zodi

Zodiacal and exozodiacal light brightness conventions for exoplanet imaging, from a single source that runs on numpy and JAX.

Status: v0.1 core (pre-PyPI). The LBTI HOSTS population distributions land next.

What it owns

• Local zodiacal light surface brightness: the Leinert et al. (1998) Table 17 position dependence and Table 19 wavelength dependence, with the interpolation, anchoring, and near-Sun conventions of the production codes stated explicitly (zodi.specific_intensity, zodi.zodi_flux_ratio).
• The exozodi chain of Stark et al. (2014), anchored at the EEID and numerically compatible with EXOSIMS calc_JEZ0, with the radial scaling measured from the EEID (zodi.exozodi_flux_ratio_v, zodi.jez0, zodi.scale_jez), the three published latitudinal models (zodi.latitudinal_factor), and the grey-scatterer band correction in both production flavors: the stellar-color scaling used by pyEDITH (zodi.exozodi_flux_ratio_band) and the scattered-plus-thermal spectrum model used by EXOSIMS, with its calibration as a closed-form least-squares fit (zodi.fit_grey_scatter_constants).
• Unit conversions between the dialects in common use: magnitudes per square arcsecond, flux ratio per square arcsecond, spectral radiance, photon rates, and MJy per steradian (zodi.units).

Coming next: the LBTI HOSTS survey n-zodi population distributions as quantile functions (callers supply their own uniform draws).

Documentation and evidence

docs/conventions.md states every model, constant, and known cross-code difference with sources and measured deltas; docs/evidence.md maps each claim to an executable check and its kind of evidence. Cross-code benchmark scripts against EXOSIMS, skyscapes, and zodipy live in scripts/ and are runnable by anyone with those packages installed. Current results: exact agreement with the EXOSIMS closed-form models and magnitude chain, float64 round-off agreement with skyscapes, and 0.95-1.23 brightness ratios against the independent Kelsall model at 1.25 um.

Design

One implementation serves both backends through the array API standard. Functions compute in the namespace of their array inputs (via array-api-compat), so numpy callers get numpy in and numpy out with no JAX anywhere in their dependency tree, while JAX callers get functions that jit, vmap, and differentiate natively. Hard dependencies are numpy and array-api-compat only; the [jax] extra exists so the test suite can run the JAX side. JAX users should enable float64:

jax.config.update("jax_enable_x64", True)

Random sampling is deliberately absent from the library: distributions ship as quantile functions (inverse CDFs), and callers bring uniforms from their own generator, whether that is numpy.random or jax.random.

Not this library

Structured circumstellar disks (rings, gaps, offsets) and scene rendering belong to scene simulators; solar-system ephemerides belong to the caller. For thermal-infrared zodiacal emission modeling, see zodipy; this library covers the reflected-light brightness conventions used in exoplanet direct imaging.

Install

PyPI release pending; for now:

pip install git+https://github.com/CoreySpohn/zodi.git