Skip to content

1.6

Latest
Latest

Choose a tag to compare

View all tags
@mesonepigreco mesonepigreco released this 25 Mar 18:19
· 1 commit to master since this release
1.6
684036b

Release Notes for CellConstructor 1.6.0

Overview

CellConstructor 1.6.0 is a significant performance-focused release that delivers substantial speedups for phonon calculations, particularly for medium-sized systems (50-200 atoms). This release includes critical bug fixes for modern Fortran compilers and several new features.

Highlights

🚀 Performance Improvements

  • SymmetrizeFCQ Optimization: Vectorized coordinate conversions and q-star copying operations, resulting in up to 25% speedup for symmetry operations on typical systems.

    • Cartesian-crystal coordinate conversions are now 6× faster through vectorized np.einsum operations
    • Q-star block copying optimized with NumPy transpose and reshape operations
    • Added cached transformation matrices to avoid redundant computations

  • DiagonalizeSupercell Optimization: Major performance boost through vectorized operations and pre-computed phase factors, significantly reducing computation time for large supercell diagonalizations.

  • GenerateSupercell Optimization: New O(N) set_tau_fast Fortran routine replaces the previous O(N²) algorithm for supercell generation, providing substantial speedups for large systems.

🐛 Bug Fixes

  • Fortran Compiler Compatibility: Fixed acoustic sum rule errors that occurred with newer Fortran compilers (gfortran 15+), ensuring compatibility with modern HPC environments.
  • ForceTensor Error Messages: Improved error handling and messages for incorrect file formats in both second-order and third-order force tensors.
  • DiagonalizeSupercell: Fixed the lo_to_split argument handling and a missing timer variable that caused NameError exceptions.
  • Spglib Integration: Removed the restrictive spglib<=2.2 requirement and fixed symmetrization bugs for better compatibility with newer spglib versions.

✨ New Features

  • get_primitive_cell() Method: New method in the Structure class to easily obtain the primitive cell from any crystal structure, simplifying workflow for high-throughput calculations.

🔧 Improvements

  • Timer Support: Added optional timer parameter to SymmetrizeFCQ, SetupQPoint, SymmetrizeDynQ, ApplyQStar, and ImposeSumRule functions for detailed performance profiling.
  • Fortran Module Updates: Enhanced symdynph_gq_new.f90 to properly enforce symmetries after small-group symmetrization.

Detailed Changes

Commits since v1.5

  1. 684036b - Optimize SymmetrizeFCQ with vectorized coordinate conversions and q-star copying
  2. d7db321 - Add get_primitive_cell() method to Structure class
  3. e5f5fff - Fix NameError: add missing t_main_loop_start timer in DiagonalizeSupercell_slow
  4. b462c05 - Optimize generate_supercell with O(N) set_tau_fast Fortran routine
  5. c49db94 - Optimize DiagonalizeSupercell with vectorized operations and pre-computed phase factors
  6. 8da226c - Fixed a typo
  7. 697666c - Fixed an error in the lo_to_split argumento of DiagonalizeSupercell
  8. 9415f83 - Fixed the error message when the fileformat is wrong in the ForceTensor
  9. a7c9f46 - Fixed a missing error message in the file_format
  10. 502225d - Fix another fortran error in third order
  11. 6be1e8e - Merge pull request #118 from SSCHAcode/fix_fortran_compiler
  12. 4228fd4 - New version. Fixed the error with older fortran compilers
  13. dd6fd10 - Fixed the error of the acoustic sum rule error for using a new fortran compiler
  14. 4a3bb93 - Fixed another spglib test
  15. 6e014e7 - Removed the requirement for spglib<=2.2
  16. d4a100a - Merge pull request #117 from SSCHAcode/spglib_fix
  17. 62649f2 - Fix a spelling
  18. cc4fe44 - Fixed a bug
  19. 49f36b6 - Fixed the spglib symmetrization
  20. 4045be7 - Fixing the interpolate mesh subroutine

Performance Benchmarks

Small System (CsSnI₃, 5 atoms, 10 irreducible q-points)

  • Total Symmetrize time: 0.249s → 0.186s (25% speedup)
  • Coordinate conversions: 0.06s → 0.01s (6× speedup)

Medium System (CsSnI₃ rhombohedral, 10 atoms, 18 irreducible q-points)

  • Total Symmetrize time: ~0.092s with negligible coordinate conversion overhead

Note: Larger systems with bigger q-point stars will see even more significant improvements from the vectorized operations.

Compatibility

  • Python: 3.8+
  • NumPy: 1.20+
  • ASE: Compatible with latest versions
  • Fortran Compiler: gfortran 15+ and Intel Fortran
  • Spglib: Now compatible with all recent versions (removed <=2.2 restriction)

Upgrade Notes

This release is fully backward compatible. No API changes were made - all optimizations maintain the existing function signatures and behavior. Users can upgrade seamlessly and immediately benefit from the performance improvements.

Contributors

Thanks to all contributors who made this release possible, particularly for the performance optimizations and compiler compatibility fixes.

Full Changelog: 1.5...1.6.0

Assets 2
Loading