Add Paper Problem 2 OCP benchmark

docs/PAPER_OCP_VALIDATION.md

@@ -23,6 +23,16 @@ units throughout. The Pyomo benchmark itself does not depend on GEKKO or
 MATLAB. GEKKO is available in the repo-local Pixi environment for upstream
 policy-segment verification.
 
+Problem 2 is now available as a second paper-reference OCP:
+
+- primary drying only, using the same moving-boundary model;
+- shelf temperature as the optimized control;
+- objective: minimize drying time;
+- constraints: product temperature at or below 240 K, interface velocity at or
+  below `2.8e-7 m/s`, and shelf temperature between 228 K and 260 K;
+- expected qualitative policy sequence: interface-velocity tracking, maximum
+  heat input, then product-temperature tracking.
+
 ## Validation Strategy
 
 The benchmark has two layers:
@@ -136,18 +146,47 @@ All rows use `nfe=12`, `ncp=3`, `LAGRANGE-RADAU`, the policy initializer, and a
 | `n_z=5` | optimal | ~6.19 h | Policy 1 -> Policy 2 |
 | `n_z=20` | optimal/acceptable | ~6.19 h | Policy 1 -> Policy 2 |
 
+## Problem 2 First-Pass Tolerances
+
+The Problem 2 validation is intentionally coarse at this stage. The paper
+reports switch times near 2.0 h and 3.9 h, with drying complete around 8.9 h.
+The slow test therefore accepts broad first-pass tolerances on the coarse
+`n_z=5`, `nfe=12`, `ncp=3` mesh:
+
+- terminal interface gap at or below `1e-7 m`;
+- product-temperature violation at or below `1e-3 K`;
+- post-initial interface-velocity violation at or below `5e-10 m/s`;
+- shelf-temperature bound violations at or below `1e-6 K`;
+- drying time within `0.7 h` of the paper value;
+- first two policy switches within `0.8 h` of the paper values.
+
+The velocity constraint is skipped at the initial collocation point because the
+paper explicitly reports an initial velocity excursion before Policy 3 quickly
+brings `dS/dt` to its setpoint. Metrics still report the initial velocity
+separately, while path-constraint checks use the post-initial trajectory.
+
+Known limitations:
+
+- The Problem 2 initializer is a deterministic policy-sequenced warm start, not
+  a full reproduction of the upstream high-index GEKKO Policy 3 solve.
+- The first validated solve is coarse. A refined `n_z=20` Problem 2 solve and a
+  MATLAB/GEKKO upstream export should be added once the upstream reference
+  tooling is extended beyond Problem 1.
+
 ## Future Work
 
 Next steps are tracked in GitHub issues:
 
 1. #28 - Prepare the first Paper Problem 1 validation PR.
-2. #29 - Add Problem 2 with the interface-velocity constraint and expected
-   Policy 3 -> Policy 1 -> Policy 2 sequence.
-3. #30 - Compare the paper-reference transcription against LyoPRONTO's existing
+2. #30 - Compare the paper-reference transcription against LyoPRONTO's existing
    quasi-steady Pyomo and scipy optimizers.
-4. #31 - If the benchmark is credible, add a LyoPRONTO-facing experimental
+3. #31 - If the benchmark is credible, add a LyoPRONTO-facing experimental
    policy API with the same rich result format.
 
 #26 is addressed by the bottom-node temperature constraint alignment, the
 smaller expression-based NLP for vapor pressure/resistance/flux/interface
 velocity, constraint scaling, and the `n_z=20` slow validation test.
+
+#29 is addressed by the Problem 2 config defaults, velocity path constraint,
+Policy 3 classifier support, policy-sequenced initializer, coarse slow solve,
+and first-pass tolerance documentation.

lyopronto/pyomo_models/README.md

@@ -88,19 +88,28 @@ drying OCP in Srisuma and Braatz, arXiv:2509.10826v1.
 
 **Key functions:**
 - `create_paper_problem1_model()` - Build the orthogonal-collocation Pyomo model
+- `create_paper_problem2_model()` - Build Problem 2 with the interface-velocity constraint
 - `solve_paper_problem1()` - Solve Paper Problem 1 with IPOPT
+- `solve_paper_problem2()` - Solve Paper Problem 2 with IPOPT
 - `generate_problem1_policy_initialization()` - Build a policy-based warm start
+- `generate_problem2_policy_initialization()` - Build the Problem 2 Policy 3 -> Policy 1 -> Policy 2 warm start
 - `initialize_paper_problem1_from_trajectory()` - Seed a model from a trajectory
 - `load_upstream_matlab_trajectory()` - Read upstream MATLAB output saved to `.mat`
 - `compare_paper_problem1_trajectories()` - Compare Pyomo and upstream metrics
-- `classify_paper_policies()` - Infer active Policy 1/Policy 2 regions
+- `classify_paper_policies()` - Infer active Policy 1/Policy 2/Policy 3 regions
 
 This module uses SI/Kelvin units from the paper/upstream code and is separate
-from LyoPRONTO's cm/Torr/degC production APIs. The validated default solve uses
-a coarse `n_z=5` mesh, and the upstream paper's `n_z=20` spatial mesh is covered
-by a slow validation test using IPOPT acceptable termination
-(`acceptable_tol=1e-3`, `acceptable_iter=5`). Both reproduce the expected
-Policy 1 -> Policy 2 sequence near the paper's reported switch time.
+from LyoPRONTO's cm/Torr/degC production APIs. Problem 1 validates the
+temperature-constrained drying-time objective and reproduces the expected
+Policy 1 -> Policy 2 sequence near the paper's reported switch time. Problem 2
+adds the 240 K product-temperature limit, 260 K shelf-temperature upper bound,
+and `2.8e-7 m/s` interface-velocity path constraint; the coarse validation
+tracks the expected Policy 3 -> Policy 1 -> Policy 2 sequence.
+
+The validated default solves use a coarse `n_z=5` mesh. For Problem 1, the
+upstream paper's `n_z=20` spatial mesh is also covered by a slow validation test
+using IPOPT acceptable termination (`acceptable_tol=1e-3`,
+`acceptable_iter=5`).
 
 Regenerate the full upstream Problem 1 reference with:
 

lyopronto/pyomo_models/__init__.py

@@ -51,10 +51,13 @@ def _is_pyomo_available() -> bool:
         classify_paper_policies,
         compare_paper_problem1_trajectories,
         create_paper_problem1_model,
+        create_paper_problem2_model,
         generate_problem1_policy_initialization,
+        generate_problem2_policy_initialization,
         initialize_paper_problem1_from_trajectory,
         load_upstream_matlab_trajectory,
         solve_paper_problem1,
+        solve_paper_problem2,
     )
     from .single_step import (
         create_single_step_model,
@@ -75,11 +78,14 @@ def _is_pyomo_available() -> bool:
         "PaperPrimaryDryingConfig",
         "PaperDiscretization",
         "create_paper_problem1_model",
+        "create_paper_problem2_model",
         "generate_problem1_policy_initialization",
+        "generate_problem2_policy_initialization",
         "initialize_paper_problem1_from_trajectory",
         "load_upstream_matlab_trajectory",
         "compare_paper_problem1_trajectories",
         "solve_paper_problem1",
+        "solve_paper_problem2",
         "classify_paper_policies",
     ]