From: Peter Råback <[email protected]>
Subject: [Elmerupdates] Elmer 7.0 released
Date: Thu, 31 May 2012 15:25:20 +0300
Hi All
Due to the LGPL update of Elmer library we changed the leading number in the
Elmer version. So the current number is 7.0. There are not other drastic
changes but as the license policy of LGPL is different from GPL we considered
that it was time for a change. So starting from version 7.0 you can now link
your fully own (not derived!) solvers with the library as you wish.
There are also fresh Windows versions and documentation on sourceforge and
nic.funet.fi.
Below are the major new features between revisions 5191 (aka 6.2.) and 5685
(aka 7.0).
Thank you very much for all the contributors!
Best regards,
ElmerTeam
Software development:
----------------------------
* Doxygen code documentation system taken into use,
http://www.elmerfem.org/doxygen
* Moved dynamically linked solvers into separate subdirectory "fem/modules"
* Moved the ElmerSolver library under LGPL license.
Element basis:
-----------------
* Edge basis functions for quad & brick & wedge elements.
* New BC section Keyword "Mass consistent normals". Use mass preserving
normals on normal-tangential conditions instead of the usual nodal averaged
normals. Works in 2D and for linear- and trilinear elements in 3D.
* Enabled requesting of Gauss Points using relative order {-1,0,+1}. This is
enabled for p-elements with generic rules and for Lagrangian elements the
number of Gauss points for specific order must be provided in "elements.def".
For the code developer the option is seen as a new optional parameter
"RelOrder" in GaussPoints funtion. This must be implemented for each solver
separately. So far only StressSolve has been modified to with the "Relative
Integration Order" keyword.
Linear system:
------------------
* Linear systems may now be scaled in the solution phase using the row
equilibration (by setting Linear System Row Equilibration = Logical True)
* It is now possible to utilize the block solver strategy for monolithic
solver (BlockSolve.src). Uses the legacy solvers only for assembly, splits
the matrix into blocks and then solves the boock system.
* Modified HYPRE interface (contribution from Jonas Thies):
- Allow separate setup and solve.
- New keyword "HYPRE Block Diagonal", which does something similar to the
BILUn serial preconditioner when calling HYPRE.
- HYPRE interface now accepts 'CG' as solver option.
* interface to linear solvers from the Trilinos libraries. This can be
enabled by compiling with -DHAVE_TRILINOS and currently works only in
parallel (through SParIterSolver). Some examples are in
fem/examples/TrilinosSolvers. This is still very experimental! (contribution
from Jonas Thies)
* Added a simple iterative method 'richardson' preconditioned with a lumped
mass matrix. The algorithm is very similar to Jacobi. Might have some use in
solving problems involving just a simple mass matrix.
Misc:
------
* Enabled keyword "Timestep size" that takes MATC definitions etc. as any
other Real variable.
* Add interface "GetLocalSolution()" to unify "GetScalarLocalSolution()" and
"GetVectorLocalSolution" utility subroutines.
* Improved control on verbosity. Additional in-line parameters -s/--silent,
default level of four for Info, and keyword check 'silent'.
* Enabled internal 1D mesh generation for the 1D outlet solver.
* Improved the speed of ListGetReal command in case when a constant is being
fecthed. Also made a new subroutine ListGetRealAtIp and a related data
structure that is intended to be called at each integration point. The data
structure includes information that is used to remember things from previous
evaluations to save time. This is not yet used in any solver so changes are
still possible, and even probable.
* Solver section keyword "Back Rotate N-T System=Logical" (default: True).
This keyword can be set to false, to disable the rotation of the solution
fields of normal-tangential boundaries to coordinate directions.
New features for old Solvers:
-----------------------------------
HelmholtzSolver:
* Enabled variable density fields in Helmholtz solver. This affects the
results only when density is not constant. Also a interface to harmonic
structural analysis was implemented.
ElasticSolve:
* Stress and strain computation enabled.
* New material models: a neo-Hookean material model and an orthotropic
material model added for handling cases with large rotations and small strains
FlowSolver:
* Add new Navier-Stokes discretization option 'Laplace Discretization'. The
diffusion term discretized is \mu div(grad(u)) instead of the default 2\mu
div(\epsilon(u)). The resulting linear system is smaller (in terms of
nonzeros) and usually easier to solve. This has, however, an effect on the
meaning of the various boundary conditions.
* introduced Glen's flow law as a separate non-linear flow law in Effective
Viscosity
StatElecSolver:
* Added infinity conditions to the electrostatic equation. These assume
spherical symmetry of the far-field.
SaveLine:
* Enabled solution of isocurve points in SaveLine: Given some field, save the
intersection points that have a predefined value.
ResultOutputSolver (all related to VTU output):
* Enabled VTK Collection for timesteps
* Enables saving of body and bc id:s by keyword "Save Geometry Ids"
* Implemented eigenmode saving with two different numbering strategies
* Enabled timestep indexes with more than four numbers in length.
* Enables saving of "elemental" information.
HeatSolver:
* Added perfusion rate to HeatSolver (contribution from Matthias Zenker).
StressSolver:
* Add more general form of Robin type of BC:s: -(tau,n) + Ku = f, with K a
dim x dim matrix given with "Spring ij" keywords.
MagnetoDynamicsCalcFields:
* Modify the results computation (MagnetoDynamicsCalcFields) as so that
default output still is nodally fitted variables. Also elementwise nodal
fields directly computed to nodes (variable names with ' E' postfix)
* If "Discontinuous Galerkin=True" then elementwise nodally fitted variables
(with no postfix on names)
* If "Discontinuous Galerkin=True" and "Average within Materials=Logical
True" then materialwise fitted variables (discontinuous at material
interfaces, with no postfix on names)
Note that default ElmerPost output only contains the nodally fitted
variables, use ResultOutputSolver with format="ep" or format="vtu" to look at
the other variations.
New Solvers:
----------------
ParticleAdvector:
* Particle based solver for diffusion-free advection of scalar fields. For
each node a particle is sent -dt backwards in time and the field value is
taken from the resulting upstream point. This provides diffusion free
advection. The robustness of the solver may still be suspectable since it is
based on the ability to follow particles in a mesh.
ThermoElectricSolver:
* First version of a tightly coupled thermoelectric solver.
ShallowNS:
* First version of the shallow water Navier-Stokes equations solver.
