sixpack-2.3.7
所属分类:其他
开发工具:Fortran
文件大小:212KB
下载次数:109
上传日期:2009-10-17 05:08:30
上 传 者:
Lovehere
说明: Sixpack is a library of solvers that may be used to solve structured finite volume and finite difference discretisations of PDE s. The solvers include static methods like Jacobi s Method, SOR, SSOR, RBSOR, incomplete factorisation methods like ILU, ILDL, SIP, MSI, Krylov space methods like CG and, BiCGSTAB (with a variety of preconditioners), and a multigrid solver using Jacobi, ILU, ILDL and SIP for smoothing. In addition there is a block tridiagonal solver for when you want to try a direct method.
文件列表:
sixpack-2.3.7\data\scripts\run_convergance_2D.sh (2038, 1999-12-25)
sixpack-2.3.7\data\scripts\run_convergance_3D.sh (2050, 1999-12-25)
sixpack-2.3.7\data\scripts\run_scaling_2D.sh (1886, 1999-12-25)
sixpack-2.3.7\data\scripts\run_scaling_2D_mg.sh (1566, 1999-12-25)
sixpack-2.3.7\data\scripts\run_scaling_3D.sh (1849, 1999-12-25)
sixpack-2.3.7\data\scripts\run_scaling_3D_mg.sh (1592, 1999-12-25)
sixpack-2.3.7\doc\compile_options (1228, 1998-10-10)
sixpack-2.3.7\doc\options (3389, 1998-10-10)
sixpack-2.3.7\include\Makefile (156, 1999-12-25)
sixpack-2.3.7\include\sixpack.h (10587, 1999-12-25)
sixpack-2.3.7\include\sixpack_generic.h (5732, 1999-12-25)
sixpack-2.3.7\include\sixpack_internal.h (74667, 1999-12-25)
sixpack-2.3.7\include\sixpack_precon.h (803, 1999-12-25)
sixpack-2.3.7\include\sixpack_version.h (463, 1999-12-25)
sixpack-2.3.7\Makefile (709, 2000-01-31)
sixpack-2.3.7\Makefile.conf (746, 2000-01-31)
sixpack-2.3.7\man\boundary5.3 (2213, 1999-11-14)
sixpack-2.3.7\man\boundary7.3 (2634, 1999-11-14)
sixpack-2.3.7\man\boundaryinterp5.3 (842, 1999-11-14)
sixpack-2.3.7\man\boundaryinterp7.3 (860, 1999-11-14)
sixpack-2.3.7\man\getcompileoptions.3 (22, 1999-11-14)
sixpack-2.3.7\man\getrealkind.3 (22, 1999-11-14)
sixpack-2.3.7\man\getversion.3 (1325, 1999-11-14)
sixpack-2.3.7\man\Makefile (833, 1999-11-14)
sixpack-2.3.7\man\sixpack.3 (4851, 1999-11-14)
sixpack-2.3.7\man\solver3.3 (4398, 1999-11-14)
sixpack-2.3.7\man\solver5.3 (5940, 1999-11-14)
sixpack-2.3.7\man\solver7.3 (6120, 1999-11-14)
sixpack-2.3.7\scripts\freezeall.sh (97, 1999-12-13)
sixpack-2.3.7\scripts\Makehead.absoft (746, 1999-12-13)
sixpack-2.3.7\scripts\Makehead.cm (987, 1999-12-13)
sixpack-2.3.7\scripts\Makehead.dec (840, 1999-12-12)
sixpack-2.3.7\scripts\Makehead.dechpf (909, 1999-12-12)
sixpack-2.3.7\scripts\Makehead.gmdhpf (739, 1999-12-13)
sixpack-2.3.7\scripts\Makehead.nagf90 (884, 1999-12-13)
sixpack-2.3.7\scripts\Makehead.pgi (791, 1999-12-13)
sixpack-2.3.7\scripts\Makehead.sgi (1109, 1999-12-12)
sixpack-2.3.7\scripts\Makehead.std (532, 1999-12-12)
sixpack-2.3.7\scripts\Makehead.sun (817, 1999-12-12)
... ...
cd to the main directory, and type
make make
to make the makefiles for your platform. Then
make
to build the libraries and test program.
The libraries can be built in three forms: do, f90 and shift, depending
on how the matrix-vector multiplies are done. Some solvers only exist
in the do form.
Not all solvers will work on the CM-5. I assume the same will be true
on other parallel platforms.
A (old and outdated) report on the solvers can be found from
http://www.maths.unsw.edu.au/~norris
Notes:
A few notes about using the solvers.
1) ilu3,sip3,msi3 seem to do the same thing as tridag3.
2) these methods fail on neumann problems unless the equations are
set at a point to be zero (ie: ae = aw = 0; ap = 1; q = 0).
similarly they fail as preconditioners unless the equations are set.
3) the other (iterative) methods don't fail, but converge faster when
set equations are neither reset or set at each iteration. reset gives
same results as using neither reset or set.
Anyhow, this presents a quandry. To converge fastest requires
different methods of zeroing the equations.
The above line is not a problem. Testing 2D and 3D versions of the
solvers shows that _all_ solve OK with no set or reset (including
the direct methods!). And all are best solves with neither set
or reset. This result was found for 1D and 2D neumann heat equations,
and the 2D pressure equation in the data directory.
4) if a preconditioner solves the equations exactly, the CG scheme
will solve in one step (as one would hope).
5) the error status returned in the stat variable is
0 no error. problem solved
1 solver did not converge
2 solver encountered a problem with the input array (such as a
zero values diagonal element ap)
3 solver not implimented on this platform
4 memory allocation failure
--
Stuart Norris norris@euler.me.su.oz.au
Mechanical Engineering, University of Sydney, NSW 2006 +(61 2) 9351-2272
近期下载者:
相关文件:
收藏者: