A short library that performs single Sérsic profile fits in python using Multinest as fitter.
First clone this repository with
git clone https://github.com/WilfriedMercier/MorphoWithMultinest.git
To install and run the code, first create a new conda environment using the project YAML file:
conda create --file environment.yaml
Activate the environment with
conda activate MorphoWithMultinest
Then use pip to install additional packages:
pip install corner pymultinest
To install MultiNest, one must first clone the Git repository, then compile the code as follows
git clone https://github.com/JohannesBuchner/MultiNest
cd Multinest/build
cmake ..
make
It the compilation fails, make sure
- that all the required dependencies are installed on the machine (cmake, git, gcc, gfortran, BLAS, LAPACK, ATLAS) - see the documentation for more information
- to add the following line to your
.bashrcfileexport CMAKE_POLICY_VERSION_MINIMUM=3.5if you have a modern version of cmake
Once cmake and make have successfully finished, the shared library is available in the directory Multinest/lib. The last step is to add the following line to your .bashrc file:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:pathToMultinest/MultiNest/lib
where you replace pathToMultinest with your path.
Once all the dependencies are installed, you can install the package with pip using the command
pip install -e .
An example of a Bayesian morphological fit is in examples/JADES-GS-z14-0 and can be run with
python run_fit.py configuration.yaml
All the information necessary for the fit (location of the input files, magnitude zero point, MultiNest sampling parameters, etc.) are given in the configuration file. The run_fit.py python script simply loads the main function from MorphoWithMultinest that starts the fit given the configuration file. Alternatively, one can do
python -c 'from MorphoWithMultinest import run; run("configuration.yaml")'
Note: The task taking most of the inference time is the psf convolution of each generated model with the psf. To speedup this process, by default and if possible, the code will try to use PyTorch in GPU mode. If there is no compatible GPU on the system, it will fallback on CPU mode and, if PyTorch is not installed, it will use Astropy instead. However, note that using PyTorch in CPU mode or Astropy significantly increases the inference time compared to the GPU mode.
Once the fit is finished, the fitter returns in the terminal the best fit results and its uncertainty as follows:
2026-05-18 15:50:59,880 - INFO xc_1 yc_1 mag_1 re_1 n_1 q_1 pa_1 mag_2 re_2 n_2
2026-05-18 15:50:59,880 - INFO ------------------------------------------------------------------------------------------------------------------------
2026-05-18 15:50:59,888 - INFO 48.336 49.107 26.134 5.35458 1.20678 0.579909 84.5067 26.3009 5.34319 0.954768
2026-05-18 15:50:59,888 - INFO 0.0264706 0.0160002 0.0221266 0.0946627 0.0401668 0.00780975 0.676399 0.0280832 0.13475 0.0732638
Then, it generates a PDF recap file showing the original images, best-fit models, and residuals:
Afterwards, it produces a plot showing the median profile in each band with the 1 and 2 sigma uncertainties around:
Finally, it produces a corner plot to check the covariance between parameters:
