Skip to content

Latest commit

 

History

History
522 lines (368 loc) · 16.1 KB

README.rst

File metadata and controls

522 lines (368 loc) · 16.1 KB

Cadabra

A field-theory motivated approach to computer algebra.

Kasper Peeters <[email protected]>

This repository holds the 2.x series of the Cadabra computer algebra system. It supersedes the 1.x series, which can still be found at https://github.com/kpeeters/cadabra.

Cadabra is a symbolic computer algebra system, designed specifically for the solution of problems encountered in quantum and classical field theory. It has extensive functionality for tensor computer algebra, tensor polynomial simplification including multi-term symmetries, fermions and anti-commuting variables, Clifford algebras and Fierz transformations, implicit coordinate dependence, multiple index types and many more. The input format is a subset of TeX. Both a command-line and a graphical interface are available, and there is a kernel for Jupyter.

Installation

Cadabra builds on Linux, macOS, OpenBSD, FreeBSD and Windows. Select your system from the list below for detailed instructions.

Binaries for most of these platforms are provided from the download page at https://cadabra.science/download.html, which links to https://github.com/kpeeters/cadabra2/releases/latest. These binaries are automatically generated on every release.

See Building Cadabra as C++ library for instructions on how to build the entire Cadabra functionality as a library which you can use in a C++ program.

See Building a Jupyter kernel for information on the Jupyter kernel for Cadabra sessions.

Linux (Debian/Ubuntu/Mint)

On Debian/Ubuntu you can install all that is needed with:

sudo apt install git cmake libpython3-dev python3-dev g++ libgmp3-dev \
      libgtkmm-3.0-dev libboost-all-dev libgmp-dev libsqlite3-dev uuid-dev  \
      python3-matplotlib python3-mpmath python3-sympy python3-gmpy2

(on Ubuntu 14.04 you need to replace cmake with cmake3 and also install g++-4.9; get in touch if you don't know how to do this). On older systems you may want to install sympy using sudo pip3 install sympy, but that is discouraged in general.

This is the development platform and issues are typically first fixed here. You can use either g++ or the clang++ compiler to build. You need to clone the cadabra2 git repository (if you download the .zip file you will not have all data necessary to build). So first do:

git clone https://github.com/kpeeters/cadabra2

Building is then done with the standard:

cd cadabra2
mkdir build
cd build
cmake ..
make
sudo make install

This will produce the command line app cadabra2 and the Gtk notebook interface cadabra2-gtk. You can also find the latter in the 'Education' menu.

Linux (Fedora 24 and later)

Fedora 24 is the first Fedora to have Python 3; you can build Cadabra using Python 2 but you are strongly encouraged to upgrade. The Fedora platform receives less testing so please get in touch if you run into any issues. You can use either g++ or the clang++ compiler.

Install the dependencies with:

sudo dnf install git python3-devel make cmake gcc-c++ \
     gmp-devel libuuid-devel sqlite-devel \
     gtkmm30-devel boost-devel \
     texlive python3-matplotlib \
     python3-pip
sudo pip3 install sympy

You need to clone the cadabra2 git repository (if you download the .zip file you will not have all data necessary to build). So first do:

git clone https://github.com/kpeeters/cadabra2

Building is then done with the standard:

cd cadabra2
mkdir build
cd build
cmake ..
make
sudo make install

This will produce the command line app cadabra2 and the Gtk notebook interface cadabra2-gtk. You can also find the latter when searching for the 'Cadabra' app from the 'Activities' menu.

Linux (CentOS/Scientific Linux)

On CentOS/Scientific Linux you need to activate The Software Collections (SCL) and Extra Packages for Enterprise Linux (EPEL) to get access to a modern C++ compiler, Python3 and all required build tools.

On CentOS first do:

sudo yum install centos-release-scl epel-release

On Scientific Linux the equivalent is:

sudo yum install yum-conf-softwarecollections epel-release

Now install all build dependencies with:

sudo yum install devtoolset-7 rh-python36 cmake3 \
              gmp-devel libuuid-devel sqlite-devel \
         gtkmm30-devel boost-devel git \
              texlive python-matplotlib

You need to enable the Python3 and C++ compiler which you just installed with:

scl enable rh-python36 bash
scl enable devtoolset-7 bash

(note: do not use sudo here!).

