axilotl is a halo-model code that computes angular power spectra for patchy dark screening -- the conversion of CMB photons into axions or dark photons the ionized gas of dark-matter halos. Theoretical and observational details are provided in the following references:

• arXiv:2307.15124
• arXiv:2405.08059
• arXiv:2406.02546
• arXiv:2409.10514

Note that the code hasn't been optimized and isn't particularly fast. This is partly because there is no need for a fast dark-screening code, but, more importantly, is consistent with the code's namesake. Indeed, according to ChatGPT, "Axolotls aren't particularly fast swimmers since they rely on their feathery gills and tail for slow movement. While there's no precise measurement of their "top speed," axolotls generally swim at speeds between 1-2 body lengths per second when moving quickly, such as during feeding or escape responses. For a typical axolotl that's about 20 cm long, this translates to approximately 20-40 cm per second (0.2-0.4 m/s) under normal circumstances. They are better adapted for stealth and ambush predation than for speed."

What it computes

Given a halo-model backend (mass function, bias model, concentration, linear power spectrum from hmvec), a gas density profile, and a magnetic-field model (not needed for dark photons), axilotl builds the resonant-conversion optical depth tau_ell(z, M) for axion-photon oscillations or photon-dark-photon mixing and can compute the following power spectra:

• Cl_TT — dark-screening temperature auto-spectrum
• Cl_Tg — dark-screening temperature x galaxy cross-spectrum
• Cl_gg — galaxy auto-spectrum

Built-in components:

• Gas profiles: BattagliaGas (Battaglia+12), NFWGas
• Magnetic fields: ConstantB, CallableB, or an IllustrisTNG-measured B(r) built from axilotl.tng_bfield (spline-interpolated over 4 redshift bins x 8 halo-mass bins; see data/TNG_profiles/)
• HOD: standard 5-parameter central/satellite model (HODModel), configurable to match e.g. Kusiak+22/23 unWISE conventions, with a choice of halo mass definition (mass_def='vir' or '200c')

Install

git clone git@github.com:samgolds/axilotl.git
cd axilotl
pip install -e .

An editable install (-e) is recommended: axilotl locates its bundled data (data/TNG_profiles/) relative to the cloned repo, so keeping the package pointed at the source tree (rather than copied into site-packages) is what you want here. Once installed, import axilotl works from anywhere — no manual PYTHONPATH needed.

Quick start

import numpy as np
from scipy.interpolate import CubicSpline
from axilotl import (
    Cosmology, HaloModelBackend, BattagliaGas, ConstantB,
    ScreeningModel, HODModel, GalaxyKernel, ScreeningKernel, Cl_Tg,
)

cosmo = Cosmology(H0=67.32, ombh2=0.0224, omch2=0.1201)
ell_range = np.unique(np.logspace(0, np.log10(4000), 100).astype(int))
backend = HaloModelBackend(cosmo, ell_range, z_min=0.01, z_max=1.5, Nz=30,
                            Mvir_min=1e11, Mvir_max=1e15, NM=30)

gas = BattagliaGas(fb=cosmo.fb)
Bfield = ConstantB(B_muG=1.0, r_lo=0.1, r_hi=1.1)
screening = ScreeningModel(backend, gas, Bfield, dark_comp='axion', nu_GHz=353.0)

z = np.linspace(0, 2, 200)
dndz = CubicSpline(z, np.exp(-((z - 0.6) / 0.25) ** 2))
# 'lambda' is a Python keyword, so the satellite-slope parameter is set via dict
hod_params = {'alpha_s': 1.30, 'sigma_log_m': 0.69,
              'log10_m_min': 12.0, 'log10_m_star': 13.0, 'lambda': 1.0}
hod = HODModel(backend, hod_params, dndz, mass_def='200c')
galaxy = GalaxyKernel(hod)

sk = ScreeningKernel(screening, coupling=1e-11, m_dark_eV=3e-13)
Cl_1h, Cl_2h = Cl_Tg(sk, galaxy)

See examples/compute_templates.py for a full worked example: it reproduces the fiducial unWISE-blue x Planck/ACT axion templates from 2409.10514, including the IllustrisTNG magnetic-field model, and writes a grid of Cl_Tg(ell) templates spanning a range of axion masses to templates/.

Repository layout

axilotl/            core package (backend, gas/B-field profiles, screening, spectra)
data/                TNG_profiles/ (IllustrisTNG B(r) measurements), unWISE_dN_dz/
templates/           precomputed Cl_Tg(ell) template grid (unWISE-blue, K22 HOD)
examples/            compute_templates.py — full template-generation example