An exact diagonalization implementation for computer simulations of quantum-dot cellular automata (QCA), a beyond-CMOS computing paradigm (Original paper proposing QCA).

The simulation is written in C++, but exposed as a Python extension module, and uses Eigen as a linear algebra library and Boost for the Python integration and unit testing.

Prebuilt OS X and Linux binary packages for use with the Anaconda Python distribution (or another distribution using the conda package manager) are available on Binstar. To use them, first install Anaconda, and then install the qca package from Binstar.

$ conda install -c https://conda.binstar.org/meznom qca

This will pull in all dependencies and install everything into the main Anaconda environment. One notable dependency is boost-python which is also provided by my Binstar channel.

Alternatively, you can create a new environment and install the qca package there.

# Add my Binstar channel permanently to your conda config (~/.condarc)
$ conda config --add channels http://conda.binstar.org/meznom
# Create a new environment 'qca' with a number of useful packages
$ conda create -n qca qca coma ipython ipython-notebook matplotlib numpy scipy
# Activate the 'qca' environment
$ source activate qca
# Do some work, as an example, start the IPython notebook
$ ipython notebook
# Deactivate the 'qca' environment
$ source deactivate
# To remove the 'qca' environment with all packages, you would run:
$ conda remove -n qca --all

Building from source can be a bit tricky. Particularly, I repeatedly had problems on OS X where it is not uncommon to have multiple versions of Python and different compilers installed (e.g. when using MacPorts or similar). On Linux compiling from source is more straightforward, in my experience.

Building requirements:

• cmake, version 2.8 or greater.
• Eigen, version 3.2.0 or greater.
• Boost, version 1.49 or greater
• Compiler and standard library supporting C++11
  • Recent versions of GCC are fine (I am using 4.8)
  • Same for recent versions of Clang (I am using 3.4)
• Python 2.7

Minimal build example:

$ mkdir Release
$ cd Release
$ cmake ..
$ make

Build example using a MacPorts g++ compiler, MacPorts Boost library, and the Anaconda Python distribution:

$ mkdir Release
$ cd Release
$ CXX=/opt/local/bin/g++-mp-4.8 \
  cmake \
  -DCMAKE_BUILD_TYPE=Release \
  -DPYTHON_LIBRARY=~/anaconda/lib/libpython2.7.dylib \
  -DPYTHON_INCLUDE_DIR=~/anaconda/include/python2.7 \
  ..
$ make

Build example explicitly setting all configurable paths and using a custom built Boost library, a custom Eigen installation, and the Anaconda Python distribution:

$ mkdir Release
$ cd Release
$ CXX=/usr/bin/clang++ \
  cmake \
  -DCMAKE_BUILD_TYPE=Release \
  -DPYTHON_LIBRARY=~/anaconda/lib/libpython2.7.dylib \
  -DPYTHON_INCLUDE_DIR=~/anaconda/include/python2.7 \
  -DBOOST_ROOT=~/anaconda/ \
  -DEIGEN_INCLUDE_DIR=~/anaconda/include/eigen3/ \
  ..
$ make

The make command builds the Python extension _qca.so which gets copied to the Python module directory, i.e. qca/_qca.so. To build and run the C++ unit tests:

$ cd Release
$ make check

To run the Python unit tests:

$ python -m unittest qca.test

Finally, to install the Python module system wide:

$ ./setup.py install

Or install the Python module in the home directory:

$ ./setup.py install --user

It is very important that the Python extension _qca.so is linked against the correct version of the Python and Boost libraries. Especially on OS X this has caused problems repeatedly. On OS X we can check which libraries are linked with

$ otool -L _qca.so

and we can change linked libraries with the install_name_tool. (On Linux ldd lists linked libraries.) The Boost Python library needs to be linked against the same Python library (again, we check with otool) and, obviously, that should be the Python version we are using. Lastly, Boost and this QCA exact diagonalization code need to be compiled with the same compiler (e.g. Clang versus GCC) and linked against the same standard library (libc++ versus libstdc++). If you get obscure compilation errors, that's likely the problem. If everything compiles correctly, but importing the Python module or running Python code fails with obscure errors, then likely something is not linked correctly.

Useful cmake variables:

• BASIS_NUMBER_OF_BITS. Number of bits for the storage of the basis. Default is 32.
• CMAKE_BUILD_TYPE. Can be either Release or Debug. Defaults to Release.
• EIGEN_INCLUDE_DIR. Where to find Eigen 3. Should be detected automatically.
• BOOST_ROOT. Where to find Boost. Should be detected automatically. You can also set BOOST_INCLUDE_DIR and BOOST_LIBRARY directly.
• USE_CXX_FLAGS_RELEASE. Compiler flags for Release build type. Reasonable defaults should be set automatically, depending on which compiler is used (GCC or Clang).
• USE_CXX_FLAGS_DEBUG. Compiler flags for Debug build type.

Conda recipes for qca and its dependencies are available in a separate repository.

Usage is via the qca python module. The unit tests in qca.test provide some examples.

The code is distributed under the two-clause BSD license. Have a look at the LICENSE file for details.

Burkhard Ritter (burkhard@ualberta.ca), May 2014