Spintrum is a library for fast and efficient spin dynamics simulations with an easy Python interface
Go to file
2017-09-02 22:01:45 +02:00
spintrum Updated licensing 2017-09-02 21:57:32 +02:00
src Updated licensing 2017-09-02 21:57:32 +02:00
CMakeLists.txt Removed scipy from requirements 2017-01-13 11:11:41 +01:00
get_resources.sh Added get_resources.sh that gets Polymath and OpenBLAS 2017-01-13 08:43:58 +01:00
get_version.py Version getting now doesn't require importing spintrum 2017-01-13 10:57:12 +01:00
LICENSE.txt Updated licensing 2017-09-02 21:57:32 +02:00
Makefile Added bash prefix to get_resources.sh call to make it work without making it executable 2017-01-14 19:06:02 +01:00
MANIFEST.in Added LICENSE to MANIFEST.in 2017-09-02 22:01:45 +02:00
README.md Updated README.md 2017-09-02 21:14:30 +02:00
setup.py Removed asking about optimization to support pip 9.0+, which has a non-verbose mode. 2017-01-13 18:04:49 +01:00

Spintrum

A high-performance library for spin simulations, written by Samer Afach as part of the Budker Group in University of Mainz

purpose

Spintrum creates an interface between a high-performance and low-level language (C/C++) and an easy to use and high level language (Python).

The library depends primarily on very good C/C++ programming to get fast result. No fancy physics is used in the library (so far).

How does it work?

As you will notice, the installation has two parts. The first one is building spintrum, which builds the shared library libspintrum.so and puts it in the package directory of Python, and the second part is installing that package into Python, which you can do if you wish. It is not mandatory.

Installation

Spintrum is supported only on Linux/Unix systems, and is primarily tested on Debian (Jessie and Stretch). For explanation, look at the FAQ below. If you'd like to compile Spintrum on operating systems other than Debian, your efforts to do it will be minimal, assuming you are not trying to do it on Windows.

Prerequisites
  • Linux/Unix (preferably Debian)
  • OpenBLAS (read below)
  • Polymath (read below)
  • cmake
  • Python 3.4+
  • g++ 4.8+
In Python, the following libraries are required
  • numpy
  • mpmath
  • matplotlib (for the examples only)
Do I need to compile OpenBLAS?

This depends on the performance you're trying to achieve. OpenBLAS, when compiled, creates binaries that are processor-specific, in order to achieve the best possible performance. In order to exploit this feature, it's recommended that you compile OpenBLAS.

The installer automatically compiles OpenBLAS for you. If you would not like to have OpenBLAS compiled for you, then use the Makefile by running make. It will ask whether you want to have OpenBLAS compiled for you. Then, you're free to use the directory spintrum independently as a package, or copy it yourself to your Python installation. You have to have OpenBLAS installed on your system to have this work (whether compiled for you or installed by your system); spintrum compilation will fail.

How to build?

In case you choose to build this manually, follow the following steps:

  • Clone the repository to some directory
  • Go to that directory with your terminal
  • Call make there.

At that point you will get a directory called spintrum, which is the Python package that you can use or copy.

Remember that the required software must be installed beforehand.

Do I need Polymath?

Definitely! Polymath is a simple C++ interface to BLAS. It was also written by the Spintrum team to separate the Maths library from the programming of Spintrum. The installer will get Polymath for you automatically from its repository.

How to install?

There are two ways:

  • Use pip to do the installation for you with the command:
# pip3 install git+https://git.afach.de/samerafach/Spintrum

Note: The # means that these commands have to be executed as root. Don't literally type that # in the terminal. This command will install some of the dependencies for you, too.

  • Or, you can use Spintrum as an independent package, by cloning the repository https://git.afach.de/samerafach/Spintrum yourself, and running make to build.
How to install missing prerequisites on Debian?

Simply run the following in the terminal (as superuser/root):

# apt-get install build-essential cmake g++ gcc gfortran python3 python3-pip

On Redhat versions of linux, find the equivalent package names and use yum or dnf to install the same dependencies.

Testing whether it works

After a successful installation, simply try the examples in the examples directory. They should all work, assuming you have everything installed correctly. The dependencies of the examples are different than Spintrum itself. For example, the examples use Matplotlib, which plots signals and spectra. You have to install this yourself. On Debian, you can sudo apt-get install python3-matplotlib to do that. Also for fitting scripts, the package scipy is used, which you can install using sudo pip3 install scipy.

Caveat about Jupyter

Spintrum works fine on Jupyter, except that the stdout and stderr output cannot be seen on Jupyter notebooks for technical reasons. So, if there's a message from the low-level C/C++, it won't be seen. Although unlikely, but in case you have an error and you can't figure out the reason, try running the script in terminal, and see whether the C/C++ interface is reporting an error.

FAQ

  • Q: Why is spintrum not supported on Windows?
  • A: Ignoring that Windows is way slower than Linux for calculations; Spintrum uses the well-known high-performance linear algebra library OpenBLAS. And since Python 3.5+ is compiled on Visual Studio 2015, and since Spintrum uses ctypes, which is the shared-library interface of Python, binary compatibility has to be achieved between Python and the compiled Spintrum library. On the other hand, OpenBLAS does not support compilation on Visual Studio with 64-bit (it supports only 32-bit), meaning that only small spin systems will be properly supported, since memory of 4+ GB cannot be used, which is not worth the effort, as too many changes have to be done to make Spintrum compatible with Windows. In the worst case, use a VMWare Workstation Virtual Machine on your Windows Machine, and you'll get the best performance.