reyna 1.1.0

A minimal overhead, vectorised, polygonal discontinuous Galerkin finite element library.

reyna

A lightweight Python package for solving partial differential equations (PDEs) using polygonal discontinuous Galerkin finite elements, providing a flexible and efficient way to approximate solutions to complex PDEs.

Features

• Support for various polygonal element types (e.g., triangles, quadrilaterals, etc.)
• Easy-to-use API for mesh generation, assembly, and solving PDEs.
• High performance with optimized solvers for large-scale problems.
• Supports both linear and nonlinear equations.
• Extensible framework: easily integrate custom element types, solvers, or boundary conditions.

Installation

You can install the package via pip. First, clone the repository and then install it using pip:

Install from PyPI:

pip install reyna

Install from source:

pip install git+https://github.com/mattevs24/reyna.git

Example Usage

Create a Simple Mesh

A simple example to begin with is the RectangleDomain object. This requires just the bounding box as an input. In this case, we consider the unit square; $[0, 1]^2$. We then use poly_mesher to generate a bounded Voronoi mesh of the domain. This uses Lloyd's algorithm, which can produce edges that are machine precision in length. To avoid this for benchmarking and other critical purposes, use the cleaned keyword, set to True.

import numpy as np

from reyna.polymesher.two_dimensional.domains import RectangleDomain
from reyna.polymesher.two_dimensional.main import poly_mesher

domain = RectangleDomain(bounding_box=np.array([[0, 1], [0, 1]]))
poly_mesh = poly_mesher(domain, max_iterations=10, n_points=1024)

Generating the Geometry Information

The DGFEM code requires additional information about the mesh to be able to run, including which edges are boundary edges and their corresponding normals as well as information on a given subtriagulation to be able to numerically integrate with the required precision. This is done using the DGFEMgeometry function.

from reyna.geometry.two_dimensional.DGFEM import DGFEMGeometry

geometry = DGFEMGeometry(poly_mesh)

Defining the Partial Differential Equation

To define the PDE, we need to call the DGFEM object. We then add data in the form of the general coefficients for a (up-to) second order PDE of the form

$$ -\nabla\cdot(a\nabla u) + b\cdot\nabla u + cu = f $$

where $a$ is the diffusion tensor, $b$ is the advection vector, $c$ is the reation functional and $f$ is the forcing functional. All of these functions must be able to take in a (N, 2) array of points and output tensors of the correct shape; (N, 2, 2), (N, 2), (N,) and (N,) respectively. An example is given

diffusion = lambda x: np.repeat([np.identity(2, dtype=float)], x.shape[0], axis=0)
advection = lambda x: np.ones(x.shape, dtype=float)
reaction = lambda x: np.pi ** 2 * np.ones(x.shape[0], dtype=float)
forcing = lambda x: np.pi * (np.cos(np.pi * x[:, 0]) * np.sin(np.pi * x[:, 1]) +
                             np.sin(np.pi * x[:, 0]) * np.cos(np.pi * x[:, 1])) + \
                    3.0 * np.pi ** 2 * np.sin(np.pi * x[:, 0]) * np.sin(np.pi * x[:, 1])

solution = lambda x: np.sin(np.pi * x[:, 0]) * np.sin(np.pi * x[:, 1])

We use the solution function here as the boundary conditions for the solver.

Adding data and Assembly

We can now call the solver, add the data and assemble.

from reyna.DGFEM.two_dimensional.main import DGFEM

dg = DGFEM(geometry, polynomial_degree=1)

dg.add_data(
    diffusion=diffusion,
    advection=advection,
    reaction=reaction,
    dirichlet_bcs=solution,
    forcing=forcing
    )

dg.dgfem(solve=True)

Setting the solve input to True generates the solution vector. If this is False, just the stiffness matrix and data vector are generated.

Visualize the solution

We also have a function to plot the data, plot_DG, but this is limited to polynomial degree 1 with limited support for polynomial degree 0. See the example below

from reyna.DGFEM.two_dimensional.plotter import plot_DG

plot_DG(dg.solution, geometry, dg.polydegree)

For the given example, we have the solution plot

Benchmarking

We have a benchmarking file that may be run availible in the main DGFEM directory. But we also provide an example of the code to be able to calculate yourself

def grad_solution(x: np.ndarray):
    u_x = np.pi * np.cos(np.pi * x[:, 0]) * np.sin(np.pi * x[:, 1])
    u_y = np.pi * np.sin(np.pi * x[:, 0]) * np.cos(np.pi * x[:, 1])

    return np.vstack((u_x, u_y)).T

dg_error, l2_error, h1_error = dg.errors(
    exact_solution=solution,
    div_advection=lambda x: np.zeros(x.shape[0]),
    grad_exact_solution=grad_solution
)

Often, the error rate is calcuated against the maximal cell diameter; the code for this is included in the DGFEM class under the h method as well as the DGFEMgeometry class under the h method (DGFEMgeometry additionally contains all the local values of h across the mesh).

h = dg.h
h = geometry.h

Note that in a purely advection/diffusion problem, some of the norms are unavailable and return a None value.

A more advanced Domain Example

There are many predefined domains in the reyna/polymesher/two_dimensional/domains folder including this more advanced CircleCircleDomain() domain;

Documentation

For detailed usage and API documentation, please visit our (soon to be) readthedocs. The above example covers most cases and the current docstrings are very thorough.

Contributing

We welcome contributions! To contribute:

Fork the repository. Create a new branch for your feature or bug fix. Write tests for your changes. Submit a pull request.

License

This project is licensed under the MIT License - see the LICENSE.md file for details.

Credits & Acknowledgements

This package was developed by mattevs24 during a PhD programme funded by the French Alternative Energies and Atomic Energy Commission. A Special thanks to the support of Ansar Calloo, Fraçois Madiot throughout the PhD so far. A further thank you to my interal supervisors Tristan Pryer and Luca Zanetti for their role in this project too and useful feedback on usability and support. Finally, a thank you to my partner Reyna who puts up with all this nonsense!

Upcoming Updates

There are many features that remain to add to this code! We hope to add support for the following features

• Mixed-type problems: support for multiple types of PDE on the same domain.
• Full readthedocs documentation to support the further developement and use of this package.