release_5.4.txt

From: Juha Ruokolainen <jpr@csc.fi>
To: elmerupdates@postit.csc.fi
Subject: [Elmerupdates] ELMER 5.4.0 RELEASED
Date: Wed, 3 Oct 2007 14:18:21 +0300 (EEST)



Elmer version 5.4.0 is out:

Source code is available from

http://nic.funet.fi/pub/sci/physics/elmer/src

Elmer 5.4 consists of the following packages:

matc 5.4.0
umfpack 4.4
mathlibs 1.0.0
elmergrid 5.4.0
meshgen2d 5.0.0
eio 5.4.0
hutiter 5.4.0
fem 5.4.0
elmerparam 1.0.0
front 5.4.0
elmerpost 5.4.0

The file elmer-5.4.0.tar.gz is also provided for convenience,
this contains all the relevant packages.

Windows binary installation package is available at

http://nic.funet.fi/pub/sci/physics/elmer/bin/windows/Elmer-5.4.0.exe.

There are new compile instructions for Windows/MINGW at the bottom
of the page

http://www.csc.fi/english/pages/elmer/download/compiling


New features of Elmer 5.4.0 include:
====================================

New manuals:
------------
* MATC manual was extracted from the html help to latex format. Now available
in as a new document called MATCManual.pdf.

* Elmer Overview manual was written. It has an emphasis on big things and 
should
help people in learning Elmer. Available in cvs as doc/overview and as a new
document called ElmerOverview.pdf.


Solver enhancements:
--------------------

* "Internal solvers" identified by Equation ="Heat Equation" | 
"Navier-Stokes" |
"Mesh Update" | "Magnetic Induction" | "Stress Analysis" may now use "Exported
Variable n" definitions in solver-sections. Previously these definitions were
silently ignored.

* Added Boundary Condition-section keyword "Target Coordinates Eps=Real".
Target nodes are only set if closer than "eps" to given coordinate values.
The default is to set closest nodes irrespective of their distance to given
coordinates.

* New Solver-section keyword "Radiation Solver=Logical" may be used to define 
the
solver responsible of taking care of the heat radiation, if a Solver with
"Equation=Heat Equation" is not present.

* Added Simulation Type=Scannig, which goes through timesteps etc., but 
doesn't set transient
mode.

* Added DefUtils function GetCReal(), which is similiar to GetReal but 
returns a scalar value.
One may use this for parameters that are time dependent (for example), but 
not spatially varying.

* Added Material-section keyword "Pressure Coefficient" which may be used to 
give the pressure
material derivative coefficient value in the heat equation. This should give 
a value for
the expression (@log(1/rho)/@log(T))_p

* Added feature to define elements passive in the middle of the simulation. 
Passive elements are
controlled by a Body Force keyword (type Real), e.g. "Temperature Passive" 
for Heat Equation.

* Possibility to change the reference shape of linear elasticity.

* Added possibility for saving scalars data in a file understood by the 
LiveGraph? program

* Added some new nonlin. and steady state convergence measures and features.

* Added a module IterativeMethods?.src where new iterative solvers may be 
placed and referred in
IterSolve?.src. New iterative solvers include 'Jacobi', 'SGS', and 
'BiCGstabL'.

* The parallel solver has received some attention, and should work much 
better than
before. Also the HYPRE interface is updated to include ParaSails - a 
parallel sparse
approximate inverse preconditioner. To activate use

Linear System Use Hypre = Logical True
Linear System Preconditioning = ParaSails
ParaSails Threshold = Real
ParaSails Filter = Real
ParaSails Symmetry = Real
ParaSails Maxlevel = Real

look at the HYPRE/ParaSails documentation for the meaning of preconditioner 
settings.


* changes in convergence measures & norms & relaxation:

Convergence measures are used to monitor the convergence of the solution process. The different measures are 

c_n = 2*||u|-|u_0|| / (|u|+|u_0|)

c_s = 2*|u-u_0| / (|u|+|u_0|)

c_r = |Ax_0-b| / |b|

where the suffixes (n,s,r) refer to values 'norm', 'solution', 'residual' of the keyword 

Nonlinear System Convergence Measure = String

The old standard way 'norm' is the default. The critical value when convergence is decleared is still terminated by the keyword 

Nonlinear System Convergence Tolerance = Real

Absolute measures (opposed to relative) may be invoked by

Nonlinear System Convergence Absolute = Logical True

Relaxation:

Relaxation is now also done within ComputeChange. The keywords related to that are 

Nonlinear System Relaxation Factor = Real
Nonlinear System Relaxation Before = Logical

The default of the second one is True (as before). This means that relaxation is applied before the convergence measure is computed. Heavy relaxation will therefore make convergence seem easier. The countereffect this the flag could be set False. This has the price that one extra norm has to be computed, and that norms are not backward compatible. 

