VASPBERRY 1.6.3 — simpler band selection and reliable Kubo output

VASPBERRY 1.6.3 — simpler Kubo selection and reliable native outputs

VASPBERRY 1.6.3 selects the Kubo result from the requested bands, detects
byte and legacy four-byte-word WAVECAR record lengths, and verifies complete,
finite curvature output before reporting success.

Run the supplied example

Clone the repository or download the versioned source.
From its root, use the site's Intel oneAPI environment and MPI allocation:

source /opt/intel/oneapi/setvars.sh
make ifx-mpi
repo_dir="$PWD"
vb_bin="$repo_dir/build/vaspberry-ifx-mpi"
mpiexec -n 4 "$vb_bin" --help kubo
mpiexec -n 4 "$vb_bin" --help bands
mkdir results-mos2-kubo
(
  cd results-mos2-kubo
  mpiexec -n 4 "$vb_bin" --task kubo \
    --wavecar "$repo_dir/examples/1H-MoS2/KPATH/2.band/WAVECAR" \
    --bands 1:18
)

The native executable reads the supplied wavefunctions and writes KUBO.csv
with 48 occupied-subspace curvature rows. Its columns are
spin,k_index,kx_frac,ky_frac,kz_frac,omega_z_A2,min_external_gap_eV.
Curvature is in Ų and the external gap is in eV. Import the CSV into your
preferred plotter, skip the # metadata lines, and plot omega_z_A2 against
k_index. Python is optional for this native-result plot. This path input
does not supply a full-zone integral.

Use a fresh result directory for each run. Intel Classic and GNU build
alternatives are in the build guide.

Band selection and migration

Requested quantity	Native Kubo arguments
One isolated band	--bands N
Trace of a multi-band subspace	--bands FIRST:LAST
Separate isolated bands	--bands FIRST:LAST --per-band 1

These rules apply to kubo, kubo-line and kubo-integral. A singleton
such as 18:18 remains one band. The trace defaults to KUBO.csv;
--curvature-csv PATH selects another filename. Individual-band CSV output
is optional, alongside the existing DAT files.

The native --bundle and -kubo_bundle flags now fail with migration guidance.
Each individual band must have a gap greater than 1e-5 eV to every other
stored band. A trace checks only the gap to excluded bands and permits
internal degeneracies. Pure legacy -kubo commands retain individual-band
selection with the same isolation check. Pair and spin-sector tasks retain
their existing selection semantics. Version 1.6.3 is maintainer-selected;
these command changes require migration even though the numerical trace
formula is unchanged.

Input and result checks

Native WAVECAR reading recognizes supported byte and four-byte-word RECL
layouts without rewriting the file. Malformed or ambiguous layouts are
rejected. Keep the supplied compiler flags, including Intel -assume byterecl.

New trace CSV V2 and individual-band CSV V3 keep their numerical columns
and add source dimensions, expected row coverage and explicit completion
status. Files begin INCOMPLETE and receive terminal PASS only after all
finite results are written successfully. A .partial file is promoted to
the final filename only after successful completion, flush and close.
Pair export uses the same staging rule while retaining its existing schema.
Readers reject missing, repeated
or nonfinite rows. Older trace V1 and band V2 inputs remain readable as
legacy_unverified; their saved data and producer metadata are preserved.
The Python bundle-hall command continues to integrate a finished trace CSV.

The technical report (PDF)
describes the methods and public material examples. Existing reference
calculations keep their original provenance; this release makes no new
material-convergence claim. See the native reference,
hands-on guide, example map,
output schemas, migration guide,
validation requirements and changelog.

Published source and checks

Source commit: 1d5d752.

• .github/workflows/ci.yml: 8 jobs passed
• .github/workflows/bi-z2-validation.yml: 1 jobs passed
• .github/workflows/intel-mpi-validation.yml: 2 jobs passed
• .github/workflows/installation-validation.yml: 2 jobs passed

All four workflows checked this exact release commit. Historical material-convergence studies retain their original provenance.