Skip to content

jesserobertson/multigrid

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

29 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

libmultigrid

Author: Jesse Robertson, Australian National University Date: 17 October 2011 URL: http://github.com/jesserobertson/multigrid Email: my name with domain csiro.au

This is a multigrid solver for solving elliptic PDEs using finite differences on a rectangular grid. It uses red-black updating with a user-specified smoother and fully weighted restriction/bilinear interpolation for solution transfer between grids. These are implemented as methods of a solver class (mgrid::LinearMultigrid) so that you don't have to specify too much to get the solver running.

Unfortunately there isn't much in the way of documentation other than this README but you can have a look at the two examples and get the general idea of what's going on.

I'm releasing this under the the Community Research and Academic Programming License (CRAPL: see http://matt.might.net/articles/crapl/ for details). Hopefully things'll work as advertised but if you have any problems then drop me a line and we'll see if we can fix them.

Installing

You'll need to install cmake, and the Boost, Blitz++ and netCDF C++ libraries to compile the multigrid library (Boost is available from here, Blitz++ from here, and the netCDF libraries are available from here). On a Mac the easiest thing to do is use Homebrew (get it here), so that this can be done in one hit with the command brew install cmake boost blitz netcdf. On debian the following works: apt-get install cmake libboost-dev libblitz-dev libnetcdfc++4 libhdf5-openmpi-dev.

Once you've installed these dependencies, you can go to the root folder (where this README is located) and run cmake . && make install. This should put the library and headers under /usr/local. Feel free to modify the install directory in the CMakeLists.txt file if you want it to go somewhere else.

When building your own script, you can also use the CMake files that are included in the two example directories. These should find the multigrid library and link it into your own executable.

Specifying your own elliptic PDEs

You can use this library by subclassing from the mgrid::LinearMultigrid class. You need to specify a differential operator, and a smoothing operator which will be specific to your given PDE. These functions are given a multigrid level, and i and j indexes which they can use to calculate the smoothing step. To make specifying these things easier, the solution attribute also implements finite differences of the specified order in the form:

solution[level].dxx(i, j)
solution[level].dzz(i, j)

...which would give you a finite difference approximation to the second derivative in the x and z directions respectively. These finite differences also take the location of the point into account, so if you're near a boundary they will automatically use forward or backward differences as required.

For example, to solve the Poisson equation for a solution $u_h$ on a grid with spacing $h = (hx, hz)$, with a source term $f_h$, we have the following differential operator:

$$ L(u_h) = d^2u/dx^2 + d^2u/dz^2 $$

and the following Gauss-Seidel update step as a smoother:

$$ S(u_h) = u_h(i+1, j)((u_h(i+1, j) + u_h(i-1, j))/(hx^2) + (u_h(i, j+1) + u_h(i, j-1))/(hz^2) - f_h(i, j))/(2(hx + hz)) $$

These are implemented in the Poisson example as

inline double differential_operator(mgrid::Level level, int i, int j) {
    return solution[level].dxx(i, j) + solution[level].dzz(i, j);
};
inline void Poisson::relaxation_updater(mgrid::Level level, int i, int j) { 
    const double hx = solution[level].spacing(0);
    const double hz = solution[level].spacing(1);
    const double xxfactor = 1/(hx*hx);
    const double zzfactor = 1/(hz*hz);
    solution[level](i, j) = 
	((solution[level](i+1, j) + solution[level](i-1, j))*xxfactor
	+ (solution[level](i, j+1) + solution[level](i, j-1))*zzfactor 
	- source[level](i, j))/(2*(xxfactor + zzfactor));
};

As you can see, you can get the spacing size from the current level of the solution. The solution and source arrays are stored as attributes containing vectors of grids, arranged from fine to coarse, and are called solution and source respectively. You shouldn't need to bother with the order too much, just pass the supplied level through.

The source term is automatically moved between grids using the same restriction/prolongation operators as used for the solutions. To make it wasier to specify initial conditions and a source term, the mgrid::LinearMultigrid class also contains finestLevel and coarsestLevel attributes, with the index of the finest and coarsest grid level respectively. You can just specify an array which gives the source function during construction of the class - everything that works to set the values in a Blitz array will work here. Here's how the Poisson example does it:

source[finestLevel] = -1.0; 
sourceIsSet = true;   

... the sourceIsSet attribute lets the class know you've specified this parameter.

