|
Dear Jena,
Yes, this is quite standard to consider
an absolute maximum residual proportional to a RHS norm. However,
the residual considered in the standard solve in GetFEM is already
a relative residual, which means that the maximal residual given
to the standard solve is already multiplied by an estimate of the
total load.
Best regards,
Yves
Le 09/05/2019 à 08:44, Moritz Jena a
écrit :
Hello Yves,
Thanks for your quick answer.
Adapting the residuum at each
iteration
step is very interesting and I will try to implement this in my
code.
However, my problem is still, to
determine/calculate
the maximum allowed residuum depending on my model.
At the moment I follow the
theory, that
the maximum allowed residuum max_res is calculated with respect
to the
external forces:
If external_forces > 1e-5 *
1000
max_res =
external_forces / 1000
else
max_res =
1e-5
end
I base my theory on observations
in
ansys. Am I right with my theory? Or is there even another
(maybe easier)
way to get a plausible max_res out of GetFEM?
Best regards,
Moritz Jena
Von:
Yves Renard
<address@hidden>
An:
Moritz Jena
<address@hidden>
Kopie:
getfem-users
<address@hidden>
Datum:
08.05.2019 17:11
Betreff:
Re: [Getfem-users]
Maximum allowed residuum
Dear Jena,
Unfortunately, from the interface you cannot adjust the
maximum residual
and change it from a Newton sub-iteration to another. The only
mean I see
from the interface is to call for only one Newton iteration
each time,
then change the maximum residual from a call to another and
loop until
the convergence criterium is not reached.
Best Regards,
Yves
----- Mail original -----
De: "Moritz Jena" <address@hidden>
À: "getfem-users" <address@hidden>
Envoyé: Mardi 7 Mai 2019 16:23:00
Objet: [Getfem-users] Maximum allowed residuum
Dear GetFEM-Users,
I have written a FEM program, with which you can calculate
technical
springs numerically.
Therefore I'm using the GetFEM-MATLAB-Interface 4.3
These calculations are highly non-linear due to self-contact
issues and
large deformations.
That is the reason why technical springs have to be solved
iteratively
in
several substeps.
The calculation by itself works very well. But I think I have
a problem
with the maximum allowed residuum that is passed to the solve
command of
getfem.
My theory is that the maximum allowed residuum has to be
variable during
the different substeps.
By investigating these question online I found a paper of an
university
in
Germany that contains the following for a equation system of
A*x=b : ||delta_x||/||x|| <= Cond(A) * ||delta_b||/||b||
Unfortunatley these _expression_ is not leading to plausible
results.
I'm also familiar using ansys. By using this commercial
software I know
that the maximum allowed residuum is changing by each
iteration and by
each substep.
All of the delivered getfem examples are using a constant
maximum allowed
residuum, no matter of high deformation or not.
So my question is, how am I able to find the maximum allowed
residuum and
by what _expression_? And do I have to define a variable maximum
allowed
residuum during the substeps to get reliable results out of
the solve
command?
Best regards,
Moritz Jena
--
Yves Renard (address@hidden) tel : (33) 04.72.43.87.08
INSA-Lyon
20, rue Albert Einstein
69621 Villeurbanne Cedex, FRANCE
http://math.univ-lyon1.fr/~renard
---------
|