Discussion:
[COMSOL_Users] Weak form and unwanted solutions
Hamidreza Adib-RamezaniI
2008-04-20 16:08:28 UTC
Permalink
Hi,

Have anybody an experience about unwanted solutions using weak form?

Hamidreza
zimmerman_wbj
2008-04-20 21:03:47 UTC
Permalink
Hi Hamidreza,

You question is so open ended, that I thought it would engender no
discussion. So I have pitched three possibles at you:

(1) My usual experience with unwanted solutions is with highly
nonlinear problems which converge to solutions that did not match my
preconceptions of where the "real" solution is. I usually use the
parametric solver to "step along" the nonlinearity parameter to find
the bifurcation point.

(2) The other type of unwanted solution I get appears when the
boundary conditions are too prescriptive and kill off the desired by
say enforcing a symmetry.

(3) If I were to conjecture what type of "weak form and unwanted
solutions" you might have, my guess would be that coupling in your
boundary conditions does not get expressed. It is one of the age old
difficulties with Comsol and its predecessor femlab that coupling of
variables / modes in the subdomain settings is easy, but coupling in
the BCs is hard -- you have to explicitly "active" the coupling by
specifying weak boundary constraints are to be used.

The major complaint that users usually have is "singular matrix",
which can be caused by just about any conceptual or coding error.
Yours, unwanted solutions, is probably the exact opposite!

Regards,
Will
Post by Hamidreza Adib-RamezaniI
Hi,
Have anybody an experience about unwanted solutions using weak form?
Hamidreza
Hamidreza Adib-RamezaniI
2008-04-21 15:26:04 UTC
Permalink
Hi, Will

Thank you so much for your comments.
I study linear and non-linear Cosserat models using comsol. This is
highly nonlinear problem which engages the displacements as well as
the microrotations. As a matter of fact, I have a highly nonlinear
problem including coupling (microrotations and displacements).
According to the mathematical point of view, I know that the problem
has a unique solution and it is stable. So, I know perfectly what I
will obtain (The mathematical issue is completely treated).


I have verified all possibilities that you mentioned in your message.
I found out that the problem comes from boundary conditions (case
NO.3). For example, I apply a torque to the cylinderical bar and I
verify the calculated torque using comsol by boundary integration
(postprocessing). They are not the same!!!! The problem becomes worse
when I increase the torque value.

Would you please explain me about the modifications on the BCs? Should
I use Non-ideal or User-defined "Constrf"?

Best regards and thanks again
Hamidreza
Post by zimmerman_wbj
Hi Hamidreza,
You question is so open ended, that I thought it would engender no
(1) My usual experience with unwanted solutions is with highly
nonlinear problems which converge to solutions that did not match my
preconceptions of where the "real" solution is. I usually use the
parametric solver to "step along" the nonlinearity parameter to find
the bifurcation point.
(2) The other type of unwanted solution I get appears when the
boundary conditions are too prescriptive and kill off the desired by
say enforcing a symmetry.
(3) If I were to conjecture what type of "weak form and unwanted
solutions" you might have, my guess would be that coupling in your
boundary conditions does not get expressed. It is one of the age old
difficulties with Comsol and its predecessor femlab that coupling of
variables / modes in the subdomain settings is easy, but coupling in
the BCs is hard -- you have to explicitly "active" the coupling by
specifying weak boundary constraints are to be used.
The major complaint that users usually have is "singular matrix",
which can be caused by just about any conceptual or coding error.
Yours, unwanted solutions, is probably the exact opposite!
Regards,
Will
Post by Hamidreza Adib-RamezaniI
Hi,
Have anybody an experience about unwanted solutions using weak form?
Hamidreza
zimmerman_wbj
2008-04-25 23:59:14 UTC
Permalink
Try non-ideal constraints. The manual states that they frequently
work better than ideal constraints.

