ipax 0.1.1__tar.gz

ipax-0.1.1/.gitignore

@@ -0,0 +1,52 @@
+# Byte-compiled / optimized
+__pycache__/
+*.py[cod]
+*$py.class
+
+# Distribution / packaging
+build/
+dist/
+*.egg-info/
+.eggs/
+wheels/
+
+# Environments
+.venv/
+venv/
+env/
+.python-version
+
+# Tooling caches
+.mypy_cache/
+.ruff_cache/
+.pytest_cache/
+.hypothesis/
+.coverage
+.coverage.*
+htmlcov/
+coverage.xml
+
+# Benchmarks (asv)
+benchmarks/.asv/
+benchmarks/results/
+benchmarks/html/
+*.prof
+
+# Generated benchmark reports (regenerated by the `qc` job; kept as CI artifacts).
+# The directory itself is tracked via benchmarks/reports/.gitkeep.
+benchmarks/reports/*
+!benchmarks/reports/.gitkeep
+
+# Docs build
+site/
+
+# Editors / OS
+.vscode/
+.idea/
+*.swp
+.DS_Store
+Thumbs.db
+
+# Datasets (TROTS etc. are downloaded, never committed)
+data/
+*.mat

ipax-0.1.1/AGENTS.md

@@ -0,0 +1,268 @@
+# AGENTS.md — `ipax`
+
+Operational guide for any agent or contributor working in this repository. This is
+the **canonical** instructions file; `CLAUDE.md` imports it.
+
+`ipax` is a pure-Python, **Python Array API–conformant** primal–dual interior-point
+solver for large-scale nonlinear constrained optimization, tuned for radiotherapy
+(RT) treatment-planning scale. The math, the KKT reduction, and the design rules
+that govern changes live in this file; cite the source papers (see References) for the
+algorithmic details.
+
+---
+
+## What this project is
+
+- General NLP: `min f(x)` s.t. `c(x)=0`, `g(x)≤0`, `x_L ≤ x ≤ x_U`, solved by a
+  log-barrier interior-point method with Lagrange multipliers.
+- **Scale target:** `1e3`–`1e5` variables; dense *or* sparse Jacobians/Hessians.
+- **Default Hessian:** L-BFGS (compact, Powell-damped) of the Lagrangian.
+- **Globalization:** filter line-search (IPOPT, default); Breedveld step controller
+  as an alternative mode.
+- **Backends:** NumPy + PyTorch in CI; CuPy/JAX (incl. GPU) supported via the same
+  code path. Inspirations: IPOPT (Wächter & Biegler 2006), Breedveld et al. (2017).
+
+---
+
+## Non-negotiable invariants (read first)
+
+1. **No concrete array library in the core.** Never `import numpy`/`torch`/`cupy`/
+   `jax` inside `ipax/` except in `ipax/backend/sparse/` and other explicitly
+   labeled adapters. Get the namespace from the input arrays:
+   ```python
+   from ipax.backend.namespace import array_namespace
+   xp = array_namespace(x)          # then use xp.*, xp.linalg.*
+   ```
+2. **Stay inside the standard.** Use only the Array API main namespace + the optional
+   `linalg` extension. The extension provides `cholesky, eigh/eigvalsh, qr, svd,
+   solve, inv, pinv, slogdet, matrix_norm, vector_norm, …`. It does **NOT** provide
+   LDLᵀ-with-inertia, triangular solve, `lstsq`, or any sparse type — do not assume
+   them. Gap-fillers go in `backend/` with a comment naming the missing primitive.
+3. **Linear algebra is injected, never hard-wired.** All Jacobians/Hessians are
+   `LinearOperator`s; all KKT solves go through the `LinearSolver` protocol. Adding a
+   solve strategy must not touch `ipm/driver.py`.
+4. **Sparsity is an adapter concern.** The core emits structure as Array-API
+   index/value vectors; per-backend wrappers in `backend/sparse/` build the actual
+   sparse matrix and factor it. The IPM never sees a backend sparse object.
+5. **No `localStorage`-style global state.** Solver state lives in explicit objects;
+   no module-level mutable singletons.
+
+If a change appears to require breaking 1–4, stop and discuss in the PR description
+instead of working around it.
+
+---
+
+## Architecture map
+
+```
+ipax/
+  typing.py options.py result.py solve.py