You also need to install sympy by hand:

sudo pip3 install sympy

Now need to clone the cadabra2 git repository (if you download the .zip file you will not have all data necessary to build):

git clone https://github.com/kpeeters/cadabra2

Building is then done with the standard:

cd cadabra2
mkdir build
cd build
cmake3 ..
make
sudo make install

This will produce the command line app cadabra2 and the Gtk notebook interface cadabra2-gtk. You can also find the latter in the 'Education' menu.

Linux (openSUSE)

For openSUSE (tested on 'Leap 15.2', probably also fine with minor changes for 'Tumbleweed') you first need to install the dependencies with:

sudo zypper install --no-recommends git cmake python3-devel gcc-c++ \
              gmp-devel libuuid-devel sqlite-devel \
              gtkmm3-devel  \
              texlive python3-matplotlib \
              python3-sympy \
              libboost_system1_71_0-devel libboost_filesystem1_71_0-devel \
              libboost_date_time1_71_0-devel libboost_program_options1_71_0-devel

You can get away with less than the full texlive.

This platform receives less testing so please get in touch if you run into any issues. You need to clone the cadabra2 git repository (if you download the .zip file you will not have all data necessary to build). So first do:

git clone https://github.com/kpeeters/cadabra2

Building is then done with the standard:

cd cadabra2
mkdir build
cd build
cmake ..
make
sudo make install

This will produce the command line app cadabra2 and the Gtk notebook interface cadabra2-gtk.

Linux (Arch/Manjaro)

The package for Arch Linux is cadabra2 https://aur.archlinux.org/packages/cadabra2/ Building and installing (including dependencies) can be accomplished with:

yay -S cadabra2

Alternatively use makepkg:

git clone https://aur.archlinux.org/cadabra2.git
cd cadabra2
makepkg -si

Please consult the Arch Wiki https://wiki.archlinux.org/index.php/Arch_User_Repository#Installing_packages for more information regarding installing packages from the AUR.

Linux (Solus)

Support for Solux Linux is experimental. To build from source on Solus Linux, first install the dependencies by doing:

sudo eopkg install -c system.devel
sudo eopkg install libboost-devel gmp-devel libgtkmm-3-devel
sudo eopkg install sqlite3-devel texlive python3-devel
sudo eopkg install git cmake make g++

Then configure and build with:

cd cadabra2
mkdir build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX=/usr
make
sudo make install

This installs below /usr (instead of /usr/local on other platforms) because I could not figure out how to make it pick up libraries there.

Any feedback on these instructions is welcome.

OpenBSD

Install the dependencies with:

pkg_add git cmake boost python-3.6.2 gtk3mm gmp gmpxx texlive_texmf-full py3-sympy

We will build using the default clang-4.0.0 compiler; building with the alternative g++-4.9.4 leads to trouble when linking against the libraries added with pkg_add.

Configure and build with:

cd cadabra2
mkdir build
cd build
cmake -DENABLE_MATHEMATICA=OFF ..
make
su
make install

The command-line version is now available as cadabra2 and the notebook interface as cadabra2-gtk.

Any feedback on this platform is welcome as this is not our development platform and testing is done only occasionally.

FreeBSD

The recommended way to install Cadabra is through:

pkg install cadabra2

It is also possible to build and install Cadabra from the port:

cd /usr/ports/math/cadabra2 && make install clean

The command-line version is now available as cadabra2 and the notebook interface as cadabra2-gtk.

Any feedback on this platform is welcome as this is not our development platform.

macOS

Cadabra builds with the standard Apple compiler, on both Intel and Apple silicon, but you do need a number of packages from Homebrew (see https://brew.sh). Install the required dependencies with:

brew install cmake boost gmp python3
brew install pkgconfig
brew install gtkmm3 adwaita-icon-theme
pip3 install sympy gmpy2

If the lines above prompt you to install XCode, go ahead and let it do that.

You also need a TeX installation such as MacTeX, https://tug.org/mactex/ . Any TeX will do, as long as 'latex' and 'dvipng' are available, so you simply do:

brew install mactex

Make sure to install TeX before attempting to build Cadabra, otherwise the Cadabra style files will not be installed in the appropriate place. Make sure 'latex' works from the terminal in which you will build Cadabra.

You can build against an Anaconda Python installation (in case you prefer Anaconda over the Homebrew Python); cmake will automatically pick this up if available.

You need to clone the cadabra2 git repository (if you download the .zip file you will not have all data necessary to build). So do:

git clone https://github.com/kpeeters/cadabra2

After that you can build with the standard:

cd cadabra2
mkdir build
cd build
cmake -DENABLE_MATHEMATICA=OFF ..
make
sudo make install

(note the -DENABLE_MATHEMATICA=OFF in the cmake line above; the Mathematica scalar backend does not yet work on macOS).