Also, try refining your grid. Probably the 4th most common cause
for "unwanted solutions" but might also explain your inconcistency in
imposed condition versus computed values of it.
Post by Hamidreza Adib-RamezaniI
Hi, Will
Thank you so much for your comments.
I study linear and non-linear Cosserat models using comsol. This is
highly nonlinear problem which engages the displacements as well as
the microrotations. As a matter of fact, I have a highly nonlinear
problem including coupling (microrotations and displacements).
According to the mathematical point of view, I know that the problem
has a unique solution and it is stable. So, I know perfectly what I
will obtain (The mathematical issue is completely treated).
I have verified all possibilities that you mentioned in your
message.
Post by Hamidreza Adib-RamezaniI
I found out that the problem comes from boundary conditions (case
NO.3). For example, I apply a torque to the cylinderical bar and I
verify the calculated torque using comsol by boundary integration
(postprocessing). They are not the same!!!! The problem becomes worse
when I increase the torque value.
Would you please explain me about the modifications on the BCs? Should
I use Non-ideal or User-defined "Constrf"?
Best regards and thanks again
Hamidreza
Post by zimmerman_wbj
Hi Hamidreza,
You question is so open ended, that I thought it would engender no
(1) My usual experience with unwanted solutions is with highly
nonlinear problems which converge to solutions that did not match my
preconceptions of where the "real" solution is. I usually use the
parametric solver to "step along" the nonlinearity parameter to find
the bifurcation point.
(2) The other type of unwanted solution I get appears when the
boundary conditions are too prescriptive and kill off the desired by
say enforcing a symmetry.
(3) If I were to conjecture what type of "weak form and unwanted
solutions" you might have, my guess would be that coupling in your
boundary conditions does not get expressed. It is one of the age old
difficulties with Comsol and its predecessor femlab that coupling of
variables / modes in the subdomain settings is easy, but coupling in
the BCs is hard -- you have to explicitly "active" the coupling by
specifying weak boundary constraints are to be used.
The major complaint that users usually have is "singular matrix",
which can be caused by just about any conceptual or coding error.
Yours, unwanted solutions, is probably the exact opposite!
Regards,
Will
Post by Hamidreza Adib-RamezaniI
Hi,
Have anybody an experience about unwanted solutions using weak form?
Hamidreza
Hamidreza Adib-RamezaniI
2008-05-02 19:42:14 UTC
Permalink
Hi, Will

I tried non-ideal constraints for all boundary conditions. It works
perfectly. So, there is no more unwanted solution for my linear
Cosserat models. It's great!


Thank you so much.
Best regards
Hamidreza
Post by zimmerman_wbj
Try non-ideal constraints. The manual states that they frequently
work better than ideal constraints.
Also, try refining your grid. Probably the 4th most common cause
for "unwanted solutions" but might also explain your inconcistency in
imposed condition versus computed values of it.
Post by Hamidreza Adib-RamezaniI
Hi, Will
Thank you so much for your comments.
I study linear and non-linear Cosserat models using comsol. This is
highly nonlinear problem which engages the displacements as well as
the microrotations. As a matter of fact, I have a highly nonlinear
problem including coupling (microrotations and displacements).
According to the mathematical point of view, I know that the
problem
Post by zimmerman_wbj
Post by Hamidreza Adib-RamezaniI
has a unique solution and it is stable. So, I know perfectly what I
will obtain (The mathematical issue is completely treated).
I have verified all possibilities that you mentioned in your
message.
Post by Hamidreza Adib-RamezaniI
I found out that the problem comes from boundary conditions (case
NO.3). For example, I apply a torque to the cylinderical bar and I
verify the calculated torque using comsol by boundary integration
(postprocessing). They are not the same!!!! The problem becomes
worse
Post by Hamidreza Adib-RamezaniI
when I increase the torque value.
Would you please explain me about the modifications on the BCs?
Should
Post by Hamidreza Adib-RamezaniI
I use Non-ideal or User-defined "Constrf"?
Best regards and thanks again
Hamidreza
Post by zimmerman_wbj
Hi Hamidreza,
You question is so open ended, that I thought it would engender
no
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
(1) My usual experience with unwanted solutions is with highly
nonlinear problems which converge to solutions that did not
match
Post by zimmerman_wbj
my
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
preconceptions of where the "real" solution is. I usually use
the
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
parametric solver to "step along" the nonlinearity parameter to
find
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
the bifurcation point.
(2) The other type of unwanted solution I get appears when the
boundary conditions are too prescriptive and kill off the
desired
Post by zimmerman_wbj
by
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
say enforcing a symmetry.
(3) If I were to conjecture what type of "weak form and
unwanted
Post by zimmerman_wbj
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
solutions" you might have, my guess would be that coupling in
your
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
boundary conditions does not get expressed. It is one of the
age
Post by zimmerman_wbj
old
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
difficulties with Comsol and its predecessor femlab that
coupling
Post by zimmerman_wbj
of
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
variables / modes in the subdomain settings is easy, but
coupling
Post by zimmerman_wbj
in
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
the BCs is hard -- you have to explicitly "active" the coupling
by
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
specifying weak boundary constraints are to be used.
The major complaint that users usually have is "singular
matrix",
Post by zimmerman_wbj
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
which can be caused by just about any conceptual or coding
error.
Post by zimmerman_wbj
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
Yours, unwanted solutions, is probably the exact opposite!
Regards,
Will
Post by Hamidreza Adib-RamezaniI
Hi,
Have anybody an experience about unwanted solutions using
weak
Post by zimmerman_wbj
form?
Post by Hamidreza Adib-RamezaniI
Post by zimmerman_wbj
Post by Hamidreza Adib-RamezaniI
Hamidreza
Loading...