+  problem/      base.py (Problem ABC), function.py, derivatives.py,
+                finitediff.py, autodiff/{jax,torch}.py
+  backend/      namespace.py, operators.py,
+                sparse/{numpy_scipy,cupy,torch,jax,_routing}.py
+  linalg/       solver.py, dense.py, krylov.py, regularize.py
+  ipm/          barrier, kkt, step, filter_ls, restoration, breedveld_ls,
+                hessian, init, driver
+  testing/      problems.py (analytic oracles), backends.py
+tests/          contracts/, unit/, property/, integration/, backends/, regression/
+examples/       minimal runnable examples for the current implementation surface
+benchmarks/     generators/, corpus/, harness, runners, reports
+```
+
+- **`Problem`** (`problem/base.py`): user-facing ABC. Required: `n_vars`,
+  `objective`. Optional (resolved by `derivatives.py`): `gradient`, nonlinear
+  `eq/ineq_constraints` + Jacobians, and `lagrangian_hessian`. **Linear constraints
+  are declared separately** (`linear_eq`/`linear_ineq`, constant data) from nonlinear
+  ones — constant Jacobian, no Hessian term, assembled once. Derivative precedence:
+  analytic → autodiff → finite-diff (grads/Jacobians); analytic → autodiff-HVP →
+  **L-BFGS** (Hessian). Never form a dense Hessian at scale.
+- **`LinearOperator`** (`backend/operators.py`): `matvec`/`rmatvec`/`matmat`;
+  subclasses `Dense, Diagonal, LowRank, LBFGSOperator, MatrixFreeJacobian,
+  SparseOperator, Composite`. All `Problem` Jacobians/Hessians normalize to these.
+- **`LinearSolver`** (`linalg/solver.py`): `DenseSolver` (Cholesky/solve),
+  `KrylovSolver` (CG/MINRES/GMRES, matrix-free — default at scale), and
+  `SparseDirectSolver` (per-backend). Auto-selected by size/density/capabilities.
+
+---
+
+## Math conventions (match the code to these)
+
+- Standard form uses slacks: `g(x)+s=0, s≥0`. Multipliers: `y` (equalities),
+  `λ` (inequalities/slacks), `z_L,z_U` (bounds). `diag(v)=V`; `e`=all-ones.
+- Two KKT solve routes: the **condensed normal-equations** route (Array-API-native,
+  default for dense & matrix-free; PD via Powell-damped L-BFGS + primal–dual
+  regularization, so **no inertia oracle required**) and the **indefinite augmented**
+  route (sparse-direct backends with LDLᵀ inertia only).
+- Regularization: Friedlander–Orban primal–dual (`δ_w` on (1,1), `−δ_c` on (2,2));
+  on Cholesky failure escalate `δ_w` from `1e-6`, doubling, per Breedveld.
+- Convergence: scaled KKT ∞-norm ≤ `ε_tol` (default `1e-8`), IPOPT scaling
+  `s_d, s_c`.
+
+Keep symbol names in code aligned with the math conventions above (`W, Sigma_x,
+Sigma_s, N, mu, tau, alpha, theta, phi`).
+
+---
+
+## Coding standards
+
+- Python ≥ 3.10, full type hints, `from __future__ import annotations`.
+- Public API documented with docstrings; numerical choices cite the paper/eq.
+- Config via frozen dataclasses in `options.py`; no magic numbers in the loop body.
+- Formatting/lint: `ruff` (format + lint) and `mypy` clean before commit.
+- Pure functions where practical; no hidden global state; explicit `xp`/device
+  threading.
+- Floats are `float64` by default; never hard-code a dtype — read it from inputs.
+
+---
+
+## Testing & verification
+
+- **Test-driven.** For every change write the failing tests first, then implement
+  to green. New protocol implementations must pass the shared **contract test
+  batteries** (`tests/contracts/`) for `Problem`, `LinearOperator`, `LinearSolver`,
+  `SparseDirectSolver` — this is what keeps "pluggable" honest.
+- **Always multi-backend.** Tests parametrize over the namespace fixture
+  (`tests/conftest.py`): NumPy + PyTorch minimum; CuPy/JAX/GPU when available.
+  `array-api-strict` is the **purity gate** (raises on out-of-standard calls).
+- **Purity check** in CI: fail if banned imports appear outside allowed adapters.
+- **Oracles:** verify KKT conditions at the solution; closed-form QP/LP optima,
+  Hock–Schittkowski set, cross-check vs `scipy.optimize` and `cyipopt` when installed.
+- **Derivative harness:** FD vs analytic vs autodiff grad/Jacobian + Hessian-vector
+  checks (also a public utility usable on any user `Problem`).
+- **Benchmarks are separate** (`benchmarks/`, not `tests/`): synthetic RT-like
+  generators (`1e3`–`1e5` vars, 5–50% density) + standard sets, tracked over commits
+  with `asv`; run nightly, not per-PR.
+- A non-trivial change is not done until the relevant tests pass on **both** NumPy
+  and PyTorch. Add a regression test with every bug fix.
+
+### Commands
+
+`scripts/check.py` is the **single verification entrypoint** — it runs the same
+gates CI runs (`format → lint → types → purity → test`), so "is my change done?"
+is one command instead of reconstructing flags from `ci.yml`, the pre-commit
+config and this file. Prefer it over invoking the tools by hand.
+
+```bash
+python scripts/check.py                      # all gates (multi-backend tests included)
+python scripts/check.py --fast               # skip the slow test gate (lint/types/purity only)
+python scripts/check.py lint types           # only the named gates
+python scripts/check.py --list               # show the gates and their commands
+
+# Equivalent raw invocations (what the gates wrap):
+pytest -q                                    # full test suite
+IPAX_BACKENDS=numpy,torch,array_api_strict pytest -q
+ruff check . && ruff format --check . && mypy ipax
+python scripts/check_purity.py               # import-purity gate (invariants #1/#4)
+asv run                                      # tracked perf benchmarks (benchmarks/)
+```
+
+Enforcement layers (same `check_purity.py` logic, different triggers): the
+**pre-commit** hook runs it at commit; a **Claude Code PostToolUse hook**
+(`scripts/hooks/purity_guard.py`, wired in `.claude/settings.json`) runs it the
+moment an edit touches `ipax/`, so a purity violation is caught immediately
+rather than at commit. Other agents (e.g. aider's `lint-cmd`) can point at the
+same script.
+
+### Agent tooling
+
+Claude Code config lives in `.claude/` and is checked in:
+
+- **`invariant-auditor`** subagent (`.claude/agents/invariant-auditor.md`) — a
+  read-only reviewer that audits a diff against the five invariants, the math/citation
+  conventions, the testing discipline, and the GPU/device-efficiency rules (the
+  *semantic* checks the purity hook and contract tests cannot see). Invoke it
+  explicitly before a commit or PR — e.g. *"use the invariant-auditor to review my
+  changes"*.
+- **`/verify`** (`.claude/commands/verify.md`) — run `scripts/check.py` and summarize.
+- **`/tdd`** (`.claude/commands/tdd.md`) — drive a change through the mandated
+  red→green→verify→regression loop, multi-backend and invariant-aware.
+
+The auditor's rubric is the static half of GPU performance work; the measured half is
+a future `benchmarks/runners/device_efficiency.py` (sync/kernel profiling on real
+hardware), deferred until GPU CI exists.
+
+---
+
+## Scope guardrails
+
+**In:** equality + inequality + bound constraints; L-BFGS + exact Hessian; dense,
+matrix-free, and sparse-direct solver routes; filter line-search + restoration;
+optional Mehrotra–Gondzio higher-order corrections; multi-backend.
+
+**Out (do not add without discussion):** Breedveld's dose-matrix `N=AᵀDA+Q+T`
+condensation, graph permutation/tiling, mixed-precision tiled BLAS, chained
