BCemu 2.0.5

Using emulators to implement baryonic effects.

BCemu

A Python package for modelling baryonic effects in cosmological simulations.

Package details

The package provides emulators to model the suppression in the power spectrum due to baryonic feedback processes. These emulators are based on the baryonification model (Schneider et al. 2019), where gravity-only N-body simulation results are manipulated to include the impact of baryonic feedback processes. For a detailed description, see Giri & Schneider (2021).

INSTALLATION

One can install a stable version of this package using pip by running the following command::

pip install BCemu

In order to use the latest version, one can clone this package running the following::

git clone https://github.com/sambit-giri/BCemu.git

To install the package in the standard location, run the following in the root directory::

python setup.py install

In order to install it in a separate directory::

python setup.py install --home=directory

One can also install it using pip by running the following command::

pip install git+https://github.com/sambit-giri/BCemu.git

The core dependencies are installed automatically. Some features require optional packages:

• BCemu2025 — differentiable backends (default is numpy, no extra install needed):

  pip install jax flax     # for backend='jax'  (CPU/GPU/TPU, differentiable via jax.grad)
  pip install torch        # for backend='torch' (differentiable via autograd)
  

• BCemu2021 emulators (BCM_7param / BCM_3param):

  pip install smt==1.0.0
  

  A clear error message is shown if you try to use these emulators without smt installed.

Tests

For testing, use pytest, which can be installed via pip. Tests are split by emulator version so you only download what you need:

Test file	Marker	What it tests
tests/test_BCemu2021.py	bcemu2021	BCM 7-param and 3-param emulators (BCemu2021)
tests/test_BCemu2025.py	bcemu2025	JAX, numpy, and torch backends (BCemu2025)

Run all tests:

python -m pytest tests

Run tests for one emulator only (avoids downloading the other):

python -m pytest tests -m bcemu2021
python -m pytest tests -m bcemu2025

📖 Citation

If you use BCemu in your research, please cite the following paper:

Giri, S. K., & Schneider, A. (2021). Emulation of baryonic effects on the matter power spectrum and constraints from galaxy cluster data. Journal of Cosmology and Astroparticle Physics, 2021(12), 046. https://doi.org/10.1088/1475-7516/2021/12/046

BibTeX entries:

@article{giri2021emulation,
  title={Emulation of baryonic effects on the matter power spectrum and constraints from galaxy cluster data},
  author={Giri, Sambit K and Schneider, Aurel},
  journal={Journal of Cosmology and Astroparticle Physics},
  volume={2021},
  number={12},
  pages={046},
  year={2021},
  publisher={IOP Publishing}
}

USAGE

BCemu2025

The 2025 emulator covers 8 BCM parameters, redshifts up to z = 3, and supports differentiable inference. The default backend is numpy (no extra packages required). Install jax+flax or torch to use differentiable backends.

import numpy as np
import BCemu

# Default numpy backend — fast, no extra dependencies
bfcemu = BCemu.BCemu2025()

Ob, Om = 0.0486, 0.306
bcmdict = {
    'Theta_co': 0.3,
    'log10Mc':  13.1,
    'mu':       1.0,
    'delta':    6.0,
    'eta':      0.10,
    'deta':     0.22,
    'Nstar':    0.028,
    'fb':       Ob / Om,
}

k, S = bfcemu.get_boost(bcmdict, z=0.5)

Differentiable usage (requires jax+flax):

import jax
import jax.numpy as jnp
import BCemu

bfcemu = BCemu.BCemu2025(backend='jax')
params = jnp.array([bcmdict[k] for k in bfcemu.param_names])

# Forward pass
S = bfcemu.get_boost_differentiable(params, z=0.5)

# Full Jacobian ∂S/∂θ
J = jax.jacfwd(bfcemu.get_boost_differentiable)(params, z=0.5)

BCemu2021

Requires smt==1.0.0 (pip install smt==1.0.0). Script to get the baryonic power suppression.

