Installation

Prerequisites for basic installation

To install petitRADTRANS, without retrievals, you need to install:

  • Python 3.9+,

  • a fortran compiler, for example gfortran.

Running retrievals: read the MultiNest installation instructions.

Linux

On Linux, install Python and the fortran compiler with:

sudo apt-get install python python-pip gfortran

On some distributions, python may need to be replaced with python3.

Note

A general Python recommendation is to use a Python virtual environment such as venv or conda, to prevent potential conflicts.

Mac OS

Important

On Mac, it is highly recommended to use a Python virtual environment such as venv or conda, to prevent potential conflicts.

Using gfortran disk images

Warning

While using homebrew should be the preferred method for installing external libraries on Mac, alternative methods exist. Use them at your own risk.

François-Xavier Coudert’s github repository provides gfortran disk images (.dmg) with which you can install gfortran like any other program for Mac, through an installation wizard. Both Apple Silicon (M1, M2, M3) and Intel chip versions are available.

Windows

Native installation prerequisites

  1. Get a fortran compiler through, for example, MSYS2 or Visual Studio.

  2. Go to the Python website, then download and execute the Python installer.

Warning

It is not possible to run parallel (fast) retrievals with a native Windows installation (see the MultiNest section).

WSL-native dual installation

pRT can be installed both on the Windows and WSL sides. Files on WSL can be accessed from the Windows side using the path \\wsl.localhost\, and files on Windows can be accessed from the WSL side using /mnt (e.g., to get into “C:\Users” from WSL: cd /mnt/c/Users). Note however than accessing files across sides is slow.

Prerequisite for retrievals: MultiNest

Warning

Windows native installation: MultiNest retrievals, that are used by default in pRT, will not work as is on Windows. This is because MultiNest requires the LAPACK and OpenMPI libraries to function. Installing LAPACK on Windows can be a tedious process, and OpenMPI support on Windows has been discontinued, meaning that it is not possible to run MultiNest retrievals in parallel, increasing significantly computation times. This can be overcome by using WSL (see installation instructions above).

If you want to use pRT’s retrieval package, you need to install the PyMultiNest package:

  1. Follow the instructions provided on the PyMultiNest website.

  2. Link the resulting library files by including the MultiNest/lib/ to your LD_LIBRARY_PATH. This can be done by adding this line at the end of your environment setup file “.bash_profile”, “.bashrc”, or “.zshrc” (depending on your operating system and shell type):

    LD_LIBRARY_PATH=/path/to/MultiNest/lib:$LD_LIBRARY_PATH
    

Warning

Using Mac+Anaconda: see the troubleshooting section.

Pre-installation packages

Before starting the installation of pRT, make sure to install the following Python packages with:

pip install numpy meson-python ninja

On some distributions, pip may need to be replaced with pip3.

Installation of petitRADTRANS via pip install

To install pRT without retrievals via pip install, open a terminal and run:

pip install petitRADTRANS --no-build-isolation

Be sure to add the --no-build-isolation flag.

To be able to use the retrieval module, execute:

pip install petitRADTRANS[retrieval] --no-build-isolation

Compiling pRT from source

  1. Download petitRADTRANS from Gitlab, or clone it from GitLab via:

    git clone https://gitlab.com/mauricemolli/petitRADTRANS.git
    
  2. In the terminal, enter the petitRADTRANS folder.

  3. No retrievals: execute the following command in the terminal:

    pip install . --no-build-isolation
    
  4. With retrievals: execute the following command in the terminal:

    pip install .[retrievals] --no-build-isolation
    

Setting up the input_data directory

By default, pRT’s input files are downloaded into the <home>/petitRADTRANS/input_data directory, where <home> is your home directory (shortcut ~ in most OS). This can be changed by modifying the pRT config file. All of this is described more in the “Getting Started” notebook. Please note that the folder that pRT stores its opacities in has to be called input_data (but it can be placed wherever you want).

Note

pRT relies on data (opacities, stellar spectra, planet data, pre-calculated chemical abundances) to perform its calculations. Those data will be downloaded automatically as needed.

Alternatively, the data can be accessed and downloaded via Keeper here. The planet data are fetched from the Nasa Exoplanet Archive.

Testing the installation

Open a new terminal window. Then open python and type:

from petitRADTRANS.radtrans import Radtrans
radtrans = Radtrans(line_species=['CH4'])

If you have not already manually downloaded the CH4 correlated-k opacities, this should trigger the download of the opacity file.

The last lines of the output should be:

Loading Radtrans opacities...
 Loading line opacities of species 'CH4' from file '/path/to/input_data/opacities/lines/correlated_k/CH4/12C-1H4/12C-1H4__YT34to10.R1000_0.3-50mu.ktable.petitRADTRANS.h5'... Done.
 Successfully loaded all line opacities
Successfully loaded all opacities

The warning about the pressure can be ignored.

Troubleshooting the installation

Temporary directory issue

When importing Radtrans, you may see one of those two errors:

# For a pip install
ModuleNotFoundError: No module named 'petitRADTRANS.<fortran_extension>'

# For an editable pip install
FileNotFoundError: [Errno 2] No such file or directory: '/a/temporary/directory/overlay/bin/ninja'

The issue is often caused by your setup installing the fortran extensions inside a temporary directory, that is then automatically removed. Try these fixes in that order:

  1. Ensure that you added the --no-build-isolation flag to the installation command. This should fix the issue in almost all cases.

  2. Ensure that all the installing elements of your setup (pip, conda, fortran compiler, etc.) are up-to-date and installed cleanly.

  3. If you are on Mac, and use Homebrew, try first to execute brew upgrade, brew update, then to follow the instructions of brew doctor, before re-trying the installation.

  4. If you are on Mac, and do not use Homebrew, the error may be related with your setup. Carefully check for libraries versions, dependencies, and duplicate installations.

  5. In last resort, you can add the --no-clean flag to the installation command. Beware however: this will create a temporary directory that will not be removed from your system, taking space on your disk. Each new installation with this flag will create a new temporary directory, but will not remove the previous one. You may need to perform manual cleaning to free space on your disk.

Mac+Anaconda known issue with MultiNest

Linking the MultiNest libraries the usual way may not work on a Mac when using anaconda. In that case you may also need to copy the MultiNest/lib/* files generated during the installation into the lib folder that your Python binary sees. This folder should be called something like /opt/miniconda3/envs/name_of_your conda_environment/lib/. You may also need the conda version of the mpi4py package, which must be installed with:

conda install mpi4py

In case of troubles, if you use Homebrew, executing brew upgrade, brew update, then following the instructions of brew doctor may help. If you do not use Homebrew, the error may be related with your setup. Carefully check for libraries versions, dependencies, and duplicate installations.

Other issues

You can take a look at the solved issues here. If you do not find an helpful answer there, do not hesitate to open a new issue.