SPICE: SPectral Integration Compiled Engine

A comprehensive Python library for modeling and analyzing stellar spectra with inhomogeneous surfaces, supporting rotation, pulsations, spots, and binary star systems.

The paper is submitted and available as a preprint.

Installation

Install from PyPI:

pip install stellar-spice

For PHOEBE integration support:

pip install stellar-spice[phoebe]

For zarr-backed model-atmosphere grid interpolation (LazyZarrInterpolator and friends):

pip install stellar-spice[grid]

Documentation

📖 Read the full documentation for detailed API reference and tutorials.

Key Features

🌟 Stellar Surface Modeling

• Mesh-based stellar surfaces using icosphere discretization
• Inhomogeneous temperature distributions across stellar surfaces
• Surface gravity variations accounting for rotation and shape distortions
• Line-of-sight velocity calculations for Doppler shift effects

🔄 Stellar Rotation

• Solid-body (rigid) rotation with a configurable axis and equatorial velocity
• Rotational broadening effects on spectral lines
• Surface velocity field calculations
• Time-dependent spectral variations due to rotation

🌊 Stellar Pulsations

• Spherical harmonic pulsation modes (l, m modes)
• Fourier series parameterization for complex pulsation patterns
• Multi-mode pulsations with different periods and amplitudes
• Surface displacement and velocity calculations

🌑 Stellar Spots

• Spherical harmonic spot modeling for complex spot distributions
• Temperature contrast between spots and photosphere
• Time-evolving spot patterns
• Spot-induced spectral variations

⭐ Binary Star Systems

• Full orbital dynamics with Keplerian orbits
• Mutual eclipses and occultations resolved on the projected meshes
• PHOEBE integration — import Roche-lobe geometry and tidally distorted binary meshes computed by PHOEBE

📊 Spectral Synthesis

• Blackbody radiation for basic stellar modeling
• Model-atmosphere grid interpolation from precomputed, zarr-backed grids
• Analytic line-profile emulators (Gaussian and physical)
• Transformer-Payne integration for ML-based spectral synthesis
• Custom spectral models via the SpectrumEmulator interface

🔍 Synthetic Photometry

• Multiple passband support (Johnson, Stromgren, Gaia, etc.)
• AB magnitude system calculations
• Bolometric luminosity computations
• Time-series photometry for variable stars

🎯 Advanced Features

• JAX-based computations for fast, differentiable calculations
• GPU acceleration support
• 3D visualization of stellar surfaces and binary systems
• Animation capabilities for time-evolving systems
• Occlusion handling for complex geometries

Quick Start

Basic Stellar Model

import numpy as np
from spice.models import IcosphereModel
from spice.models.mesh_transform import add_rotation
from spice.spectrum import simulate_observed_flux, Blackbody

bb = Blackbody()

# Create a solar-like star
star = IcosphereModel.construct(
    n_vertices=1000,                 # Mesh resolution (number of vertices)
    radius=1.0,                      # Solar radii
    mass=1.0,                        # Solar masses
    parameters=bb.solar_parameters,
    parameter_names=bb.parameter_names,
)

# Add solid-body rotation (equatorial velocity in km/s)
star = add_rotation(star, rotation_velocity=2.0)

# Generate the disc-integrated spectrum. simulate_observed_flux expects *log10*
# wavelengths and returns an (n_wavelengths, 2) array whose columns are the
# disc-integrated emulator channels (flux and continuum; column 0 is the flux).
wavelengths = np.logspace(3, 4, 1000)            # 1000-10000 Å
flux = simulate_observed_flux(bb.intensity, star, np.log10(wavelengths))

Binary Star System

import numpy as np
import jax.numpy as jnp
from spice.models import IcosphereModel, Binary
from spice.models.binary import add_orbit, evaluate_orbit_at_times
from spice.models.mesh_view import get_mesh_view
from spice.spectrum import simulate_observed_flux, Blackbody, AB_passband_luminosity
from spice.spectrum.filter import GaiaG

bb = Blackbody()
los = jnp.array([0.0, 1.0, 0.0])  # line of sight

# Create binary components (cast to the line of sight for occlusion handling)
primary = get_mesh_view(
    IcosphereModel.construct(1000, 1.0, 1.0, bb.solar_parameters, bb.parameter_names), los)
secondary = get_mesh_view(
    IcosphereModel.construct(1000, 0.8, 0.8, bb.solar_parameters, bb.parameter_names), los)

