Simulating carbon nanostructures.

Short Summary

The Hubbard Model is an approximate model that describes fermions on a lattice that have a contact interaction. The Hamiltonian includes a nearest-neighbor "hopping", in which the fermions can go from one site to another, and an interaction that two fermions feel if they're on the same site. In solid-state and condensed matter physics, sites can represent atoms in some material, while the fermions are the electrons in the outer-most shells.

It's thought that the Hubbard model approximates carbon nanostructures, such as graphene, nanotubes, and fullerenes, very well. This code is intended to allow us to study the Hubbard model, focusing on carbon nanostructures. However, it has enough generality to allow the study of the Hubbard model on arbitrary graphs. Some graphs have special features—the hexagonal lattice of graphene is bipartite, for example. In some circumstances, these special features may unlock numerical speedup.

Requirements

C++

C++14 compiler (tested with recent versions of gcc and clang)
blaze
BLAS/LAPACK

Python

Python 3 version 3.7 or later
Pybind11 >= 2.2.4
numpy
SciPy library
h5py
scikit-learn (optional)
matplotlib (optional)

Setup

TL;DR

See other sections for an explanation and details.

python3 setup.py configure --build-type=RELEASE --blaze=<path-to-blaze>
python3 setup.py build -j <number-of-threads>
python3 setup.py install

Configuring

This project is installable via pip. However since the C++ extension is rather complex, you need to configure the build first by running

python3 setup.py configure

with suitable arguments. The supported arguments are (run python3 setup.py configure --help):

compiler: Choose a C++ compiler. isle is tested with current versions of gcc and clang. The intel compiler is not supported because of a bug in the compiler.
build-type: Set the CMake build type. Can be DEVEL (default), RELEASE, DEBUG
blaze: Point to blaze installation (has to contain blaze/Blaze.h)
blaze-parallelism: Select parallelism used by blaze. Allowed values are NONE (default), OMP (OpenMP), CPP (C++ threads)
blas-vendor: Select vendor of BLAS/LAPACK library. Supported values are:
- Generic - Reference implementation (default)
- Intel10_32, Intel10_64lp, Intel10_64lp_seq, Intel - MKL, see https://cmake.org/cmake/help/v3.0/module/FindBLAS.html
parallel-blas: Specify this flag if the BLAS/LAPACK implementation is parallelized. Otherwise blaze might parallelize on top of it which can slow down the program.

This does not actually run CMake to configure the C++ build but merely performs some rudimentary checks and saves the parameters.

Note that specifying the compiler or build type when running python3 setup.py build does not work; they need to be set when configuring.

Building

For users: Compile and install the package using

python3 setup.py build -j <number-of-threads>

python3 setup.py install [--user]

For developers: Compile and install in development mode using

python3 setup.py develop [--user]

And just re-run the command after changing the code.

Testing

Unit tests are automatically compiled when using the above commands (sorry for the extra compilation time, we are working on a way around that). They only work however, if you are in development mode! You can run them via

python3 setup.py test

Keep in mind that some tests take a long time because they need to process fairly big matrices in an inefficient way to make sure the efficient algorithms are correct.

Documentation

The documentation is available online at https://jl-wynen.github.io/isle/

You can generate it locally by running

python3 setup.py doc

to generate the source code documentation using Doxygen. Then open docs/html/index.html in a browser.

Note that the C++ interface is not identical to the Python interface. Documentation of the Python side is incomplete as of now.

There is additional documentation available under docs/algorithm; run make to generate it. This needs LaTeX to compile PDFs.