+dose-influence products; hard-coded RT cost-functions (LTCP/gEUD/dose-volume);
+multi-criteria / Pareto / beam-angle layers. Keep dimensionality & sparsity in
+mind, but the RT-specific kernels belong in a separate downstream layer.
+
+---
+
+## Direction
+
+`ipax` is **beta**. The solver surface above is complete; near-term work
+is **maintenance and measurement**, not new features. The project is also actively
+**looking for adopters** — early real-world use on the `Problem`/`solve` API is what
+drives the road to `1.0.0`.
+
+### Road to 1.0.0
+
+`1.0.0` is the point where the public API carries a semver stability promise. It is
+gated on these exit criteria (the day-to-day work below is *how* we get there, not a
+new feature list):
+
+1. **API stability.** The public surface — `solve`, `Problem`, `Options`, `Result`,
+   and the `LinearOperator`/`LinearSolver` protocols — holds stable across the `0.x`
+   series, with any deprecations carried for at least one minor release.
+2. **Published documentation.** The MkDocs site (`docs/`: API reference + guides) is
+   built and published.
+3. **GPU & sparse-direct verification.** The CuPy/cuDSS and Torch/JAX-CUDA sparse
+   routes and the MUMPS inertia path are exercised on real hardware in CI, or
+   documented as provisional until GPU CI exists.
+4. **Adopter validation.** At least one external workload runs on `ipax`, with the
+   feedback folded back into the API.
+5. **Tracked baselines.** `asv` performance baselines and the QC accuracy sweep are
+   tracked across releases, with cross-checks vs IPOPT/SciPy/OSQP green.
+
+Near-term work is **maintenance and measurement**, not new features:
+
+- **Bug-fixing.** Tighten correctness on hard nonconvex/ill-conditioned problems;
+  every fix ships with a regression test (`tests/regression/`).
+- **Performance.** Drive the `benchmarks/` scaling and micro suites (`asv`), reduce
+  per-iteration cost and memory across the dense/matrix-free/sparse routes, and
+  profile device efficiency once GPU CI exists.
+- **Accuracy benchmarking.** Grow the QC sweep and reference cross-checks
+  (vs IPOPT/SciPy/OSQP), track scaled-KKT accuracy and robustness across the corpus
+  and backends over time.
+
+New algorithmic features beyond the current scope are out until discussed.
+
+---
+
+## Documentation
+
+Documentation lives in the folder docs and uses mkdocs.
+
+Minimal runnable examples live in `examples/`. Keep them short, feature-scoped,
+and runnable from the repository root with the local virtual environment. Examples
+may import a concrete backend such as NumPy at the application edge; the `ipax/`
+core must remain backend-agnostic.
+
+---
+
+## References
+
+- Wächter, A. & Biegler, L. T. (2006). "On the implementation of an interior-point
+  filter line-search algorithm for large-scale nonlinear programming."
+  *Mathematical Programming* 106(1), 25–57. <https://doi.org/10.1007/s10107-004-0559-y>
+- Breedveld, S., van den Berg, B. & Heijmen, B. (2017). "An interior-point
+  implementation developed and tuned for radiation therapy treatment planning."
+  *Computational Optimization and Applications* 68(2), 209–242.
+  <https://doi.org/10.1007/s10589-017-9919-4>. TROTS dataset <http://www.trots.eu>.
+- Array API `linalg` extension:
+  <https://data-apis.org/array-api/latest/extensions/linear_algebra_functions.html>
+- `array-api-compat` / `array-api-extra`:
+  <https://data-apis.org/array-api-compat/> · <https://data-apis.org/array-api-extra/>
+- Friedlander, M. P. & Orban, D. (2012). "A primal–dual regularized interior-point
+  method for convex quadratic programs." *Mathematical Programming Computation* 4(1),
+  71–107. <https://doi.org/10.1007/s12532-012-0035-2>