This will produce the command line app cadabra2 and the Gtk notebook interface cadabra2-gtk.

Feedback from macOS users is very welcome because this is not the main development platform.

Windows

On Windows compilation is easiest by using the MSYS2 system, as their gtkmm-3.0 packages just work and the whole system can be driven from the command line. We used to build Cadabra using the vcpkg packages, but they no longer provide packages for gtkmm-3.0, and in general the lack of binary packages means that build times are on the order of many, many hours, instead of just a few minutes with MSYS2. More info on building and packaging gtk apps on windows at https://www.gtk.org/docs/installations/windows/.

Install MSYS2 from https://www.msys2.org and start a UCRT64 shell. First update with (if you don't do this you may end up not being able to install some of the required packages due to version conflicts):

pacman -Suy

Then install a compiler and the dependencies of Cadabra with:

pacman -S mingw-w64-ucrt-x86_64-gcc
pacman -S mingw-w64-ucrt-x86_64-gtkmm3
pacman -S mingw-w64-ucrt-x86_64-boost
pacman -S mingw-w64-ucrt-x86_64-sqlite3
pacman -S mingw-w64-ucrt-x86_64-cmake
pacman -S mingw-w64-ucrt-x86_64-python-matplotlib
pacman -S mingw-w64-ucrt-x86_64-python-sympy
pacman -S mingw-w64-ucrt-x86_64-osslsigncode
pacman -S git

Checkout Cadabra and build:

git clone https://github.com/kpeeters/cadabra2
cd cadabra2
mkdir build
cd build
cmake ..
ninja
ninja install

This will leave an installation in Program Files (x86)/Cadabra, from where you can start cadabra2-gtk.

To build an installer, simply run cpack after having built Cadabra.

Building a Jupyter kernel

As of version 2.3.4 the standard build process (as described above) also creates a Jupyter kernel, which is written in Python on top of ipykernel (thanks to Fergus Baker). This should work on most platforms out-of-the-box; you do not need to do anything else. The Jupyter kernel allows you to use Cadabra notation inside a Jupyter notebook session.

The distribution also still contains code for the 'old' Jupyter kernel, which is written in C++ on top of xeus. Building this kernel is more complicated mainly because of this dependency, and there is not much of an advantage over the Python kernel; it's mainly left in the tree for future reference, For full instructions on how to build the old xeus-based kernel, see https://github.com/kpeeters/cadabra2/blob/master/JUPYTER.rst.

Creating an AppImage

When building Cadabra for bundling as an AppImage, the GUI will be configured to use MicroTeX (https://github.com/NanoMichael/MicroTeX) for typesetting (this dependency is included in the Cadabra source). MicroTeX is a rendering library which does not rely on an existing LaTeX installation. Configure and build with:

cmake -DAPPIMAGE_MODE=ON -DCMAKE_INSTALL_PREFIX=/usr ..
make
make install DESTDIR=AppDir

This installs everything in the AppDir folder ready for packaging. Then run:

make appimage

to create the AppImage itself. If you run into trouble with this, please first consult the comments in the top-level CMakeLists.txt file about linuxdeploy and friends.

Tutorials and other help

Please consult https://cadabra.science/ for tutorial-style notebooks and all other documentation, and https://cadabra.science/doxygen/html/ for doxygen documentation of the current master branch. The latter can also be generated locally; you will need (on Debian and derivatives):

sudo apt-get install doxygen libjs-mathjax

For any questions, please contact [email protected] .

Building Cadabra as C++ library

If you want to use the functionality of Cadabra inside your own C++ programs, you can build Cadabra as a shared library. To do this:

mkdir build
cmake -DBUILD_AS_CPP_LIBRARY=ON ..
make
sudo make install

There is a sample program simple.cc in the c++lib directory which shows how to use the Cadabra library.

Special thanks

Special thanks to José M. Martín-García (for the xPerm canonicalisation code), James Allen (for writing much of the factoring code), Dominic Price (for the meld algorithm implementation, many additions to the notebook interface, the conversion to pybind and the Windows port), Fergus Baker (for the new Jupyter kernel), Isuru Fernando (for the Conda packaging), the Software Sustainability Institute and the Institute of Advanced Study. Thanks to the many people who have sent me bug reports (keep 'm coming), and thanks to all of you who use Cadabra, sent feedback or cited the Cadabra papers.

Licenses

Cadabra itself is licensed under the GPL-3.0. It includes some dependencies which have the following licenses: