joss: address reviewer Tangi

[jbcaillau]

Context

Addresses the JOSS review comments of @tmigot, connected to openjournals/joss-reviews#10630. This PR collects all of the reviewer's issues; it is kept as a draft until they are all covered.

Issue	Topic	Status
#802	Minor typos / wording suggestions	✅ done
#803	Paper review (citations, problem class)	✅ done
#805	Repository-level comments	✅ done

All changes so far are in joss/ (paper.md, paper.bib). joss/paper.pdf is rebuilt and auto-committed by the Compile JOSS paper workflow (.github/workflows/JOSS.yml) on push.

The reviewer's line numbers refer to the rendered PDF; the mapping to joss/paper.md source lines is given below where relevant.

---

Issue #802 — minor typos / wording

All 12 suggestions applied verbatim; editorial only, no technical content changed.

#	PDF line	paper.md	Change
1	52	57	In contrast with InfiniteOpt.jl that is designed → In contrast to InfiniteOpt.jl, which is designed
2	54	57	Besides, contributing to… → Moreover, contributing to…
3	57	57	Pontryagin's Maximum Principle and that heavily rely → Pontryagin's Maximum Principle, which heavily rely
4	74	67	Direct method tools → Direct-method tools
5	75	67	Strong direct method implementations: → Strong direct-method implementations:
6	87	69	PDE's → PDEs
7	121	85	arbitrary user defined functions → arbitrary user-defined functions
8	146	89	Non-regression on complex problems is increasingly driven by OptimalControlProblems.jl, used as a curated problem bank. → Regression testing on complex optimal control instances is increasingly supported by OptimalControlProblems.jl, which serves as a curated problem bank.
9	162	97	Both approaches resort to iterative solvers → Both approaches rely on iterative solvers
10	163	97	may converge or not, e.g. depending on… → may or may not converge, e.g., depending on…
11	181	107	The package serves educational purposes through Tutorials.jl covering topics… → The package also serves educational purposes through Tutorials.jl, which covers topics…
12	190	109	across the control-toolbox ecosystem packages → across the control-toolbox ecosystem

Existing markdown links (items 8 and 11) and the *e.g.* italics (item 10) were preserved.

---

Issue #803 — paper review

1. Uniform styling / linking of .jl packages

A single explicit convention is now applied throughout the paper:

every package or tool is hyperlinked at each of its occurrences, and carries its bibliography tag [@key] only at the first occurrence.

• Missing hyperlinks added: ForwardDiff.jl, DifferentiationInterface.jl, NonlinearSolve.jl, NLPModelsIpopt.jl, Knitro, InfiniteOpt.jl, JuMP, ADNLPModels, ExaModels, MadNLP, CUDSS, and every package in the Julia packages bullet of State of the Field.
• Duplicated tags removed (kept at first occurrence only): @Patterson2014, @Verschueren2021, @Andersson2019, @arnstrom2022daqp, @Gagnon_*, @pulsipher2022unifying, @QuantumControl_jl, @shin2023accelerating, @SHIN2024110651, @SHIN2021693, @ADNLPModels_jl, @OptimalControlProblems_jl, @Wachter2006, @rackauckas2017differentialequations.
• QuantumControl.jl is now cited: the QuantumControl_jl entry was present in paper.bib but unused.

The only names left unlinked are those inside the bold bullet labels of State of the Field (**Legacy tools (COTCOT, HamPath, NutoPy)**, **Direct-method tools (BOCOP, ACADO, GPOPS-II, acados, nosnoc)**), which are enumerations of what the bullet is about; each is hyperlinked in the body of its bullet.

2. MINPACK citation

Added @More1980 to paper.bib — Moré, Garbow & Hillstrom, User Guide for MINPACK-1, ANL-80-74, Argonne National Laboratory, 1980, doi:10.2172/6997568 — and cited it at the MINPACK.jl mention in the Indirect methods paragraph.

3. Companion example citation / version

The reviewer is right that v1.0.0 was stale (v1.0.1 was released on 2026-06-11). Rather than bumping to the version DOI, the entry now uses the Zenodo all-versions (concept) DOI 10.5281/zenodo.19663649, which always resolves to the latest release and cannot go stale again. The now-misleading version field was dropped.

4. Ipopt citation in Modelers and solvers

IPOPT is now hyperlinked in that paragraph. Its @Wachter2006 tag stays at the first occurrence (Summary), per the convention in §1 above — repeating the tag would break the rule applied to every other tool.

5. Supported problem class

New Problem Class section, placed right after the Summary, giving the Bolza formulation and stating the scope explicitly. Every capability listed was verified against docs/src/manual-abstract.md:

• fixed and free final time — the time bounds t₀, t_f may be fixed or be components of the optimisation variable v, covering minimum time problems;
• endpoint constraints — boundary constraints;
• path constraints — state, control and mixed (control and state);
• box constraints on state, control and variable;
• Mayer, Lagrange and Bolza costs — Bolza is the general form, Mayer (f⁰ = 0) and Lagrange (g = 0) the special cases; maximisation via minimisation of the opposite cost;
• autonomous and non-autonomous dynamics — dependences detected by the parser;
• multiphase — not supported;
• DAEs — not supported, explicit ODEs only;
• hybrid / nonsmooth systems — not supported.

The section also records the finite-dimensional optimisation variable v (a parameter optimised together with the trajectory), which has no counterpart in the reviewer's list but is part of the supported class.

---

Issue #805 — repository-level comments

#	Item	Status	Notes
1	.zenodo.json DOI mismatch with CITATION.bib/CITATION.cff	✅ done	.zenodo.json now uses 10.5281/zenodo.13336563, the Zenodo concept DOI (verified via the Zenodo API: 13336652 was a stale version-specific DOI for v0.11.2).
2	Spurious deps in Project.toml (Documenter, ADNLPModels, NLPModels, SolverCore)	✅ done	Documenter and NLPModels removed entirely ([deps]+[compat]) — confirmed unused anywhere in src/ or test/. ADNLPModels and SolverCore are unused in src/ but are used by the test suite (SolverCore for an AbstractExecutionStats signature check, ADNLPModels — which will also trigger the CTSolversADNLPModels extension once CTSolvers ships it), so both were moved to [extras]/[targets] instead of dropped; ExaModels, also flagged for review, was checked and kept as-is — it's genuinely imported in src/imports/examodels.jl.
3	README.md basic usage: using NLPModelsIpopt unexplained	✅ done	Added inline comments on using NLPModelsIpopt and using Plots (same issue applies there) in README.md and docs/src/index.md, explaining they activate package extensions. docs/src/manual-solve.md already documented this in depth, so it was left untouched.
4	CITATION.cff missing version/date-released	⬜ not fixed (deliberate)	These fields would go stale at every release without an automated bump, which this repo doesn't have yet. Left as-is rather than adding fields that immediately drift.
5	test/README.md Unix-only / shell-heavy commands	✅ done	File removed — it was an internal doc, not part of the public API surface.
6	Tests don't run on Windows (CTBase.run_tests path-separator bug)	✅ fixed upstream	Root-caused and fixed in CTBase.jl#512. Bumping OptimalControl.jl's CTBase compat to pull in that fix is deferred to a follow-up PR alongside other pending upstream updates.
7	Spurious DataFrames dep in docs/Project.toml	✅ done	Removed from [deps]/[compat] — confirmed unused anywhere under docs/.
8	CPU/GPU Julia version inconsistency in CI.yml	✅ done	Removed the pinned versions: input from both jobs; they now inherit the reusable CTActions workflow's default (["1.10", "1.12"]), so CPU and GPU always test the same two Julia versions instead of diverging.

See commit 7eb98a15.

---

Closes #802
Closes #803
Closes #805

🤖 Generated with Claude Code

[jbcaillau]

@tmigot can you please review?