GWSurrogate is an easy-to-use interface to gravitational wave surrogate models.

Surrogates provide a fast and accurate evaluation mechanism for gravitational waveforms, which would otherwise be found through solving differential equations. These equations must be solved in the ``building" phase, which was performed using other codes.

If this package contributes to a project that leads to a publication, please acknowledge this by citing the relevant paper(s). Please see the How to Cite section at the bottom of this README file.

gwsurrogate is available at https://pypi.python.org

gwsurrogate requires:

1. gwtools. If you are installing gwsurrogate with pip you will automatically get gwtools. If you are installing gwsurrogate from source, please see https://bitbucket.org/chadgalley/gwtools/

2. gsl. For speed, the long (hybrid) surrogates use gsl's spline function. To build gwsurrogate you must have gsl installed. Fortunately, this is a common library and can be easily installed with a package manager.

Note that at runtime (ie when you do import gwsurrogate) you may need to let gsl know where your BLAS library is installed. This can be done by setting your LD_PRELOAD or LD_LIBRARY_PATH environment variables. A relevant example:

>>> export LD_PRELOAD=~/anaconda3/envs/python27/lib/libgslcblas.so

The python package pip supports installing from PyPI (the Python Package Index). gwsurrogate can be installed to the standard location (e.g. /usr/local/lib/pythonX.X/dist-packages) with

>>> pip install gwsurrogate

If there is no binary/wheel package already available for your operating system, the installer will try to build the package from the sources. For that, you would need to have gsl installed already. The installer will look for GSL inside /opt/local/. You may provide additional paths with the CPPFLAGS and LDFLAGS environment variables.

In the case of an homebrew installation, you may install the package like this:

>>> export HOMEBREW_HOME=`brew --prefix`
>>> 
>>> export CPPFLAGS="-I$HOMEBREW_HOME/include/"
>>> export LDFLAGS="-L$HOMEBREW_HOME/lib/"
>>> pip install gwsurrogate

gwsurrogate is on conda-forge, and can be installed with

>>> conda install -c conda-forge gwsurrogate

Certain gwsurrogate modules are implemented as C-extensions and require NumPy’s C-API headers at build time. By default, pip install . uses the NumPy 2.x headers (as pinned in pyproject.toml) but produces binaries that remain compatible with NumPy >=1.7 at runtime. If you explicitly need to build against NumPy >=1.7 headers, update the NumPy requirement in pyproject.toml before installing.

To create a Conda environment with Python 3.11 and NumPy < 2.0:

conda create -n myenv python=3.11 "numpy<2.0"

First, please ensure you have the necessary dependencies installed (see above). Next, git clone this project, to any folder of your choosing. Then run

git submodule init
git submodule update

For a "proper" installation, run the following commands from the top-level gwsurrogate folder containing setup.py

>>> python -m pip install .            # option 1
>>> python -m pip install --editable . # option 2

where the "--editable" installs an editable (development) project with pip. This allows your local code edits to be automatically seen by the system-wide installation.

Explore our Jupyter Notebooks for a comprehensive overview of individual models and the user-level API. For an introductory explanation of the surrogate modeling methodology used in GWSurrogate, check out these videos:

• Introduction to the GWSurrogate package
• Overview of surrogate modeling methodology

To get a list of all available surrogate models, do:

>>> import gwsurrogate
>>> gwsurrogate.catalog.list()
>>> gwsurrogate.catalog.list(verbose=True)      # Use this for more details

The most up-to-date models trained on numerical relativity data are listed below, along with links to example notebooks.

• NRSur7dq4: For generically precessing BBHs, trained on mass ratios q≤4. Paper: arxiv:1905.09300.
• NRHybSur3dq8: For nonprecessing BBHs, trained on mass ratios q≤8. Paper: arxiv:1812.07865.
• NRHybSur2dq15: For nonprecessing BBHs, trained on q≤15, chi1≤0.5, chi2=0. Paper: arxiv:2203.10109.
• NRHybSur3dq8_CCE: For nonprecessing BBHs, trained on CCE (Cauchy-characteristic evolution) waveforms of mass ratios q≤8. Unlike all of the other models, NRHybSur3dq8_CCE includes memory effects. Paper: arxiv:2306.03148.

The most up-to-date models are trained on point-particle blackhole perturbation data and calibrated to numerical relativity (NR) in the comparable mass regime.

• BHPTNRSur1dq1e4: Nonspinning BBHs, trained on mass ratios q≤10000 and harmonics up to ell=10. Paper: arxiv:2204.01972.

The most up-to-date effective one body surrogate models.

• SEOBNRv4PHMSur: precessing binary black hole with 2<=ell<=5 modes in inertial frame. Trained on mass ratios q ≤20. Paper: arxiv:2204.01972.

