The emulator is developed using the following packages:

• python 3.7

1. tensorflow==1.14.0 pip install tensorflow==1.14.0
2. gpflow=1.5.1 pip install gpflow==1.5.1
3. scikit-learn=0.21.3 pip install scikit-learn

To install the package in developer mode, clone the git repo locally and then use a pip development installation by running:

git clone https://github.com/LSSTDESC/mgemu.git
cd mgemu
python setup.py install --user   # Add --user flag to install it locally

from the top-level of the repo. From the same directory, run the test suite:

pytest

A script for running the emulator is also provided in `mgemu/notebooks/'.

from mgemu import emu

# LCDM parameters
h = 0.67 # See notes regarding the value of h below
Omh2=(h**2)*0.281
ns=0.971
s8=0.82

# Hu-Sawicki model parameters
fr0 = 1e-5
n=1

# Redshift
z= 0.3

pkratio, k = emu(Omh2=Omh2, ns=ns, s8=s8, fR0=fr0, n=n, z=z)

1. The emulator is trained within the range:

• 0.12 ≤ Ω_mh² ≤ 0.15
• 0.85 ≤ n_s ≤ 1.1
• 0.7≤ σ₈ ≤0.9
• 10⁻⁸ ≤ f_R0 ≤10⁻⁴
• 0 ≤ n ≤4
• 0 ≤ z ≤ 49

Hence the estimation is only reliable in this range. Outside of this, the GP performs extrapolation, and may provide unreliable results.

2. Emulator is trained with COLA simulations with h = 0.67, in agreement with Planck. For small differences of h, the emulator output does not vary much. However the user must exercise caution, and limit the use of h = 0.67 as much as possible.

3. Output from the mgemu.emu function will be the binned power spectra enhancement P_MG(k)/P_LCDM(k) and wavenumber k in 213 bins.

4. Although the k range of the emulator is up to 0 ≤ k ≤ 3.5 h/Mpc, we do not advocate the use of the emulator for small scales beyond k ~ 1.0 h/Mpc.

This software was developed within the LSST DESC using LSST DESC resources, and so meets the criteria given in, and is bound by, the LSST DESC Publication Policy for being a “DESC product”.

The list of released versions of this package can be found here, with the master branch including the most recent (non-released) development.

This repository is maintained by Georgios Valogiannis (@gvalgioa) and Nesar Ramachandra (@nesar). If you have comments, suggestions or questions, please write us an issue.

This is open source software, available for re-use under the modified BSD license.