v1.0.0

v1.0.0 — Space Symmetries in MBPT and SEET

Summary

This is the first stable release of the 1.0 line and the culmination of the v1.0.0-alpha series. The headline change is full symmetry adaptation across the mean-field and post-HF pipeline: space-group representations are now propagated through the k-mesh on the scalar side, and the relativistic (X2C/x2c1e) path carries the matching double-group SU(2) lift so spinor quantities transform correctly across the irreducible Brillouin zone.

The rewritten k-point symmetry engine is not just more correct, it is substantially faster on CPU (up to ~4× on 4x4x4 mesh, with even higher speed-up for denser meshes), and it removes a symmetry-contamination failure mode that could make self-consistent GW diverge. Two long-standing correctness issues (the X2C CPU kernel and black-phosphorus symmetry breaking) are fixed and now verified against a public benchmark suite (see Validation & benchmarks).

Highlights

• Symmetry adaptation. Space-group representations on the scalar path and the double-group SU(2) lift on the X2C/x2c1e path, applied consistently in the HF, GW, and GF2 kernels.
• Faster CPU k-point handling. The rewritten symmetry code cuts CPU wall-clock by ~1.4–4.2× across the benchmark solids, with the gain growing with the number of k-points in the star (Silicon 4×4×4: 4.18× in GW total, 4.25× in GF2 total, at identical energies). Single-k molecular cases show no speedup, as expected.
• X2C CPU kernel fixed. The all-electron full-X2C (spinor) relativistic path was broken on the v0.3.2 CPU kernel; the GPU kernel was already correct. v1.0.0 fixes the CPU path so it matches GPU.
• SEET preprocessing rewrite. kstruct-driven orthogonalization with full Löwdin / symmetric-Löwdin / MO / natural-orbital coverage (Löwdin variants supported for the X2C spinor path), backed by an HF/GW/GF2 orthogonalization basis-invariance regression test.
• Release provenance. All executables now report the build version via --version, driven by GREEN_RELEASE.txt.

Validation & benchmarks

v1.0.0 was validated against the previous v0.3.2 line on the public green-benchmarks HPC regression suite, on both CPU and GPU. Full write-up and numbers:

• SUMMARY.md — hand-written overview.
• RESULTS.md — per-system observables, all versions.
• COMPARISON.md — pairwise differences and timing ratios.

Three findings:

1. Same physics, faster. Total energies (ehf, e1b, ecorr) from v0.3.2 and v1.0.0 agree to the GW/SCF convergence threshold (≲1e-7 Ha) on both CPU and GPU — while v1.0.0's CPU path runs up to ~4× faster on the 4x4x4 mesh. v1.0.0 reproduces v0.3.2 wherever v0.3.2 was correct, so the two fixes below are genuine, not numerical drift.

2. Germanium — X2C CPU kernel bug, fixed. On v0.3.2 the CPU X2C (spinor) kernel produced non-physical energies (gw/ehf off by ~215 Ha, no band structure); the GPU kernel was already correct. v1.0.0's CPU result now matches GPU. (Scalar-relativistic sfx2c1e was not affected.)

3. Black phosphorus — symmetry contamination resolved. In a self-consistent GW run on black-phosphorous (Cmce, #64; orthorhombic, non-symmorphic), v0.3.2's Green's function accumulates growing symmetry-breaking error and the total energy diverges (~-52.5 Ha → +20 Ha around iteration 12–14). v1.0.0 stays flat and converges monotonically to ~2e-6 Ha. This is the clearest single demonstration of the release's value.

   

Bug fixes

• Count the Madelung/Ewald correction once per k-point in the X2C HF kernel.
• Distribute the GW self-energy tau over the solver object's context.
• Optimize X2C HF exchange and fix an MPI_Bcast namespace issue; broadcast the self-energy result to all ranks after population on rank 0.
• Inchworm impurity solver: don't double-count μ when extracting the self-energy.
• Reset timing statistics after each kernel/solver run so per-iteration timings no longer accumulate (HF, GW, GF2 CPU paths).

What's Changed

• add local integral decomposition by @sdirnboeck in #19
• Seet by @sdirnboeck in #20
• double counting using mbpt by @sdirnboeck in #24
• fix parameter names, remove unused. params by @gauravharsha in #26
• Rebase SEET onto main, and fix critical bug related to number of orbitals in impurity by @gauravharsha in #30
• add extra include for safety by @gauravharsha in #31
• Add k-point shape validation for X_inv_k after IBZ filtering by @yangyu-code with @Copilot in #35
• Fix 4 fatal bugs in inchworm post-processing self-energy update by @yangyu-code in #34
• update energy in seet inchworm loop by @gauravharsha in #36
• add automatic tag creation workflow by @gauravharsha in #41
• Manage version number / branch using GREEN_RELEASE.txt by @gauravharsha in #42
• add owner info to Github App Token by @gauravharsha in #43
• fix gh api command in github green_tag.yaml by @gauravharsha in #44
• Space group symmetries by @gauravharsha in #45
• Seet main merge by @egull in #46
• reducint dependencies of mbpt by @egull in #47
• Fix/context macro conflict by @gauravharsha in #48
• Remove redundant mpi_context parameter from HF solver interface by @gauravharsha in #49
• Apply space-group symmetry transform in X2C GW and HF kernels by @gauravharsha in #50
• Fix C++17 warning: copy structured binding before lambda capture by @egull in #51
• Update GitHub Actions to Node.js 24-compatible versions by @egull in #52
• fix(inchworm): don't double-count mu when extracting self-energy by @gauravharsha in #54
• feat: SEET kstruct migration + orth basis-invariance regression test by @gauravharsha in #53
• fix: reset statistics timings after GW and HF kernels by @gauravharsha in #55
• feat: report GREEN_RELEASE via --version in all executables by @gauravharsha in #56
• Fix MPI-context correctness in the X2C HF and GW CPU kernels by @gauravharsha in #66

New Contributors

• @yangyu-code with @Copilot made their first contribution in #35
• @yangyu-code made their first contribution in #34
• @egull made their first contribution in #46
• @agnieszkajazdzewska performed the end-to-end benchmarking of the code, as reported in https://github.com/Green-Phys/green-benchmarks

Full Changelog: v0.3.1...v1.0.0