Pick a model, let's say NRSur7dq4 and download the data. Note this only needs to be done once.

gwsurrogate.catalog.pull('NRSur7dq4')       # This can take a few minutes

Load the surrogate, this only needs to be done once at the start of a script

sur = gwsurrogate.LoadSurrogate('NRSur7dq4')

q = 4                           # mass ratio, mA/mB >= 1.
chiA = [-0.2, 0.4, 0.1]         # Dimensionless spin of heavier BH
chiB = [-0.5, 0.2, -0.4]        # Dimensionless of lighter BH
dt = 0.1                        # timestep size, Units of total mass M
f_low = 0                       # initial frequency, f_low=0 returns the full surrogate

# optional parameters for a precessing surrogate models
precessing_opts = {'return_dynamics': True}

# h is dictionary of spin-weighted spherical harmonic modes
# t is the corresponding time array in units of M
# dyn stands for dynamics, do dyn.keys() to see contents
t, h, dyn = sur(q, chiA, chiB, dt=dt, f_low=f_low, precessing_opts=precessing_opts)

There are many more options, such as using MKS units, returning the polarizations instead of the modes, etc. Read the documentation for more details.

help(sur)

Jupyter notebooks located in tutorial/website give a more comprehensive overview of individual models.

You can also evaluate any gwsurrogate model through PyCBC’s waveform API.

1. Install

   pip install gwsurrogate pycbc

2. Example Usage

   from pycbc.waveform import get_td_waveform
   import gwsurrogate as gws
   import matplotlib.pyplot as plt
   
   # PyCBC waveform
   hp_pcbc, hc_pcbc = get_td_waveform(approximant="GWS-NRSur7dq4",mass1=30, mass2=30, delta_t=1.0/2048,f_lower=20.0,f_ref=20.0)
   
   # gwsurrogate waveform
   sur = gws.LoadSurrogate("NRSur7dq4")
   t, h, dynamics = sur(q=1.0, chiA0=[0, 0, 0], chiB0=[0, 0, 0], M=60.0, dt=1.0/2048, f_low=20.0, dist_mpc=1.0, units="mks", inclination=0.0, phi_ref=0.0, f_ref=20.0)
   
   # Plot comparison
   plt.plot(hc_pcbc.sample_times, hp_pcbc, 'b', label='h₊ via PyCBC')
   plt.plot(t, h.real, 'r--', label='h₊ via gwsurrogate')
   plt.xlabel("Time [s]")
   plt.ylabel("Strain")
   plt.legend()
   plt.show()

3. Supported Approximants See the full list of PyCBC entry-point names in setup.py.

If you have git cloned this project and installed (and intalled it using the --editable option), its a good idea to run some regression tests.

>>> cd test                              # move into the folder test
>>> python download_regression_models.py # download all surrogate models to test
>>> python test_model_regression.py      # (optional - if developing a new test) generate regression data locally on your machine
>>> cd ..                                # move back to the top-level folder
>>> pytest                               # run all tests
>>> pytest -v -s                         # run all tests with high verbosity

We welcome contributions! Here's how you can get involved:

1. Report Bugs or Suggest Enhancements:
   Use the GitHub issue tracker to report bugs or suggest new features. Before submitting, consider browsing through existing issues to see if your concern has already been addressed. A developer will respond to issues that are opened on GitHub.

2. Contribute Code:
   We use the fork and pull request model for code contributions. Fork the repository, make your changes, and submit a pull request. We use Ruff for linting and auto-fixes. If you’re on VS Code, install the Ruff extension.

Please ensure you follow our Code of Conduct when contributing.

If this package contributes to a project that leads to a publication, please acknowledge this by citing the GWSurrogate article in JOSS. The paper has the following bibtex entry

@article{Field:2025isp,
    author = "Field, Scott E. and Varma, Vijay and Blackman, Jonathan and Gadre, Bhooshan and Galley, Chad R. and Islam, Tousif and Mitman, Keefe and Pürrer, Michael and Ravichandran, Adhrit and Scheel, Mark A. and Stein, Leo C. and Yoo, Jooheon",
    title = "{GWSurrogate: A Python package for gravitational wave surrogate models}",
    eprint = "2504.08839",
    archivePrefix = "arXiv",
    primaryClass = "astro-ph.IM",
    doi = "10.21105/joss.07073",
    journal = "J. Open Source Softw.",
    volume = "10",
    number = "107",
    pages = "7073",
    year = "2025"
}

Please also cite the relevant paper(s) describing your specific model. This information can be found by doing

>>> import gwsurrogate
>>> gwsurrogate.catalog.list(verbose=True) 

This package is based upon work supported by the National Science Foundation under PHY-1316424, PHY-1208861, and PHY-1806665.

Any opinions, findings, and conclusions or recommendations expressed in gwsurrogate are those of the authors and do not necessarily reflect the views of the National Science Foundation.