stringjax 0.1.0

Umbrella metapackage and tooling for the StringJAX ecosystem of differentiable string-theory codes.

The StringJAX project: Compiling String Theory

Differentiable tools for string compactifications in JAX.

StringJAX is the umbrella project for a family of interoperating, separately versioned and separately citable Python packages. This repository hosts the lightweight stringjax metapackage (a one-command installer and an environment helper) and the documentation hub. It contains no physics code — that lives in the member packages below.

Why StringJAX?

Constructing and analysing string-theory vacua has traditionally meant stitching together bespoke scripts, incompatible data formats, and slow, non-differentiable numerics. StringJAX exists to make that pipeline fast, differentiable, and reproducible from end to end. Every member package is built on JAX, so the same code runs on CPU, GPU, or TPU and exposes exact gradients through the whole geometry → effective theory → moduli-stabilisation chain. The packages are small, single-purpose, and independently versioned, so you install only what you need and cite exactly what you used.

Install

pip install stringjax            # CPU; JAXVacua engine + CYTools + bundled models
pip install "stringjax[all]"     # + databases (StringForge) and the flux-bounding extra

GPU/TPU/Apple-Silicon need one extra step for JAX — see the installation guide.

Note: stringjax is permissively licensed (MIT), but using the ecosystem pulls in jaxvacua, which is GPL-3.0; your usage is subject to those terms.

Packages

Package	Role	Install	Docs	GitHub
JAXVacua	Type IIB flux-vacuum engine	pip install jaxvacua	DOC	(GitHub)
JAXPolyLog	Differentiable polylogarithms	pip install jaxpolylog	DOC	(GitHub)
StringForge	Calabi–Yau databases & vacua vault	pip install stringforge	DOC	(GitHub)

StringJAX itself is the project/brand and is not an installable physics package.

First five minutes

import jaxvacua as jvc

jvc.set_precision("float64")                       # before heavy compute
model  = jvc.FluxEFT(h12=2, model_ID=1, limit="LCS", maximum_degree=5)
finder = jvc.FluxVacuaFinder.from_model(model)
vacua  = finder.sample_SUSY_flux_vacua(N=100, mode="ISD")
print(len(vacua), "candidate vacua")

Check your environment at any time:

stringjax doctor      # versions, JAX backend/devices, float64 status, Gurobi availability
stringjax versions    # just the member-package versions

Documentation

The documentation hub lives at https://stringjax.readthedocs.io. Each member package has its own authoritative documentation; the hub orients newcomers and links into them.

Roadmap & long-term goals

StringJAX is released incrementally, and the umbrella is deliberately thin so that member packages can evolve at their own pace.

• Near term — stabilise the JAXVacua flux-vacuum engine and its public API, publish archival releases with DOIs, and grow the StringForge data layer.
• Medium term — broaden coverage of the low-energy effective theory (the Kähler sector and α′/string-loop corrections, axion physics, and uplift scenarios) as further member packages that plug into the same differentiable interfaces.
• Long term — a unified, differentiable environment in which large regions of the landscape can be sampled, optimised, and connected to phenomenology and machine-learning workflows, with strong interoperability and reproducibility guarantees across the ecosystem.

The about page sets out the motivation and strategy in more detail; the compatibility matrix tracks current version windows.

Citing

Please cite the specific package(s) you used rather than the umbrella; see CITATION.cff and each member's citation file. BibTeX entries are collected in the documentation.

Contributing

See CONTRIBUTING.md and the code of conduct. Questions are welcome in GitHub Discussions.