Source code for tengri.components.spatial.flat_slab

# SPDX-License-Identifier: BSD-3-Clause
"""Flat-slab (uniform aperture) spatial profile.

A uniform top-hat disk: intensity = 1 inside ``radius_kpc``, 0 outside.
This is the implicit "all-of-the-galaxy-is-here" model that classical
SED-fitting codes use when they scale a spectrum by a single aperture
factor. Architecturally explicit here so users can verify (or contest)
the assumption.

Notes
-----
The hard top-hat edge is smoothed with a small ``softness`` factor so the
function stays differentiable for gradient-based inference. The softened
profile is

.. math::

    I(r) = \\mathrm{sigmoid}\\!\\left(\\frac{R - r}{R \\cdot 0.01}\\right)

which is essentially indistinguishable from a hard top-hat at any
distance more than ~1% of R from the edge.
"""

from __future__ import annotations

from typing import ClassVar

import jax
import jax.numpy as jnp

from tengri.components.spatial_model_component import SpatialModelComponent
from tengri.parameters.priors import Uniform

__all__ = ["FlatSlab"]


[docs] class FlatSlab(SpatialModelComponent): """Uniform-disk (flat-slab) profile. Free parameters --------------- radius_kpc : Uniform(0.1, 50.0) [kpc] Disk radius. Intensity = 1 inside, 0 outside, with a small sigmoidal smoothing at the edge for differentiability. Notes ----- JIT/grad/vmap-compatible. The edge softening is a numerical convenience for gradient-based inference; it is not a physical claim. """ name = "flat_slab" parameter_prefix = "spatial_" radius_kpc = Uniform( 0.1, 50.0, description="Disk radius", units="kpc", default=1.0, ) reads: ClassVar[dict[str, str]] = {} publishes: ClassVar[dict[str, str]] = {"spatial_profile_2d": ""}
[docs] def predict(self, p, profile_in, grid_kpc): x, y = grid_kpc r = jnp.sqrt(x**2 + y**2) R = p["radius_kpc"] softness = R * 0.01 intensity = jax.nn.sigmoid((R - r) / softness) return intensity, {}