import numpy as np
import matplotlib.pyplot as plt
import BCemu

bfcemu = BCemu.BCM_7param(Ob=0.05, Om=0.27)
bcmdict = {'log10Mc': 13.32,
           'mu'     : 0.93,
           'thej'   : 4.235,  
           'gamma'  : 2.25,
           'delta'  : 6.40,
           'eta'    : 0.15,
           'deta'   : 0.14,
           }

z = 0
k_eval = 10**np.linspace(-1,1.08,50)
p_eval = bfcemu.get_boost(z, bcmdict, k_eval)

plt.semilogx(k_eval, p_eval, c='C0', lw=3)
plt.axis([1e-1,12,0.73,1.04])
plt.yticks(fontsize=14)
plt.xticks(fontsize=14)
plt.xlabel(r'$k$ (h/Mpc)', fontsize=14)
plt.ylabel(r'$\frac{P_{\rm DM+baryon}}{P_{\rm DM}}$', fontsize=21)
plt.tight_layout()
plt.show()

The package also has a three-parameter baryonification model. Model A assumes all three parameters to be independent of redshift while model B assumes the parameters to be redshift-dependent via the following form:

.

Below an example fit to the BAHAMAS simulation result is shown.

import numpy as np 
import matplotlib.pyplot as plt
import BCemu
import pickle

BAH = pickle.load(open('examples/BAHAMAS_data.pkl', 'rb'))

bfcemu = BCemu.BCM_3param(Ob=0.0463, Om=0.2793)
bcmdict = {'log10Mc': 13.25, 
           'thej'   : 4.711,  
           'deta'   : 0.097}

zs = [0,0.5]
k_eval  = 10**np.linspace(-1,1.08,50)
p0_eval1 = bfcemu.get_boost(zs[0], bcmdict, k_eval)
p1_eval1 = bfcemu.get_boost(zs[1], bcmdict, k_eval)

bfcemu = BCemu.BCM_3param(Ob=0.0463, Om=0.2793)
bcmdict = {'log10Mc': 13.25, 
           'thej'   : 4.711,  
           'deta'   : 0.097,
           'nu_Mc'  : 0.038,
           'nu_thej': 0.0,
           'nu_deta': 0.060}

zs = [0,0.5]
k_eval  = 10**np.linspace(-1,1.08,50)
p0_eval2 = bfcemu.get_boost(zs[0], bcmdict, k_eval)
p1_eval2 = bfcemu.get_boost(zs[1], bcmdict, k_eval)

plt.figure(figsize=(10,4.5))
plt.subplot(121); plt.title('z=0')
plt.semilogx(BAH['z=0']['k'], BAH['z=0']['S'], '-', c='k', lw=5, alpha=0.2, label='BAHAMAS')
plt.semilogx(k_eval, p0_eval1, c='C0', lw=3, label='A', ls='--')
plt.semilogx(k_eval, p0_eval1, c='C2', lw=3, label='B', ls=':')
plt.axis([1e-1,12,0.73,1.04])
plt.yticks(fontsize=14)
plt.xticks(fontsize=14)
plt.legend()
plt.xlabel(r'$k$ (h/Mpc)', fontsize=14)
plt.ylabel(r'$\frac{P_{\rm DM+baryon}}{P_{\rm DM}}$', fontsize=21)
plt.subplot(122); plt.title('z=0.5')
plt.semilogx(BAH['z=0.5']['k'], BAH['z=0.5']['S'], '-', c='k', lw=5, alpha=0.2, label='BAHAMAS')
plt.semilogx(k_eval, p1_eval1, c='C0', lw=3, label='A', ls='--')
plt.semilogx(k_eval, p1_eval2, c='C2', lw=3, label='B', ls=':')
plt.axis([1e-1,12,0.73,1.04])
plt.yticks(fontsize=14)
plt.xticks(fontsize=14)
plt.xlabel(r'$k$ (h/Mpc)', fontsize=14)
plt.ylabel(r'$\frac{P_{\rm DM+baryon}}{P_{\rm DM}}$', fontsize=21)
plt.tight_layout()
plt.show()

CONTRIBUTING

If you find any bugs or unexpected behaviour in the code, please feel free to open a Github issue. The issue page is also good if you seek help or have suggestions for us.

References

[1] Schneider, A., Teyssier, R., Stadel, J., Chisari, N. E., Le Brun, A. M., Amara, A., & Refregier, A. (2019). Quantifying baryon effects on the matter power spectrum and the weak lensing shear correlation. Journal of Cosmology and Astroparticle Physics, 2019(03), 020. arXiv:1810.08629.

[2] Giri, S. K. & Schneider, A. (2021). Emulation of baryonic effects on the matter power spectrum and constraints from galaxy cluster data. Journal of Cosmology and Astroparticle Physics, 2021(12), 046. arXiv:2108.08863.