allocation is the V10016G resource partition, requesting
one GPU under normal permissions/billing; the invoked executable is CPU VASP,
not GPU VASP. This example is an interface workflow, not a GPU speed result.
The excluded TC partition members are queried at every submission and V10032G
is always excluded. Keep the normal Slurm hooks and account permissions.

The ordinary account needs a valid ATAT VASP configuration and permission to
use the installed VASP and Al PBE potential. No POTCAR, VASP binary, OUTCAR,
WAVECAR or vasprun.xml is redistributed. A SHA256 of the expected potential
is provided to keep the five energy points consistent with the example.

Inputs: fixed one-atom FCC primitive cell, a=4.04 Angstrom; strains -0.02,
-0.01,0,0.01,0.02; PBE, ENCUT500eV, EDIFF1e-6, Gamma8x8x8, ISMEAR1,SIGMA0.1.
No ionic/cell relaxation is performed inside a point. Native calcbulk scales
the cell, runs actual VASP, extracts energy, fits its quadratic E(linear strain)
and writes bulkmod (eV/Angstrom^3), strain.out, energy and str_relax.out.

Independent analysis reads full-precision cell vectors from POSCAR rather than
the lower-precision displayed OUTCAR volume. It compares the exact native
quadratic convention and an independent Birch-Murnaghan fit; these are different
fit models and can have slightly different bulk moduli. This five-point scan
is an operational example, not a convergence study or experimental prediction.

Toolchain: ATAT3.52 GCC13.3; VASP6.4.2 CPU wrapper loading Intel oneAPI2024.2/IntelMPI. ASE3.29.0 Python provides independent EOS and plotting.
