PR 4/7: Pyomo multi-period DAE model

[bernalde]

Summary

Full trajectory optimization model. Discretizes the drying process into multiple time periods, encoding the physics (vapor pressure, product resistance, energy balance, mass transfer) as Pyomo constraints. Supports both finite-difference and collocation discretization. Log-transform for vapor pressure to avoid numerical issues with exponentials.

PR 4 of 7 in the Pyomo integration series.

Changes

• lyopronto/pyomo_models/model.py — Full multi-period DAE model
• tests/test_pyomo_models/test_model_multi_period.py — Multi-period tests (741 lines)
• tests/test_pyomo_models/test_model_advanced.py — Advanced model tests
• tests/test_pyomo_models/test_model_validation.py — Validation against scipy
• tests/test_pyomo_models/test_physics_equations.py — Physics equation tests

Key Physics Encoded as Constraints

Psub = 2.698e10 * exp(-6144.96 / (Tsub + 273.15))     # Vapor pressure (log-transformed)
Rp = R0 + A1 * Lck / (1 + A2 * Lck)                   # Product resistance
dmdt = Ap / Rp * (Psub - Pch)                          # Sublimation rate
Kv * Av * (Tsh - Tbot) = dmdt * dHs                    # Energy balance

PR Chain

#	PR	Status
0	Sync upstream (#5)	Merged
1	CI/CD for Pyomo (#6)	Open
2	Pyomo dependencies (#7)	Open
3	Utils & single-step (#8)	Open
4	Multi-period model (this PR)
5	Optimizer functions	Pending
6	Benchmarks	Pending
7	Docs & examples	Pending

Testing

~2075 lines of new tests covering model construction, discretization, physics constraints, and validation against scipy baseline.

[bernalde]

I found several blocking issues in the core model formulation and in the new test coverage. The main concerns are that the product-temperature constraint is applied to the wrong variable/direction, scaled solves return scaled rather than physical units, warmstart initializes dmdt 10x too high, the cake-length ODE does not match the existing scipy drying-rate update, and one new Pyomo test fails when networkx is not installed.

[bernalde]

Review summary:

1. Blocking issues: the default scaled create_multi_period_model() result does not expose the documented unscaled model attributes and cannot be passed to warmstart_from_scipy_trajectory; dmdt[0] is unconstrained but is returned as part of the solution trajectory.
2. Nonblocking issues: none.
3. Questions: none.
4. Tests run and outcomes: PYTHONPATH=/tmp/LyoPRONTO_pr9 pytest tests/test_pyomo_models -m "pyomo and not slow" -n 0 --tb=short --maxfail=10 passed with 67 passed, 9 skipped, 14 deselected; RUN_SLOW_TESTS=1 PYTHONPATH=/tmp/LyoPRONTO_pr9 pytest tests/test_pyomo_models/test_model_multi_period.py::TestModelOptimization::test_optimization_runs tests/test_pyomo_models/test_model_validation.py::TestOptimizationComparison::test_optimization_improves_over_scipy tests/test_pyomo_models/test_model_validation.py::TestOptimizationComparison::test_optimized_solution_satisfies_constraints -n 0 --tb=short --maxfail=3 passed with 3 passed; ruff check lyopronto/pyomo_models/model.py tests/test_pyomo_models/test_model_advanced.py tests/test_pyomo_models/test_model_multi_period.py tests/test_pyomo_models/test_model_validation.py tests/test_pyomo_models/test_physics_equations.py passed; git diff --check origin/pr/pyomo-utils-singlestep...HEAD passed.
5. Merge as-is: no. I would not merge this until the blocking issues above are addressed.

[bernalde]

Propagating review feedback from the lower Pyomo PRs where the same issue still applies.

[bernalde]

Review summary:

1. Blocking issues: final_dryness stops the optimizer at 95% dried while optimize_multi_period() reports t_final as total drying time.
2. Nonblocking issues: none.
3. Questions: none.
4. Tests run and outcomes: targeted non-slow Pyomo suite passed; targeted slow multi-period IPOPT tests passed; CI-style Pyomo marker suite passed; ruff passed; git diff --check passed. The broader non-Pyomo CI command was interrupted after reaching 99% with one xdist worker still running for several minutes and no failures observed before interruption.
5. Merge as-is: no. I would not merge this until the blocking issues above are addressed.

