Fix kinetics custom params and equilibrium indexing

[parkyr]

Summary

• Fix RxnKinetics.set_params so user-defined kinetic_model instances require, store, flatten, and restore custom params_f correctly.
• Fix vectorized reversible equilibrium rates so array-temperature reverse terms divide by per-time/per-reaction Keq values.
• Add regression coverage for the issue High: [Kinetics] Custom rate-law setup and vectorized equilibrium reverse term crash #71 custom-model and vectorized-equilibrium failures.

Closes #71.

Triage / Acceptance Criteria

Uses the maintainer triage artifact at #71 (comment).

Covered:

• Custom RxnKinetics(..., kinetic_model=fn, params_f=arr) constructs and stores the f-parameters.
• Missing custom params_f still raises the intended RuntimeError.
• Flat parameter-vector updates preserve custom params_f shape and update values.
• Vectorized array-temperature equilibrium_model matches scalar per-time evaluation.
• 2-D deltah_rxn from heat-of-reaction style inputs no longer goes through np.outer flattening.

Related Issues

• Medium: [Kinetics] Equilibrium reverse term uses product stoichiometry as reaction order #36, High: [Kinetics] Reversible-reaction analytical Jacobian ignores the equilibrium reverse term #23, and Low: [Kinetics] Reaction-order matrix truncates fractional custom orders #44 are adjacent kinetics defects called out by the issue, but are not duplicates of this fix and are intentionally left to their own scopes.

Tests

• Red first: .venv/bin/python -m pytest tests/test_kinetics_custom_and_equilibrium.py -v failed on unchanged base with the recorded custom-model UnboundLocalError and vectorized-equilibrium mismatch.
• Green after fix: .venv/bin/python -m pytest tests/test_kinetics_custom_and_equilibrium.py -v
• Green after fix: .venv/bin/python -m pytest tests/ -m "not assimulo" (38 passed, 8 deselected)
• Green after fix: .venv/bin/python -m pytest --collect-only (46 tests collected)
• Green after fix: git -c core.whitespace=cr-at-eol diff --check

Local environment: repository .venv with Python 3.11.15. The changed kinetics code is pure NumPy and does not require the assimulo backend.

Branch Hygiene

• Base branch: master
• Source branch: fix/issue-71-kinetics-custom-equilibrium
• Branch point: origin/master at 1a3ab3155216f07e9a18c60687a7b6b1c5959202
• upstream/master was fetched at 400e1e13ad4ecd3ba202d5fae8c70ec73bdf9c3e and is already an ancestor of origin/master.
• Open PR overlap check found no same-file overlap for PharmaPy/Kinetics.py or tests/test_kinetics_custom_and_equilibrium.py.
• No prerequisite or stacked PR.

[parkyr]

Maintainer review — PR #111 (fix for #71)

Reviewed the diff against origin/master merge-base. The PR resolves both defects from the #71 triage artifact: the custom-kinetic_model construction crash in set_params and the array-temp equilibrium reverse-term indexing. The fix is broader than the minimal patch (handles the parameter-estimation round-trip and 2-D deltah_rxn), and the changes are internally consistent.

What I verified

• Correctness of the fix path. set_params dict-branch else now assigns params_f/records params_f_shape for elem_flag=False; the flat-vector else restores it via reshape(self.params_f_shape); concat_params ravels custom params_f. The round-trip is coherent and the elem_flag=True path is untouched.
• equil_term guards 2-D deltah_temp with inv_temp[:, None] * deltah/gas_ct instead of np.outer (which flattened), and equilibrium_model now selects keq_temp[:, ind] when keq_temp is 2-D (array temp) while keeping keq_temp[ind] for the scalar-temp/1-D case. Both indexing cases are correct.
• Tests (local, repo .venv, Python 3.11.15):
  • pytest tests/test_kinetics_custom_and_equilibrium.py -v → 5 passed. The two equilibrium tests use the trusted scalar/1-D path as an implementation-independent oracle; the flat-update test genuinely exercises the concat_params↔set_params round-trip.
  • pytest tests/ -m "not assimulo" → 38 passed, 8 deselected.
  • pytest tests/ -m assimulo → 8 passed (reactor/flowsheet integration; no regression in the touched kinetics paths).
  • git -c core.whitespace=cr-at-eol diff --check → clean.
• GitHub checks (head 1066d67): Core tests SUCCESS, Assimulo integration tests SUCCESS.
• Closes #71 is the correct link — the PR fully resolves both defects in that issue.

Findings

Nonblocking — Custom-model coverage stops at construction and parameter plumbing; no test drives a user kinetic_model through an actual rate evaluation (e.g. in a reactor integration) to confirm the restored params_f produces correct rates end to end.

• Why it matters: the regression suite proves the feature no longer crashes and that parameters round-trip, but not that the custom rate law is wired correctly downstream.
• Concrete step: this is appropriately scoped for the #71 regression fix (an end-to-end custom-model reactor run needs the assimulo backend and is broader than the defect). Fine to defer; worth a follow-up issue if custom rate laws become a supported/documented path.

Summary

• Blocking: none.
• Nonblocking: 1 (custom-model end-to-end coverage — deferrable).
• Questions: none.
• Tests run: regression (5), non-assimulo suite (38), assimulo integration (8), whitespace check — all green.
• Merge-readiness: mergeStateStatus: CLEAN, PR-head checks green. No blocking findings.

This account authored the PR, so GitHub does not accept a self-APPROVE; posting as COMMENT. reviewDecision is currently empty (no reported approval gate) — if branch protection later requires an approval, an eligible maintainer will need to provide it.