You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Agents: read the guide for agents in #310 first. Specification: this description, #306 and its comments.
Sub-issue of #306, part 5 of 5. Independent, low priority.
Scope
From the table of #306: :tags (controlled vocabulary defined in src and checked: e.g. :hard, :analytic, :large_state), :references (bibliographic sources), :corleone (optional pointer to a related problem of SciML's OptimalControlBenchmarks), :reference_solution (optional analytic x(t), u(t)). Each in METADATA_SCHEMA with its type, checked by the consistency tests.
Done when
Fields in the schema, typed and tested; filled where known.
:reference_solution, optional, in ext/MetaData/<p>.jl: a function of the parameters giving the analytic solution of the continuous problem, (state, control, objective, source), or nothing where it is not known. Read with OptimalControlProblems.reference_solution_data(problem, parameters).
Filled for beam and bryson_denham: the Bryson–Denham solution with a boundary arc (ℓ ≤ 1/6), J = 8/(9ℓ) and 4/(9ℓ) (Bryson, Denham & Dreyfus 1963). The formulas live once, in src/solutions.jl.
Checked against the OptimalControl model: at the analytic point, the constraint violation and the objective gap of the trapezoidal discretization vanish at order 2. The test fails on a wrong objective.
Every description cites its sources: a test checks for a References section with at least one entry. The sources stay in the descriptions, so there is no :references field.
What was dropped
:tags. :hard is a judgment. :large_state and :analytic can be computed, from features and from :reference_solution.
For later. A description of the structure of a problem, made of characteristics rather than judgments: singular arcs, boundary arcs, stiffness, turnpike. It depends on the parameters, as :expected_structure did in #306.
Activity
Merge pull request #403 from control-toolbox/feat/379-reference-solution
ocots commentedon Nov 7, 2026
Done in #403.
What was done
:reference_solution, optional, inext/MetaData/<p>.jl: a function of the parameters giving the analytic solution of the continuous problem,(state, control, objective, source), ornothingwhere it is not known. Read withOptimalControlProblems.reference_solution_data(problem, parameters).beamandbryson_denham: the Bryson–Denham solution with a boundary arc (ℓ ≤ 1/6), J = 8/(9ℓ) and 4/(9ℓ) (Bryson, Denham & Dreyfus 1963). The formulas live once, insrc/solutions.jl.Referencessection with at least one entry. The sources stay in the descriptions, so there is no:referencesfield.What was dropped
:tags.:hardis a judgment.:large_stateand:analyticcan be computed, fromfeaturesand from:reference_solution.:corleone. We write our own problems ([Problems] Proof of concept: a ModelingToolkit modeling, solved with Corleone, with our own bryson_denham #308). A pointer to a SciML problem that differs from ours, and that nothing checks, would suggest an equivalence that does not hold.For later. A description of the structure of a problem, made of characteristics rather than judgments: singular arcs, boundary arcs, stiffness, turnpike. It depends on the parameters, as
:expected_structuredid in #306.Checked after #406:
mainCI green at c3ade68.