starry-rust

A Rust numerical port of starry, with a new NumPy-based Python interface. The numerical kernels run in Rust; Python handles arrays and optimization orchestration. Upstream C++ and Theano are not runtime dependencies.

The supported domain and intentional numerical differences are specified in NUMERICAL_LIMITS.md. This is a numerical port, not a Python/Theano drop-in replacement.

Upstream starry depends on Theano and a C++ extension that are difficult to build against current Python. This port exists to generate spotted-star light curve templates without that toolchain, and runs anywhere Rust and NumPy do.

Features

• Emitted, reflected, finite-source, oblate and gravity-darkened maps.
• Occultations, overlapping silhouettes, rotation and radial velocities.
• Keplerian multi-body systems, light delays and exposure integration.
• Native first Jacobians for Map flux/RV, System flux/RV and Doppler spectra; orbital state Jacobians and Hessians.
• Doppler convolution, map/spectrum inversion, L1 and tempered solvers.
• Scalar/spectral Gaussian map and System inference with full covariances.
• Surface rendering, spots, numeric-image loading, harmonic/pixel transforms, smoothing, minimization and spectral cube factorization.
• Harmonic maps and combined products through degree 32, with the filter limits and singular boundaries documented in the numerical contract.

Use System(..., orbit_convention='starry') for the original orbital/RV conventions. The default barycentric mode is a documented extension. Use DopplerMap(..., continuum_index=...) for source-style sampled-continuum normalization; its default is integrated-continuum normalization.

Build and use

cargo build --release --offline
python -m unittest discover -s python -v
python setup.py bdist_wheel
python -m pip install dist/starry_rust-0.1.0-*.whl

Source builds require Rust, setuptools and wheel. Installed wheels require NumPy. Built and tested on Ubuntu Linux and Windows x64; the wheel is named for whichever platform builds it. macOS is not tested.

from starry_rust import Map
m = Map(3, udeg=2)
m.y[3] = .2
m.u[1:] = [.3, .1]
flux = m.flux(xo=.2, yo=.4, ro=.1)
jacobian = m.flux_jacobian(xo=.2, yo=.4, ro=.1)

For source-tree use, add python/ to PYTHONPATH; the loader finds the release library. STARRY_RUST_LIBRARY selects an explicit native library.

Documentation and verification

• Numerical API and executable examples
• Numerical limits and source differences
• Validation evidence and commands

Release verification passes 47 Rust tests, 77 Python tests and 200 selected upstream tests, together with the reference suites and installed-wheel examples on Windows and Linux.

python validation/record_run.py runs formatting, strict Clippy, Rust/Python tests, reference comparisons, source audits and selected original upstream tests. Reference compilation needs the separate ../starry-upstream checkout and a C++ compiler; prebuilt local reference executables can be reused.

Upstream's MIT license is retained in LICENSE. The separately vendored exoplanet validation reference retains its own license under validation/. No source-reference Python is used by the installed runtime.

Acknowledgements

This is a port of Rodrigo Luger's starry package and its C++/Theano core. The spherical harmonic basis, rotation and occultation solvers, reflected and oblate map models, Doppler machinery and Keplerian system layer are all ports of the original code. Numerical results agree with the original across roughly a thousand comparison cases: 2.37e-9 maximum absolute discrepancy on reflected flux over 120 cases, 1.10e-9 on oblate flux over 54, 7.28e-12 on RV filters and their derivatives over 60, and 2.71e-14 on analytic circular system flux over 88. VALIDATION.md gives the full table, and NUMERICAL_LIMITS.md records where this port departs from the original on purpose.