New open source solver

jugghead · March 25, 2009, 05:48

What would be necessary to develop a:

Parallel Coupled (fully-implicit) compressible unstructured non-isothermal non-Newtonian multi-phase (n phases) flow solver?

Code: Object-oriented C++
Rigorous unit testing and version control.

RenardP · March 25, 2009, 09:57

Unix (Linux, Mac or FreeBSD).
gcc 4.x.x
g++ 4.x.x
gsl (Gnu scientific library)
BLAS, LAPACK (C++ versions)
OpenMP
MPI
git for version control
===============================
1. To many open source codes lack proper documentation, a first step will be to write down a short presentation of the theory you will implement in the code.

2. Document well the code.

3. Prepare a set of simple tutorials.

4. Use standard C++ as much as possible.

Hope this will be helpfull.

jugghead · March 25, 2009, 10:07

I agree. There should be emphasis on unit testing and documentation. Code documentation and user manuals.

Git or Mercurial seem to be the best now.

Also several numerical schemes would have be tested before implementation. Don't you think? Maybe there is a library missing which could do this. Such as a object-oriented library to calculate divergence, gradient, etc.

RenardP · March 25, 2009, 10:14

I never used it, but I've heard only good things about OpenFoam, maybe you can use this for differential operators (only second order when I've checked the documentation).

hjasak · March 25, 2009, 11:51

Nope - not a good fit. In order to do a good job with a density-based solver in OpenFOAM, (and block coupled pressure-based at the same time, why not?), you really want to write the discretisation differently. This is because of cross-field coupling: we still want rho, U, E etc to be self-contained fields but the matrix must be closely coupled.

However, the mesh, geometric field, linear solver, boundary conditions etc. classes can be re-used in full.

jugghead · March 25, 2009, 11:57

Wouldn't you also have to solve the non-linearity of Navier-Stokes equations using for example a Newton's based solver?

March 25, 2009, 05:48	New open source solver	#1
jugghead New Member Join Date: Mar 2009 Posts: 27 Rep Power: 17	What would be necessary to develop a: Parallel Coupled (fully-implicit) compressible unstructured non-isothermal non-Newtonian multi-phase (n phases) flow solver? Code: Object-oriented C++ Rigorous unit testing and version control.

March 25, 2009, 11:51		#5
hjasak Senior Member Hrvoje Jasak Join Date: Mar 2009 Location: London, England Posts: 1,905 Rep Power: 33	Nope - not a good fit. In order to do a good job with a density-based solver in OpenFOAM, (and block coupled pressure-based at the same time, why not?), you really want to write the discretisation differently. This is because of cross-field coupling: we still want rho, U, E etc to be self-contained fields but the matrix must be closely coupled. However, the mesh, geometric field, linear solver, boundary conditions etc. classes can be re-used in full. __________________ Hrvoje Jasak Providing commercial FOAM/OpenFOAM and CFD Consulting: http://wikki.co.uk

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
OpenFOAM on MinGW crosscompiler hosted on Linux	allenzhao	OpenFOAM Installation	127	January 30, 2009 19:08
OpenFOAM with IBM AIX	matthias	OpenFOAM Installation	20	March 25, 2008 02:36
Grid generation - Open source codes	Chris	Main CFD Forum	5	May 23, 2002 08:16
MAC Solver Source Code Page Updated	Maciej Matyka	Main CFD Forum	0	July 6, 2001 06:41
UDF Scalar Code: HT 1	Greg Perkins	FLUENT	8	October 20, 2000 12:40

March 25, 2009, 09:57		#2
RenardP Member Join Date: Mar 2009 Posts: 62 Rep Power: 17	Unix (Linux, Mac or FreeBSD). gcc 4.x.x g++ 4.x.x gsl (Gnu scientific library) BLAS, LAPACK (C++ versions) OpenMP MPI git for version control =============================== 1. To many open source codes lack proper documentation, a first step will be to write down a short presentation of the theory you will implement in the code. 2. Document well the code. 3. Prepare a set of simple tutorials. 4. Use standard C++ as much as possible. Hope this will be helpfull.

March 25, 2009, 10:07		#3
jugghead New Member Join Date: Mar 2009 Posts: 27 Rep Power: 17	I agree. There should be emphasis on unit testing and documentation. Code documentation and user manuals. Git or Mercurial seem to be the best now. Also several numerical schemes would have be tested before implementation. Don't you think? Maybe there is a library missing which could do this. Such as a object-oriented library to calculate divergence, gradient, etc.

March 25, 2009, 10:14		#4
RenardP Member Join Date: Mar 2009 Posts: 62 Rep Power: 17	I never used it, but I've heard only good things about OpenFoam, maybe you can use this for differential operators (only second order when I've checked the documentation).

March 25, 2009, 11:57		#6
jugghead New Member Join Date: Mar 2009 Posts: 27 Rep Power: 17	Wouldn't you also have to solve the non-linearity of Navier-Stokes equations using for example a Newton's based solver?