A library for implicitly solving parabolic PDEs
A package for solving parabolic differential equations, in particular the Schrodinger equation and the Gross-Pitaevskii equation using the Crank Nicolson method.
To install the package from PyPI use pip install sinope, or download the source distribution sinope-0.1.tar.gz and run pip install sinope-0.1.tar.gz.
Alternatively if you prefer to keep the code in its own directory, download the github repository and compile the extension module with make.
To run tests use sinope_run_tests.py. This requires the package hypothesis. which can be installed with pip install hypothesis.
Examples for how to use the code are available in scripts. The user code should define a grid class which inherits from one of Grid1D or Grid2D. The user then needs to define four methods:
- initialise_solvers(self, **kwargs):
This is called when the grid is initialised using with the keyword arguments that are passed to the grid constructor. In this method the chosen solvers (to be found in the solvers module) should be constructed.
- set_time_step(self, dt):
This method needs to initialise all of the needed solvers with a common step size, which is dt. This method should have some mechanism to choose the step size if it is called with dt as None. In which case the longest_time_step methods of the solver classes should be used.
This advances the solution by a single time step using the previously defined solvers.
Additionally the method n_components(self) can be overridden (default function returns 1) if the solution u has more than one component. This allows for multiple (possibly dependent) variables to be evolved together. Note that each component is necessarily real, so if one of the variables is complex then it requires two components (one for each of the real and imaginary parts).
The solvers are defined in the solver module. The available solvers are Explicit (for a real parabolic PDE) using a FTCS explicit scheme. A Crank Nicolson implicit solver also for real PDEs is also available using Cholesky factorisation (cholesky module). This is implicit and is unconditionally stable while the explicit one has a strict stability condition. Crank Nicolson is therefore faster overall, especially on large grids. On the other hand it is less easily parallelised than an explicit scheme.
The solvers are entirely one dimensional and are intended to be used on the 2D grid with dimensional splitting. See the scripts folder for examples.
The tst package implements the multiplying of TST (tridiagonal symmetric Toeplitz) matrices by vectors using c++ and a swig interface. This is significantly faster than the equivalent numpy code since it cannot be implemented using elementwise operations, but requires slicing.