From: Juha Ruokolainen <[email protected]>
Subject: [Elmerupdates] Elmer Version 5.3 Released
Date: Tue, 30 Jan 2007 10:15:53 +0200 (EET)
Elmer version 5.3.0 has been relased
====================================
- New Windows binary is available from:
http://www.csc.fi/elmer/download
or directly from
ftp://ftp.funet.fi/pub/sci/physics/elmer/bin/Elmer-5.3.0.exe
The binary has been compiled under MINGW using the default
GCC + gfortran v4.3 available from:
http://gcc.gnu.org/wiki/GFortranBinaries
- Sources available through
http://www.csc.fi/elmer/download
or directly from
ftp://ftp.funet.fi/pub/sci/physics/elmer/src
New Features include:
=====================
o Nodal Loads & Nodal Dirichlet conditions:
In boundary condition-section give
Target Nodes(n) = Integer
or
Target Coordinates(n,dim) = Real
The coordinates, if given, will be used to search closest nodes.
Dirichlet conditions and nodal loads may be set the usual way for
these boundary conditions, i.e.
Temperature = Real value
Temperature Load = Real value
o Bulk Nodal Loads & Bulk Dirichlet Conditions
In body force-section variable conditions & loads may be given as
Temperature = Real value
Temperature Load = Real value
o New Harmonic analysis solver-section keywords
Harmonic Analysis = Logical
Frequency(n) = Real
these keywords may be set to activate harmonic analysis for
second order time derivative equations (StressSolve etc.):
( (-w^2M+K) + iwD ) X = F
where M is mass matrix, K stiffness matrix, and D damping matrix.
The solution is over given frequency (f=w/(2pi)) values.
Current limitations include:
x Dirichlet conditions are real
x Only real part of the force may be given
x Only real part of the solutions appears in output
o New solver-section keyword
Constant Bulk System = Logical
which may be set to 'True' if the bulk matrix & rhs are known
not to change during simulation. The boundary conditions may
change the matrix and vector as needed. Currently only used
by StressSolve & HeatSolve.
o New solver-section keyword
Constant Bulk Matrix = Logical
which may be set to 'True' if the bulk matrix is known not to
change during simulation. The boundary conditions may change the
matrix as needed. The RHS vector may also change. Currently used
only by StressSolve.
o New solver-section keyword
Linear System Refactorize = Logical
may be set to 'False' if it is known that the coefficient matrix does
not change during simulation. StressSolve also compensates for changing
timestep by reactivating factorizing when the timestep changes, for other
solvers this keyword must not be used with varying timestep.
This keyword applies to
Linear System Solver = Direct
Linear System Direct Method = UMFPack
as well as to iterative solver preconditioning.
o New 'Output File' format, including binary output by simulation-section
keyowrd:
Output Binary = Logical
New format saves space and execution time.
o Output of solution fields to VTK and OpenDX formats
o New boundary condition-section keyword
Normal-Tangential var_name Rotate = Logical
may be used to disable (by setting to value False) the change
for variables (x,y,z) -> (n,t1,t2) on a N-T boundary if not
needed (no dirichlet conditions).
o New keyword in equation-section
Element = [n:#dofs d:#dofs p:#dofs b:#dofs e:#dofs f:#dofs]
where n means nodal dofs, d discontinuous galerkin element dofs,
p stands for p-element 'p', b is number of bubble degrees of
freedom, e- and f- give edge and face degrees of freedom
respectively.
This may be used, instead of the mesh defintion, to enable
unusual (not Lagrange type) element types.
Element = [p:2]
would give p=2 element over the basic element type the mesh
defines.
Element = [d:4]
gives DG element with 4 internal DOFs. Specifying d:0 has the
special meaning that DG dofs equals to the #nodes within the
mesh defined element.
Element = [p:1 b:1]
would give one bubble DOF over a linear element, etc.
If equation-section defines more than one solver
Active Solvers(n) = Integer
the element defintion may be different for individual solvers
Element[1] = ...
Element[2] = ...
o The field variable components may be given individual names:
Variable = var_name[cname 1:#dofs cname 2:#dofs ... ]
instead of the default component names "var_name 1", "var_name 2", ...
These given component names may be used in boundary conditions,
initial conditions, etc. and are seen as field variables
as usual.
o All solvers, including FlowSolve, StressSolve, HeatSolve, may
now be used as "external solvers", specifying i.e.
Procedure = "FlowSolve" "FlowSolver"
Variable = Flow[Veloc:3 Pres:1]
in solver section.
o If variable "primary_var loads" exists the linear system solver
will compute boundary reaction forces as a field variable after
solution:
F=Ax-b
where A and b are the bulk matrix and RHS vector respectively
(the system matrix & vector before applying any boundary
conditions).