Nonlinar vs. steady state:

The convergence tolerances are used to determine the convergence of a single equation, and that of the coupled system. Therefore all the above keywords have also a counterpart starting with string 'Steady State'. 

Steady State Convergence Measure = String
Steady State Convergence Tolerance = Real
Steady State Convergence Absolute = Logical
Steady State Relaxation Factor = Real
Steady State Relaxation Before = Logical

Norms
-----
By default Elmer uses the L2 norm. Now also other norms are possible in scalar computation. 

Nonlinear System Norm Degree = Integer

Note that value 0 refers to infinity, or max, norm,

Nonlinear System Norm Dofs = Integer

By default all degrees of freedom of the result vector are taken into account. Sometimes, it is desirable to look at only some components. Like in the Navier-Stokes equation the pressure may be a poorer indicator of the accuracy, and also it does not necesseraly have same scale as velocity. Therefore by default the Navier-Stokes solver sets the value of this keyword equal to DIM. 

The norm used in the convergence measures is always the same for steady state as for nonlinear norms. For the linear systems the only possible norm is still L2. 

What now
--------
Everything should be backward compatible. The most useful new feature will probably be the new convergence measure types that are more robust for some applications. 


New application solvers:
========================

PhaseChangeSolve
----------------
There is a new solver that is applicable to 2D phase change problems, 
occurring for example in the field of
crystal growth. Previously only Eularian phase change solution was supported. 
The Lagrangian approaches
should give superior accuracy with the same mesh density. The solver includes 
both steady state algorhitm
and a transient algorhitm.

The steady-state algorithm geometrically constructs a levelset and maps the 
old interface to the new
position. Here a flux condition must be set on the interface if there is a 
constant pull in the system
(look at PhaseChange2 in tests).

The transient algorithm computes the melting velocity over the interface and 
uses the velocity to advance
the position of the surface. With time the interface should obtain the 
steady-state solution, if such
exists.

FluxSolver
----------
There is a new solver that computes fluxes. For example, heat fluxes may be 
computed when the temperature
field T and heat conductivity k are known from
q = -k\grad{T}
The heat flux is computed with the Galerkin approximation one component at a 
time in order to optimize
memory consumption.

NormalSolver
-------------
There is a new solver that computes surface normals. Usually, the direction 
of the normals are assumed to
be defined by the boundary element alone. However, there are some cases when 
this may lead to slight
inconsistancies. Then the nodal values may be computed by this solver that 
uses the Galerkin method to
define nodal values for the normals. The normals are computed one component 
at a time in order to optimize
memory consumption.


Modifications to solvers
========================

SaveScalars
-----------
SaveScalars has now some additaional operators that may be used to save the 
current norms and changes to
external files in order to monitor the bahavior. The operators are 'norm', 
'nonlinear system change', and
'steady state change'. Also some simple operators operating on a boundary 
were added: boundary max,
boundary min, boundary max abs, boundary min abs.

The data in scalars may now be saved in comma-separated-values (CSV) format 
which is understood by the
LiveGraph program. This is envoked by setting a flag 'Live Graph = Logical 
True'.

The commands above could be used to monitor convergence behavior in real-time 
by adding the following
solver to the sif-file:

Solver 3
Name = "SaveScalars"
Procedure = "SaveData" "SaveScalars"
Filename = File "f.dat"
Variable 1 = Temperature
Operator 1 = norm
Operator 2 = nonlinear system change
End



ElmerGrid enhancements
======================

* The import of Universal file (UNV) and gmsh (ver 2) formats was improved.

* In ElmerGrid enhanced the functionality of the '-increase' (element order) 
flag also to 3D geometries.
The order may be increased from 1 to 2 only.


... and countless number of other enhancements, additions, bug fixes....


ElmerPost
==========

Added modules for exporting pictures/video clips in jpg/mpg format. The 
modules can be found
in the source tree (post/src), but they are not compiled automativally due to 
dependencies on
non-standard libraries. These are distrubted with the windows binaries 
however. More information
can be found from the headers of the source files.

Usage in Elmer-Post:

savejpg file.jpg quality [ quality: 1=bad ... 100=good ]
savempg start file.mpg [ initializes video compressor ]
savempg append [ adds current frame to video ]
savempg stop [ finalizes video compressor ]

Have fun,

The Elmer Team