jaxoplanet.starry.surface_is_physical

jaxoplanet.starry.surface_is_physical#

Functions#

fibonacci_grid(oversample, lmax)

surface_min_intensity(surface, oversample, lmax[, ...])

Fully JAX, end-to-end differentiable approximate global min:

Module Contents#

jaxoplanet.starry.surface_is_physical.fibonacci_grid(oversample: int, lmax: int)[source]#
jaxoplanet.starry.surface_is_physical.surface_min_intensity(surface: jaxoplanet.starry.surface.Surface, oversample: int, lmax: int, newton_iters: int = 12, damping: float = 0.001, step: float = 1.0)[source]#
Fully JAX, end-to-end differentiable approximate global min:

  1. seed from tiny equal-area grid

  2. run fixed M Newton steps from each seed in parallel

  3. take global min across seeds

Parameters:

  • surface – jaxoplanet Surface object with .intensity(lat, lon)

  • oversample – define N = oversample * lmax^2 seeds

  • lmax – define N = oversample * lmax^2 seeds

  • newton_iters – fixed Newton iterations per seed (no line search)

  • damping – Levenberg-Marquardt diagonal added to Hessian

  • step – Newton step scaling (e.g., 1.0 or 0.5)

Returns:

(lat_min, lon_min), min_val