cpraveen / dflo

Discontinuous Galerkin solver for compressible flows

dflo

Discontinuous Galerkin solver for compressible flows. Some features of the code are

• cartesian and quadrilateral cells
• Qk basis: nodal Lagrange polynomials on Gauss points
• Pk basis: Legendre polynomials
• TVB limiter
• Positivity preserving limiter
• Flux functions: Lax-Friedrichs, Roe, HLLC, KFVS

###Getting dflo

git clone https://github.com/cpraveen/dflo

To compile code in "src"

You have to first compile deal.II with Trilinos library, UMFPACK, threads. Serial versions of these libraries are sufficient since the code in "src" is a serial code.

• cd dflo/src
• You must set the variable DEAL_II_DIR to your deal.II installation directory. It is good to set this in your bashrc file.
• cmake .
• make

By default, this compiles a DEBUG version of the code which is good for code development but runs very slowly. In order to compile optimized version, do

make release

To compile code in "src-mpi"

This is the mpi version which uses p4est. So you must install p4est and compile deal.II with p4est support. It does not require trilinos.

To install p4est, see this page

https://www.dealii.org/developer/external-libs/p4est.html

Obtain latest version of deal.II from github

git clone https://github.com/dealii/dealii

Change into dealii directory

cd dealii
mkdir build
cd build

Configure deal.II. A basic setup is given in the file dealii_mpi.sh. Just run this inside build directory

./dealii_mpi.sh

Now compile deal.II and install it

make all
make install

After this you can compile dflo.

Running the code

Many test cases are included in the examples directory. In most examples, a geo file is provided to generate the mesh. You need Gmsh to generate the mesh as follows

gmsh -2 grid.geo

Then run dflo

dflo input.prm

Some parts of dflo are parallelized with threads. You can specify number of threads to use while starting dflo

dflo input.prm 4

which will run with 4 threads.

To run the mpi version, do

mpirun -np 4 dflo input.prm

dflo's Issues

Periodic boundary conditions

Optimize compute_cell_average()

Since we use QGauss nodal basis, we can optimize this function. No need to call get_function_values.

QGauss leads to diagonal mass matrix. Simplify code.

With Gauss nodes for basis functions, mass matrix is diagonal but we use full matrix which is computationally inefficient. Exploit the diagonal structure of the mass matrix to speed-up computations.

Implement MOOD method

Improve cell average computation of neighbours on adapted grids for limiter

For TVD limiter, we need cell average of neighbouring cells. If they are at different levels, then we have to compute cell average carefully. If neighbour cell is coarser, then compute cell average by integrating on subcell.

Implement limiter for quadrilateral cells

Allow dx, dy, dz to be different in cartesian grids

Currently we assume square/cubic cells.

Prevent islands in shock indicator

Include path not correct

The include path in source files are not correct. So the CMake generated make file will not work.
For example, inside src/assemble_explicit.cc, the include statement reads

#include <base/quadrature_lib.h>

which should be

#include <deal.II/base/quadrature_lib.h>

Parallelize with MPI

Make the code 3-d

Some parts of code like limiter, numerical flux (some of them), etc. are written only for 2-d case.

Add entropy stable flux

Apply BC for slip boundary in limiter

If cell is adjacent to slip boundary, then ghost cell state must be used to compute backward/forward difference in the limiter.

src_gll: remove updates_values from cell assembly

We dont need shape values for cell assembly.

src_gll: potential can be precomputed

With gravity, potential can be computed and stored as an FE vector rather than recompute every time. update_q_points can then be removed from cell assembly.

scr_mpi runtime error in dealii::FEValuesBase<dim, spacedim>::get_quadrature_points()

Hello,

when running dflo (built against deal.ii >=8.3) from src_mpi, and after fixing external_force function parser initialization, I get a runtime error:

An error occurred in line <4373> of file </home/pkestene/local/dealii_4_dflo/include/deal.II/fe/fe_values.h> in function
const std::vectordealii::Point& dealii::FEValuesBase<dim, spacedim>::get_quadrature_points() const [with int dim = 2; int spacedim = 2]
The violated condition was:
this->update_flags & update_quadrature_points
The name and call sequence of the exception was:
ExcAccessToUninitializedField("update_quadrature_points")
Additional Information:
You are requesting information from an FEValues/FEFaceValues/FESubfaceValues object for which this kind of information has not been computed. What information these objects compute is determined by the update_* flags you pass to the constructor. Here, the operation you are attempting requires the <update_quadrature_points> flag to be set, but it was apparently not specified upon construction.

Following what is done in src, if I add the requested flag, everything seems OK.
I'll place a pull request for that. Please ignore if you can fix this otherwise.

best regards,
Pierre Kestener.

Can this solver solve problems on non-cartesian mesh?

I tested with the naca0012 mesh but found the solution not satisfactory. Is it because this code can not apply to non-cartesian mesh yet ? If I want to add this feature, which part of the code should I modify?

Roe scheme with cell average for dissipation matrix