Release-time HPC regression suite for the Green ecosystem
(green-mbpt, green-mbtools, …). Complements the per-PR unit tests
in the source repos by running calculations that genuinely need a
supercomputer allocation.
See DESIGN_BRIEF.md for the rationale and design
discussion. See ADMIN.md for the operational guide.
systems/
n2/ molecular, basic correlation sanity
si/ PBC + pseudopotential (reference solid)
ge_sfx2c1e/ all-electron + scalar relativistic (sfx2c1e)
ge_x2c1e/ all-electron + full X2C (spinor path)
h_chain/ 1D periodic correlation
<13 solids>/ GW symmetry-coverage set — see "Solid-state coverage" below
tools/ collection, diff, manifest lint, version comparison
releases/ per-release benchmark write-ups (SUMMARY + generated
RESULTS.md / COMPARISON.md tables; do not hand-edit tables)
Each system has manifest.yaml (the run contract), results/ (one JSON
summary per Green release), and — only when it needs non-default SLURM
resources — an overrides.sh. One driver, tools/submit.sh <system>,
chains the init → mbpt → ac SLURM jobs for any system. Heavy run outputs
live in scratch ($BENCH_SCRATCH, e.g. /pauli-storage) — only the
extracted JSON summaries are committed here.
A set of solids chosen to span distinct space groups, to exercise the k-point
symmetry machinery (symmorphic vs non-symmorphic, centrosymmetric vs polar,
insulators vs metals). Each runs GW on a 3×3×3 k-mesh with
gth-dzvp-molopt-sr / gth-pbe (auxbasis: none, etb_beta: 2.5), PBE
mean-field. Geometries use experimental equilibrium lattice parameters
(space group verified with spglib).
| System | Formula | SG # | Symbol | Crystal system | Atoms/cell | Notes |
|---|---|---|---|---|---|---|
| lih | LiH | 225 | Fm-3m | Cubic | 2 | Rock-salt, symmorphic |
| si | Si | 227 | Fd-3m | Cubic | 2 | Diamond (reference, pre-existing) |
| gaas | GaAs | 216 | F-43m | Cubic | 2 | Zinc-blende, non-centrosymmetric |
| diamond | C | 227 | Fd-3m | Cubic | 2 | Diamond, light atoms |
| c_bn | BN | 216 | F-43m | Cubic | 2 | Zinc-blende, non-centrosymmetric |
| na | Na | 229 | Im-3m | Cubic | 1 | BCC metal, symmorphic |
| hbn | BN | 194 | P6₃/mmc | Hexagonal | 4 | Layered, non-symmorphic, centrosymmetric |
| aln | AlN | 186 | P6₃mc | Hexagonal | 4 | Wurtzite, polar, non-centrosymmetric |
| mg | Mg | 194 | P6₃/mmc | Hexagonal | 2 | HCP metal |
| mgf2 | MgF₂ | 136 | P4₂/mnm | Tetragonal | 6 | Rutile-type, non-symmorphic |
| mgh2 | MgH₂ | 136 | P4₂/mnm | Tetragonal | 6 | Rutile-type, lightest |
| tio2 | TiO₂ | 141 | I4₁/amd | Tetragonal | 12 | Anatase, conventional cell, non-symmorphic |
| black_p | P | 64 | Cmce | Orthorhombic | 8 | Black phosphorus, conventional cell |
| al2o3 | Al₂O₃ | 167 | R-3c | Trigonal | 10 | Corundum, rhombohedral primitive |
Symmetry spread:
- Symmorphic: Na (229), c-BN / GaAs (216), LiH (225); all others non-symmorphic.
- Non-centrosymmetric: c-BN (216), GaAs (216), AlN (186).
- Metals: Na (229), Mg (194).
- 10 distinct space groups: 225, 227, 216, 229, 194, 186, 136, 141, 64, 167.
TiO₂ (anatase) and black P use their conventional cells — PySCF detects the full space group more reliably from conventional than from reduced primitive cells.
Initial round:
| Component | Versions under test |
|---|---|
| green-mbpt | 0.3.2, 1.0.0-alpha |
| green-mbtools | 0.3.0, 1.0.0-alpha |
Results land in systems/<sys>/results/<mbpt-version>_<mbtools-version>.json
and aggregate to releases/v1.0.0/RESULTS.md
(see releases/v1.0.0/SUMMARY.md for the
narrative).
- Add a new release: see
ADMIN.md - Add a new system: see
ADMIN.md - Manifest schema:
tools/verify_manifest.py