ipax-0.1.1/CLAUDE.md

@@ -0,0 +1,27 @@
+# CLAUDE.md
+
+The canonical instructions for this repository live in **[AGENTS.md](./AGENTS.md)**.
+They are imported below so Claude Code loads them automatically:
+
+@AGENTS.md
+
+---
+
+## Claude-specific notes
+
+- Treat `AGENTS.md` as the single source of truth. If guidance here ever conflicts
+  with it, `AGENTS.md` wins — and update one of the two files so they agree again.
+- Before non-trivial work, read the math conventions and KKT reduction in
+  `AGENTS.md`, and the source papers they cite (full citations in the AGENTS.md
+  References section).
+- The five **non-negotiable invariants** in `AGENTS.md` (no concrete array library
+  in the core, stay inside the Array API standard, injected linear algebra, sparsity
+  as an adapter concern, no global mutable state) are hard constraints. If a task
+  seems to need breaking one, say so and propose options instead of working around
+  it silently.
+- When you touch numerics, add or run the **multi-backend** tests (NumPy **and**
+  PyTorch at minimum) and the derivative-check harness before declaring done.
+- Cite the relevant paper/equation in code comments when implementing an algorithmic
+  step (e.g. `# Wächter & Biegler 2006, eq. (19)` / `# Breedveld 2017, eq. (18)`).
+- Consult the source papers (full citations with DOIs in the AGENTS.md References
+  section) rather than guessing at the algorithm details.

ipax-0.1.1/CONTRIBUTING.md

@@ -0,0 +1,91 @@
+# Contributing to `ipax`
+
+Thanks for working on `ipax`. The canonical, detailed contributor guide is
+[`AGENTS.md`](./AGENTS.md) — it is the single source of truth for the math
+conventions, architecture, and the rules below. This file is the short version.
+
+## Non-negotiable invariants
+
+These are hard constraints. A change that appears to require breaking one should be
+raised for discussion in the PR, not worked around silently.
+
+1. **No concrete array library in the core.** Never import `numpy`/`torch`/`cupy`/
+   `jax` inside `ipax/` except in `ipax/backend/sparse/` and `ipax/problem/autodiff/`.
+   Get the namespace from the input arrays:
+   ```python
+   from ipax.backend.namespace import array_namespace
+   xp = array_namespace(x)
+   ```
+2. **Stay inside the Array API standard** — main namespace + the optional `linalg`
+   extension only. Gap-fillers (triangular solve, `lstsq`, …) go in `backend/` with a
+   comment naming the missing primitive.
+3. **Linear algebra is injected, never hard-wired.** Jacobians/Hessians are
+   `LinearOperator`s; all KKT solves go through the `LinearSolver` protocol. Adding a
+   solve strategy must not touch `ipm/driver.py`.
+4. **Sparsity is an adapter concern.** The core emits structure as Array-API
+   index/value vectors; per-backend wrappers in `backend/sparse/` build and factor the
+   sparse matrix. The IPM never sees a backend sparse object.
+5. **No global mutable state.** Solver state lives in explicit objects.
+
+The purity boundary (invariants #1/#4) is enforced by `scripts/check_purity.py` — at
+commit via a pre-commit hook and, for Claude Code, via a PostToolUse hook the moment
+an edit touches `ipax/`.
+
+## Coding standards
+
+- Python ≥ 3.10, full type hints, `from __future__ import annotations`.
+- Public API documented with docstrings; numerical choices cite the paper/equation
+  (e.g. `# Wächter & Biegler 2006, eq. (19)`).
+- Config via frozen dataclasses in `options.py`; no magic numbers in the loop body.
+- Floats are `float64` by default — never hard-code a dtype; read it from inputs.
+- `ruff` (format + lint) and `mypy` clean before commit.
+
+## Testing
+
+- **Test-driven.** Write the failing test first, then implement to green.
+- **Always multi-backend.** Numerical tests parametrize over the namespace fixture
+  (NumPy + PyTorch minimum; CuPy/JAX/GPU when available); `array-api-strict` is the
+  purity gate.
+- New protocol implementations must pass the shared **contract batteries**
+  (`tests/contracts/`).
+- Every bug fix ships with a regression test in `tests/regression/`.
+
+## Verification
+
+`scripts/check.py` is the single verification entrypoint and runs exactly what CI
+runs (`format → lint → types → purity → test`):
+
+```bash
+python scripts/check.py            # all gates (multi-backend tests included)
+python scripts/check.py --fast     # skip the slow test gate
+python scripts/check.py --list     # show the gates and the raw commands they wrap
+```
+
+## Direction
+
+`ipax` is **beta** and **looking for adopters** — trying it on a real
+workload and reporting back is one of the most valuable contributions right now.
+The solver surface is complete; near-term work is **maintenance and measurement**
+rather than new features:
+
+- **Bug-fixing** — correctness on hard nonconvex / ill-conditioned problems, each fix
+  with a regression test.
+- **Performance** — drive the `benchmarks/` scaling and micro suites (`asv`); reduce
+  per-iteration cost and memory across the dense / matrix-free / sparse routes.
+- **Accuracy benchmarking** — grow the QC sweep and reference cross-checks
+  (vs IPOPT/SciPy/OSQP); track scaled-KKT accuracy and robustness over time.
+
+### Road to 1.0.0
+
+`1.0.0` carries a semver stability promise for the public API. It is gated on:
+**(1)** a stable public surface (`solve`, `Problem`, `Options`, `Result`,
+`LinearOperator`/`LinearSolver`) across the `0.x` series, deprecations carried one
+minor release; **(2)** the published MkDocs site; **(3)** GPU/sparse-direct
+(CuPy/cuDSS, Torch/JAX-CUDA, MUMPS inertia) verified on real hardware in CI or
+documented as provisional; **(4)** at least one external adopter workload with
+feedback folded back in; **(5)** tracked `asv` and QC baselines with reference
+cross-checks green. See [`README.md`](./README.md#road-to-100) and
+[`AGENTS.md`](./AGENTS.md#road-to-100) for the full version.
+
+New algorithmic features beyond the current scope (see the scope guardrails in
+`AGENTS.md`) are out until discussed.

ipax-0.1.1/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Niklas Wahl
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

ipax-0.1.1/NOTICE

@@ -0,0 +1,28 @@
+ipax
+Copyright 2026 Niklas Wahl
+
+Licensed under the Apache License, Version 2.0. See the LICENSE file for the
+full terms.
+
+------------------------------------------------------------------------------
+Algorithmic attribution
+------------------------------------------------------------------------------
+
+ipax is an independent, clean-room implementation written from the published
+literature. It contains no source code from the works below; they are cited as
+the academic sources of the algorithms it implements:
+
+  - Wächter, A. & Biegler, L. T. (2006). "On the implementation of an
+    interior-point filter line-search algorithm for large-scale nonlinear
+    programming." Mathematical Programming 106(1), 25-57.
+    https://doi.org/10.1007/s10107-004-0559-y
+
+  - Breedveld, S., van den Berg, B. & Heijmen, B. (2017). "An interior-point
+    implementation developed and tuned for radiation therapy treatment
+    planning." Computational Optimization and Applications 68(2), 209-242.
+    https://doi.org/10.1007/s10589-017-9919-4
+
+  - Friedlander, M. P. & Orban, D. (2012). "A primal-dual regularized
+    interior-point method for convex quadratic programs." Mathematical
+    Programming Computation 4(1), 71-107.
+    https://doi.org/10.1007/s12532-012-0035-2