Specifying boundary conditions

Boundary conditions can be set in the initialisation routine of your solver class which has been subclassed from mgrid::LinearMultigrid. You set them for the solution attribute.

Here's how it's done for the Poisson example, with two homogeneous Dirichlet and two Neumann conditions

// Set boundary conditions for velocity array 
solution.boundaryConditions.set(mgrid::leftBoundary,   mgrid::zeroNeumannCondition);
solution.boundaryConditions.set(mgrid::rightBoundary,  mgrid::zeroDirichletCondition);
solution.boundaryConditions.set(mgrid::topBoundary,    mgrid::zeroNeumannCondition);    
solution.boundaryConditions.set(mgrid::bottomBoundary, mgrid::zeroDirichletCondition); 

If your conditions are not homogeneous, you can specify a value using the BoundaryPoint class. For example, the conditions above were specified as:

const mgrid::BoundaryPoint zeroDirichletCondition = { mgrid::dirichlet, 0.0 };

A one-dimensional blitz array of BoundaryPoints makes an instance of mgrid::Boundary - you can also pass one of these to mgrid::BoundaryConditions::set to set the boundary conditions.

Specifying solver settings

Most of the settings are fairly self-explanatory:

struct Settings { 
    double aspectRatio; 		# Aspect ratio of the grids
    int numberOfGrids; 			# Total number of grids
    int minimumResolution;		# Resolution on the coarsest grid
    double residualTolerance;		# Stopping tolerance for solver
    int maximumIterations;		# Maximum allowable iterations for the solver
    CycleType mgCycleType;		# Either mgrid::wCycle or mgrid::vCycle
    unsigned long preMGRelaxIter;	# Number of relaxation iterations on way down
    unsigned long postMGRelaxIter;	# Number of relaxation iterations on way back up
};

...although a couple need a bit more explanation. The library will work out all the grid sizes based on your minimum grid size and the total number of grids by simply doubling the resolution in both the x and z direction with each grid.

You specify the minimum grid size as the minimum resolution on the smallest side of the grid - the library will adjust the x and z spacings to have as close to the same resolution in both directions as possible using the aspect ratio setting.

Running the solver

You can run the multigrid solver using the mgrid::LinearMultigrid::multigrid method. There's an optional mgrid::LinearMultigrid::solve method that you can do more complicated stuff with. For example the viscoplastic channel flow example requires a linear elliptic PDE to be solved at each step, and the source term updated from the last solution. The solve method deals with this recalculation of the source term and then calls the multigrid method.

Output

Solution output is in netCDF format. To specify file names programmatically you can overload the mgrid::LinearMultgrid::filename method. This method takes a string and returns a string with a filename. This makes it easy to insert the current parameters into the output filename when you're running a bunch of solutions at once. To actually write the solution to file you can call the mgrid::LinearMultigrid::write method and pass two arguments: the first integer is the variables to write out ('1' writes out just the solution, '2' writes the solution and its derivatives, and '3' writes out the solution, derviatives and the error between the solution and the differential equation), and the second string is the string that gets passed to the filename method (use this to pass a folder reference if you like).

In the Poisson problem example this is done like so:

std::string Poisson::filename(std::string root) {
    std::ostringstream name;  
    name.precision(1);  // Print variables to one decimal place
    name << root << "A" << std::fixed << aspect; 
    return name.str();
} 

and the writing call is

problem->write(1, "poisson_");

where problem is an instance of Poisson, a subclass of mgrid::LinearMultigrid. This method will write a file poisson_A2.nc for a problem with an aspect ratio of 2.

License

I'm releasing this code for academic use under the Community Research and Academic Programming License. You can read the terms of this license here. Specifically, this licence is designed to achieve the following:

Most open source licenses (1) require source and modifications to be shared with binaries, and (2) absolve authors of legal liability.

An open source license for academics has additional needs: (1) it should require that source and modifications used to validate scientific claims be released with those claims; and (2) more importantly, it should absolve authors of shame, embarrassment and ridicule for ugly code.

The Author reserves all rights to the Program, except for any rights granted under any additional licenses attached to the Program.

Basically I'm pretty happy to let you use this code for non-commercial/academic/research use, provided you cite my work when/if you publish your work. However, if this is too restrictive for you then drop me a line and we can have a chat.

About

Simple linear multigrid PDE solver written in C++

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published