For example specifying
Procedure = "FlowSolve" "FlowSolver"
Variable = Flow[Veloc:3 Pres:1]
Exported Variable 1 = Flow Loads[Stress Vector:3 CEQ Residual:1]
will compute boundary stress vector which could be used i.e. in FSI
simulation to specify the reaction of the boundary to flow field in
a boundary condition section:
Displacement 1 Load = Variable Stress Vector 1
Real MATC "-tx"
Displacement 2 Load = Variable Stress Vector 2
Real MATC "-tx"
Displacement 3 Load = Variable Stress Vector 3
Real MATC "-tx"
similarily
Procedure = "HeatSolve" "HeatSolver"
Variable = Temperature
Exported Variable 1 = Temperature Loads
will give nodal equivalent heat flux. Summing the 'Temperature Loads'
variable
over a boundary (using SaveData f.ex.) will give heat flux over the
boundary.
Currently implemented for StressSolve, HeatSolve and FlowSolve.
o New operators in SaveScalars
There are some new simple operators in SaveScalars: 'sum', 'dofs', 'boundary
sum', 'boundary dofs' and 'boundary mean'. These are particularly useful in
extracting the node based information. Assume that you would have computed
'Temperature Loads' as in above example. Then you could extract the sum of
loads i.e. the total heat flux over the domain and of a boundary by adding
the following solver:
Solver n
Equation = String "SaveScalars"
Procedure = File "SaveData" "SaveScalars"
Filename = File "f.dat"
Variable 1 = String "Temperature Loads"
Operator 1 = "sum"
Variable 2 = String "Temperature Loads"
Operator 2 = "boundary sum"
End
Additionally a flag 'Save Scalars = Logical True' at the boundary
of interest should also be set.
o Initial support for HYPRE parallel linear solver library
x compile and install the HYPRE library. HYPRE is found at
http://www.llnl.gov/CASC/linear_solvers/
x configure mathlibs for parallel computing
./configure --prefix=... --with-mpi
or
./configure --prefix=... --with-mpi-dir=...
make install
x configure fem for parallel computing and HYPRE
export CFLAGS "-Impi_include_dir -Ihypre_include_dir ..."
./configure --prefix=... --with-mpi-dir=... \
--with-hypre="-Lhypre_lib_dir -lHYPRE"
make install
and set the solver-section keyword in the solver input file
Linear System Use HYPRE = Logcal True
HYPRE solver used by the elmer interface is BiCGStab with (global)
ILU(n) preconditioning. The n in ILU(n) may be selected the usual way
Linear System Preconditioning = ILUn
Later versions may include support to different HYPRE linear equation
solvers, like AMG, etc.
o Partitioning (ElmerGrid)
Improved partitioning capabilities. Most importantly, interfaces to some
dual graph based partioning routines of Metis library implemented. This
makes the Metis partitioning available also for mixed meshes. The new
options are available using an extra integer parameter (optional) in the
command line. See the Metis documentation for more info.
-metis 4 0 % METIS_PartMeshNodal (default)
-metis 4 1 % METIS_PartGraphRecursive
-metis 4 2 % METIS_PartGraphKway
-metis 4 3 % METIS_PartGraphVKway
For the simple geometric partitioning there are two possibilities. You may
either first partition the elements and then define the nodal partitioning
accordingly, or vice versa.
-partition 2 2 1 0 % First partition elements (default)
-partition 2 2 1 1 % First partition nodes
o Conservation of names
There is a new possibility to conserve name information in some mesh
import routines. Currently this is supported only for the newest Ansys
script and for the FDNEUT format. In mesh format the name information is
saved to a separate file (mesh.names) that may be included in the sif
file, and thereafter the material names may be used as MATC variables.
o Removal of extra BCs
There is are some new options in ElmerGrid which to remove lower
dimensional BCs and to map the BC numbering to be continous. This may be
useful for meshers (e.g. Gmsh) that automatically create a bunch of
geometric entities making it difficult to handle them in later phases.
o Shell-element solver including large deflections
o Kohn-Sham equations solver for density functional theory
o Many new test cases
The test cases are in the fem-5.3.0.tar.gz package in the fem-5.3.0/tests
directory. After building and installing ELMER you should be able to run
all the tests by commanding
make check
in the fem-5.3.0 directory. To run individual tests enter the tests
directory and give the command
./runtests test_dir
To use the test cases as a starting point for developing your own case,
please note that
x Output is disabled by default, to re-enable uncomment either or both of
the lines
!Output File=..
!Post File=...
in Simulation-section of the .sif file.
x Solver output messages are restricted by
Max Output Level = 3
increse the number or remove the line to get more verbose output.
There are in total 69 test cases in this release, running through them takes
few minutes depending on your computer capabilities, compilers and compiler
options used.
o Numerous bug fixes and smaller enhancements
Have Fun,
The Elmer Team
