ParSub v0.2.1

Packaging and documentation release; ParSub's code is unchanged from 0.2.0.

• Docker image of the REST API: ghcr.io/psubrat29/parsub for amd64 and arm64, built, tested and published by GitHub Actions; tags 0.2.1, 0.2 and latest.
• compose.yaml to start the API with docker compose up -d, and a new Docker page.
• The PyPI page now shows the current README, with links to the website, the Docker image and the changelog.
• Release workflow: TestPyPI rehearsals can be repeated (unique .devN versions), PyPI uploads check the version first, and every upload is followed by installing it from the index.

Install: pip install -U parsub · Docker: docker pull ghcr.io/psubrat29/parsub:0.2.1

ParSub v0.2.0

First release on PyPI. ParSub reads the mathematics in a LaTeX document, works out what can be computed, writes a ready-to-run Python script, and numerically checks the paper's identities.

Install

pip install parsub

Requires Python 3.9 or newer.

Highlights

• Working end-to-end pipeline. LaTeX → SymPy → computation tasks → generated script → plots and data. Version 0.1.0 extracted no expressions at all.
• Special-function notation: hypergeometric functions {}_pF_q(a; b; z), Pochhammer symbols (a)_n, Laguerre/Gegenbauer/Jacobi polynomials L_n^{(a)}(x), Bessel functions J_\nu(x), indexed functions w_{\alpha}(z) and derivatives w''(z).
• Uses the document's own definitions. Functions defined in the paper are substituted where they are used, parameter definitions such as \vartheta = \alpha + (b+1)/2 set consistent values, and i = \sqrt{-1} is honoured.
• Checks the mathematics. Identities, repeated definitions of the same function and differential equations are verified numerically, with a clear verdict for each.
• Goal detection from the prose: evaluate, plot, solve, optimize, integrate, differentiate, series, verify.
• Robust generated scripts. Every task runs in isolation with a time limit; outputs are 300 DPI plots, CSV/TSV/Excel/JSON data and a summary.json.
• Three ways to use it: a parsub command (analyze, run, demo, version), a Python API (parsub.analyze_latex) and a REST API (parsub-api).

Validated on a real paper

On examples/sample.tex (25 numbered equations on generalized Bessel functions), ParSub:

• converts 38 of 39 formulas and runs 28 computations with no failures;
• confirms 19 of 22 identities;
• flags two misprints, Eq. (6) (Kummer's first transformation) and Eq. (22) (Laguerre index), plus the condition α > 0 for w_α(0) = 0.

Each finding was confirmed independently with mpmath. See the validation report: https://psubrat29.github.io/parsub/docs/validation.html

Also in this release

• REST API with sandboxed file paths, correct HTTP status codes and a POST /run endpoint.
• Rewritten documentation and a project website: https://psubrat29.github.io/parsub/
• Automated tests on Python 3.9, 3.11 and 3.13; automated website checks and PyPI publishing.

Upgrading from 0.1.0

0.1.0 was never published on PyPI and did not work. Install 0.2.0 with pip install parsub. If you used the Python API, task dictionaries now carry independent_variables, fixed_parameters and srepr fields; see the API reference.

Full changelog: v0.1.0...v0.2.0