# Assemble the system
binary = Binary.from_bodies(primary, secondary)

# Add orbital elements
binary = add_orbit(
    binary,
    P=1.0,                    # orbital period [years]
    ecc=0.1,                  # eccentricity
    T=0.0,                    # time of periastron passage [years]
    i=np.pi / 3,              # inclination [rad]
    omega=0.0,                # argument of periastron [rad]
    Omega=0.0,                # longitude of the ascending node [rad]
    mean_anomaly=0.0,         # mean anomaly at the reference time [rad]
    reference_time=0.0,       # reference time [years]
    vgamma=0.0,               # systemic velocity [km/s]
    orbit_resolution_points=50,
)

# Evaluate the orbit across phases (eclipses/occlusions resolved internally)
times = jnp.linspace(0.0, 1.0, 100)
primaries, secondaries = evaluate_orbit_at_times(binary, times)

# Combined Gaia G-band light curve
wavelengths = np.linspace(900, 40000, 1000)
gaia_g = GaiaG()
light_curve = [
    AB_passband_luminosity(
        gaia_g,
        wavelengths,
        simulate_observed_flux(bb.intensity, p1, np.log10(wavelengths))[:, 0]
        + simulate_observed_flux(bb.intensity, p2, np.log10(wavelengths))[:, 0],
    )
    for p1, p2 in zip(primaries, secondaries)
]

PHOEBE Integration

import numpy as np
import phoebe
from phoebe.parameters.dataset import _mesh_columns
from spice.models import PhoebeBinary
from spice.models.binary import evaluate_orbit
from spice.models.phoebe_utils import PhoebeConfig
from spice.spectrum import simulate_observed_flux, Blackbody

bb = Blackbody()

# Build a PHOEBE binary and compute a mesh dataset (standard PHOEBE workflow)
b = phoebe.default_binary()
times = np.linspace(0, 1, 10)
b.add_dataset('mesh', compute_times=times, columns=_mesh_columns, dataset='mesh01')
b.run_compute(coordinates='uvw', overwrite=True)

# Wrap the PHOEBE meshes for SPICE (PHOEBE models are read-only inside SPICE)
config = PhoebeConfig(b, 'mesh01')
pb = PhoebeBinary.construct(config, ['teff', 'logg', 'abun'])

# Evaluate the components at a snapshot time, then synthesise the spectrum
primary, secondary = evaluate_orbit(pb, config.times[0])
wavelengths = np.logspace(3, 4, 1000)
flux = simulate_observed_flux(bb.intensity, primary, np.log10(wavelengths))

Performance

• JAX-powered computations for fast, vectorized operations
• GPU acceleration support for large-scale calculations
• Efficient mesh operations with optimized occlusion algorithms
• Memory-efficient spectral synthesis for time-series data

Known internals caveat

Mesh area conventions (slated for cleanup)

MeshModel currently exposes two area arrays that follow different unit conventions:

Any code that integrates over the stellar surface must apply the matching cgs prefactor:

• simulate_observed_flux integrates m.visible_cast_areas (already in R⊙²) and applies only the dimensionless dilution (R⊙ / pc)² / d_pc².
• simulate_monochromatic_luminosity integrates m.areas (unit-sphere) and applies R² × R⊙²[cm²] to convert to cgs.

If you add a new integrator over either array, mirror the matching pattern. This convention asymmetry is a known maintainability hazard and is slated for future cleanup: once m.areas is normalised the same way as m.visible_cast_areas, the cgs prefactor in simulate_monochromatic_luminosity should be reduced to R⊙²[cm²] only, matching the structure of simulate_observed_flux.

Citation

If you use stellar-spice in your research, please cite:

@misc{spice,
      title={SPICE -- modelling synthetic spectra of stars with non-homogeneous surfaces}, 
      author={M. Jabłońska, T. Różański, L. Casagrande, H. Shah, P. A. Kołaczek-Szymański, M. Rychlicki and Yuan-Sen Ting},
      year={2025},
      eprint={2511.10998},
      archivePrefix={arXiv},
      primaryClass={astro-ph.SR},
      url={https://arxiv.org/abs/2511.10998}, 
}

Contributing

We welcome contributions! Please see our contributing guidelines for details.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Built with JAX for fast, differentiable computations
• Integrates with PHOEBE for binary star modeling
• Uses Transformer-Payne for ML-based spectral synthesis

A preprint describing SPICE is available on arXiv. See Citation if you use it in your work.