GitHub rejected REQUEST_CHANGES because the authenticated account owns this PR, so this review is submitted as COMMENT.

[bernalde]

Blocking: This terminal constraint lets the optimizer report t_final when the cake is only 95% dry. Since the objective minimizes t_final, the optimum stops at this threshold; with the standard warmstart I get Lck[-1] / Lpr0 = 0.94999999. That underestimates primary drying time compared with the existing calc_knownRp.dry completion event, which targets Lpr0 - L == 0 when the schedule permits. Please either enforce full completion, or add an explicit documented target_dry_fraction parameter that defaults to complete drying, and add a regression that checks the returned final Lck/percent dried reaches that target.

[bernalde]

Addressed in fe5fc82. target_dry_fraction now defaults to 1.0, final_dryness enforces that target, and the optimized-solution validation now checks complete drying by default. Added tests for default, custom, and invalid terminal targets.

[bernalde]

Commits pushed:

• fe5fc82 Require complete drying in multi-period model

Main changes made:

• Added target_dry_fraction to create_multi_period_model() and optimize_multi_period(), defaulting to 1.0 for complete drying.
• Changed final_dryness to enforce target_dry_fraction * Lpr0 instead of a hardcoded 95% threshold.
• Added regression coverage for the default complete-drying target, custom target values, invalid target values, and updated optimized-solution validation.

Tests run and results:

• PYTHONPATH=/tmp/lyopronto_pr9_review MPLCONFIGDIR=/tmp/mplconfig pytest tests/test_pyomo_models/test_model_multi_period.py::TestModelStructure::test_final_dryness_defaults_to_complete_drying tests/test_pyomo_models/test_model_multi_period.py::TestModelStructure::test_final_dryness_target_can_be_configured tests/test_pyomo_models/test_model_multi_period.py::TestModelStructure::test_final_dryness_target_validates_fraction -n 0 --tb=short: 5 passed.
• RUN_SLOW_TESTS=1 PYTHONPATH=/tmp/lyopronto_pr9_review MPLCONFIGDIR=/tmp/mplconfig pytest tests/test_pyomo_models/test_model_validation.py::TestOptimizationComparison::test_optimized_solution_satisfies_constraints -n 0 --tb=short --maxfail=1: 1 passed.
• PYTHONPATH=/tmp/lyopronto_pr9_review MPLCONFIGDIR=/tmp/mplconfig pytest tests/test_pyomo_models -m "pyomo and not slow" -n 0 --tb=short --maxfail=10: 82 passed, 1 skipped, 14 deselected, 1 xfailed.
• RUN_SLOW_TESTS=1 PYTHONPATH=/tmp/lyopronto_pr9_review MPLCONFIGDIR=/tmp/mplconfig pytest tests/test_pyomo_models/test_model_multi_period.py::TestModelOptimization::test_optimization_runs tests/test_pyomo_models/test_model_validation.py::TestOptimizationComparison::test_optimization_improves_over_scipy tests/test_pyomo_models/test_model_validation.py::TestOptimizationComparison::test_optimized_solution_satisfies_constraints -n 0 --tb=short --maxfail=3: 3 passed.
• python -m ruff check lyopronto/pyomo_models/model.py tests/test_pyomo_models/test_model_multi_period.py tests/test_pyomo_models/test_model_validation.py: passed.
• git diff --check: passed.

Comments intentionally not addressed:

• Previously resolved or outdated review threads were left unchanged; the current code already contains their fixes.
• No top-level discussion comments required separate code changes.

Remaining risks or follow-up items:

• The broader non-Pyomo CI command was not rerun after this isolated Pyomo change. During review it reached 99% and then one xdist worker remained running for several minutes, so I kept validation focused on the modified Pyomo model and tests.