v5.0.0a2

What's new in v5.0.0a2

Highlights

• q2mm/elements.py — Single source of truth for periodic table data (masses, covalent radii, symbols) with cited provenance (NUBASE2020, Cordero 2008). All consumers import from this one module.
• Diagnostics & benchmark CLI — q2mm-benchmark entry point with streaming tables, quality-aware status, per-combo summaries, ANSI colors, and --data-dir for wheel installs.
• JAX differentiable backend — Analytical gradients via JAX-MD for OPLSAA force fields.
• OpenMM XML export — Export optimized force fields to OpenMM XML format.
• Method D/E eigenvalue fitting — Seminario Method E hybrid pipeline (Lime & Norrby 2015).

DRY & maintainability (since a1)

• Consolidated element data eliminates 4 duplicate definitions across modules
• Generic _build_param_maps() replaces 5 near-identical functions in forcefield.py
• Generic _match_mode() replaces 4 near-identical functions in seminario.py
• Shared build_leaderboard_rows() used by both CLI and report modules
• test/_shared.py consolidates 16 path constants and 3 molecule factories from 13 test files
• Deleted dead _patterns.py module and duplicate MacroModel constants
• Fixed from_molecules() feature drift (silently lacked eigenmatrix params)
• Fixed 0.0 score falsiness bug in leaderboard rows

Packaging & robustness

• [all] extras now include jax/jaxlib; [dev] includes openmm
• Lazy scipy import — module loads without scipy installed
• Configurable Tinker bond_tolerance (was hardcoded 1.3)

CI/CD

• Pre-built container images for backend tests (OpenMM, Tinker, JAX, Psi4)
• Manual Build CI Images workflow for dependency updates
• Backend-specific CI jobs with conda environments

Full changelog

v5.0.0a1...v5.0.0a2

What's Changed

• docs: update installation instructions for PyPI by @ericchansen in #72
• refactor: Clean up package architecture, merge Param classes, fix critical bugs by @ericchansen in #79
• feat(parsers): Add eq and hash to Param by @ericchansen in #80
• test: Add end-to-end Seminario validation for ethane by @ericchansen in #81
• test: Add parity stability tests and Seminario benchmark by @ericchansen in #82
• feat(hessian): Implement Method D/E eigenvalue fitting by @ericchansen in #83
• fix(hessian): Address code review findings from PRs #81-83 by @ericchansen in #84
• feat(seminario): Add Method E hybrid pipeline (Limé & Norrby 2015) by @ericchansen in #85
• feat(hessian): Add eigenmatrix training data pipeline by @ericchansen in #86
• docs: Add ReferenceData factory methods and end-to-end README example by @ericchansen in #87
• feat(export): Add OpenMM XML export for force fields by @ericchansen in #90
• feat(backends): Add JAX differentiable MM backend with analytical gradients by @ericchansen in #92
• feat: Diagnostics package and backend-optimizer benchmark matrix by @ericchansen in #103
• ci: switch backend tests to pre-built container images by @ericchansen in #105
• fix: Address code review findings (bugs, DRY, lint, docs) by @ericchansen in #106
• refactor: Consolidate element data, DRY up codebase, fix packaging by @ericchansen in #107

Full Changelog: v5.0.0a1...v5.0.0a2

v5.0.0a1

What's Changed

• Syncing up ericchansen/q2mm and q2mm/q2mm by @ericchansen in #26
• Fixed bug associated with torsion check error by @arosale4 in #27
• Sync with Q2MM/q2mm by @ericchansen in #29
• Many changes to help with issue #22 by @ericchansen in #37
• Iss22 by @ericchansen in #38
• Added more ligands by @jesswahlers in #45
• Update Param Tethering by @arosale4 in #43
• Create README.md by @egulotty in #46
• Merge Tony's changes by @ericchansen in #47
• Pull updates from q2mm/q2mm by @ericchansen in #48
• refactor: Modernize packaging, fix imports, add CI by @ericchansen in #50
• test: fix broken tests and add parser tests (1 to 28 passing) by @ericchansen in #51
• refactor: restructure into modular subpackages (v5 architecture) by @ericchansen in #52
• feat: add Psi4 and Tinker backend engines with SN2 test data by @ericchansen in #53
• feat: add clean data models and Seminario/QFUERZA pipeline by @ericchansen in #54
• feat: Modernize Q2MM with clean models, OpenMM backend, and Seminario parity by @ericchansen in #55
• feat(optimizers): Add scipy-based FF optimizer by @ericchansen in #56
• feat(validation): Cross-backend optimization validation with OpenMM + Tinker by @ericchansen in #57
• refactor(parsers): Split monolithic files into focused modules by @ericchansen in #58
• chore: delete legacy plumbing files (-9,254 lines) by @ericchansen in #59
• fix: Address multi-model architecture review findings by @ericchansen in #60
• fix: Address multi-model architecture review findings by @ericchansen in #61
• docs: Add MkDocs Material documentation site by @ericchansen in #62
• docs: Improve tutorial, fix benchmarks, add molecule constructors by @ericchansen in #66
• chore: Python 3.10+, remove legacy code, clean repo by @ericchansen in #67
• chore: remove legacy Schrodinger screening code by @ericchansen in #70
• ci: add PyPI publishing on tag push by @ericchansen in #71

New Contributors

• @jesswahlers made their first contribution in #45
• @egulotty made their first contribution in #46

Full Changelog: https://github.com/ericchansen/q2mm/commits/v5.0.0a1