Using Mathematica From Python

- -

In which I show you how to programmatically interface with a Mathematica kernel from Python.

Mathematica, Python, and Scientific Computation

Mathematica is the flagship product of Wolfram Research. It’s a very sophisticated computer algebra system with the best notebook interface on the market if you ask me. It provides the computational power to WolframAlpha and is available on “thousands of colleges and universities in over 50 countries” according to Wolfram Research’s ad copy. Unfortunately Mathematica is not open source and comes with a hefty price tag, but your institution might have a site license that allows you to install it on your personal computer.

Python has gained a lot of ground in the scientific computation space. With packages like SciPy and SymPy and the comprehensive computer algebra system Sage, Python has access to very sophisticated computational abilities. You know what would be really great, though? If we could interact with a Mathematica kernel directly from our Python code.

The Good News

It turns out we can. Mathematica ships with something called MathLink that allows developers of C and C++ (and some other languages) to communicate with a Mathematical kernel programmatically. If you poke around in the Mathematica installation directory you will discover Python bindings for MathLink and a couple of example programs.

The Bad News—And a Fix

Unfortunately the MathLink Python bindings are undocumented, unsupported, and very outdated. Here I show you how I got it working on my system running Mac OS X Yosemite, Mathematica 10.0, and Python 2.7. It’s not too hard. I’ll assume you have the usual command line developer tools installed.

First we locate the necessary MathLink library, namely libMLi3.a. On my system it is found here:


I’m using the “AlternativeLibraries” version as per the README file located in that directory. We’ll also need the mathlink.h header file here:


Now we find the MathLink Python bindings and example code:


Since we will be editing these files, go ahead and copy them to a folder in which to work. That way, if we mess something up we have a backup of the original files.

One of those files, the file, uses Python’s distutils facility to compile and install the Python MathLink bindings extension to our Python environment’s site-packages directory. Edit the file to reflect your mathematica version (this might not be strictly necessary):

mathematicaversion = "10.0"

Now find your platform in the if-elif block (in my case “darwin”) and edit include_dirs and library_dirs to be the location of the mathlink.h header file and the library file libMLi3.a respectively. Here’s what that piece of code now looks like for my system:

elif('darwin', sys.platform)):
  setup(name="mathlink", version=pythonlinkversion,
        include_dirs = ["/Applications/"],
        library_dirs = ["/Applications/"],
        libraries = ["MLi3"],

The Python extention is mathlink.c. We need to make a minor adjustment to mathlink.c by defining MLINTERFACE to be 3 before we include the mathlink.h header file:

#include "mathlink.h"

That’s it for the Python bindings! Let’s compile and install:

$ python build
$ sudo python install

You should now be able to run the example script with the following command:

python -linkname "math -mathlink"

If you look inside you’ll find a strange attempt by the author to add the Mathematica bin directory to the path. Since the math command is already in my path, this line is unnecessary. The script sets up the mathlink connection using the -linkname argument. Consider these things as opportunities for improvement as you monkey with this example code.

Now go write some good Python code!