Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
Use of global equations - An unexpected behavior depending on equation sign
Posted 18 feb 2013, 07:55 GMT-5 1 Reply
Please login with a confirmed email address before reporting spam
I have submitted the following case in October 12, 2012 11:44....I do not yet get an acceptable solution. Maybe one of you has a solution to this.
Dear Support,
please have a look to my attached model.
This model solves the temperatures and thermal stresses in a beam cross-section.
actually model work good. I introduced two global variables for axial deformation in z-dir and rotation about x-dir.
Since the cross-sectin should have axail force and moment balance intop(ts.sz)=0 and intop(ts.sz*y)=0
this are imposed on global equations.
PROBLEM AS FOLLOWS
The model works perfectly if the global equations are defined by with a minus.
w: -intop(ts.sz)
ry: -intop(ts.sz*y)
The model works perfectly if the global equations are defined by without a minus.
w: intop(ts.sz)
ry: intop(ts.sz*y)
Since the value should be zero, it is sign must be not important... BUT I GOT TOTALLY DIFFERENT RESULTS..
ONLY THE ONE WITH MINUS IS CORRECT... it took many days to findout that :( Do you have any explanations?
One more remark: as I see very strange automatic scales for the variables (when they set to automatic)
MINUS case :
Scales for dependent variables:
mod1.u: 0.0091
mod1.T: 7.7e+02
mod1.ODE1: 9.3
PLUS case:
mod1.u: 6.7e-10
mod1.T: 0.0035
mod1.ODE1: 7.7e+02
I strongly suspect a bug
There should be no difference of roots if I solve f(u)=0, or -f(u)=0 or f(-u)=0, etc.
Regards
Dr.-Ing. Kaymak
Dear Support,
please have a look to my attached model.
This model solves the temperatures and thermal stresses in a beam cross-section.
actually model work good. I introduced two global variables for axial deformation in z-dir and rotation about x-dir.
Since the cross-sectin should have axail force and moment balance intop(ts.sz)=0 and intop(ts.sz*y)=0
this are imposed on global equations.
PROBLEM AS FOLLOWS
The model works perfectly if the global equations are defined by with a minus.
w: -intop(ts.sz)
ry: -intop(ts.sz*y)
The model works perfectly if the global equations are defined by without a minus.
w: intop(ts.sz)
ry: intop(ts.sz*y)
Since the value should be zero, it is sign must be not important... BUT I GOT TOTALLY DIFFERENT RESULTS..
ONLY THE ONE WITH MINUS IS CORRECT... it took many days to findout that :( Do you have any explanations?
One more remark: as I see very strange automatic scales for the variables (when they set to automatic)
MINUS case :
Scales for dependent variables:
mod1.u: 0.0091
mod1.T: 7.7e+02
mod1.ODE1: 9.3
PLUS case:
mod1.u: 6.7e-10
mod1.T: 0.0035
mod1.ODE1: 7.7e+02
I strongly suspect a bug
There should be no difference of roots if I solve f(u)=0, or -f(u)=0 or f(-u)=0, etc.
Regards
Dr.-Ing. Kaymak
Attachments:
1 Reply Last Post 3 mar 2013, 10:15 GMT-5
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
indeed I see 2 things:
1) changing from Global -intop1(...) to +intp1(...) does indeed displace the initial scaling values strangely, as if there is an index pointer issue and "T" gets the scaling value of "u" and ODE gets the scaling values of "T" !?
This looks very strange and suspicious, I agree, but still ...?
2) I never like to rename COMSOL's internal variables, so I replaced your w by my_w, rx by my_rx and wZ by my_wZ in the formulas and in the equations inside the physics node, then I could no longer get a solution, which indicates for me that the internal variables of COMSOL of similar names (even if not explicitly used for 2D might be making issues too: so my advice: never use variable names that can be mixed up with COMSOL's default ones (fur full 3D mode)
3) to come back to case 1) by changing the sign in the global equation you might change the sign estimate of the derivative (as both time initial conditions are all "0"), which could give some difference in convergence time, but in principle should not give any difference in results, if unique ?
Now I'm still scratching my head to fully understand what your formula tweaking is really doing, and as you have not added them in the energy etc ( COMSOL has very many other variables that should, in your case, be updated too ;)
I would propose, try a 3D model with some more realistic initial conditions, perhaps set up first by a stationary solver case, then run your time series. For the time series of diffusion problems, I mostly use time series in powers:
i.e. 2^{range(-8,0.5,8)}
I did my tests in 4.3a, but I did NOT rebuild fully your model, so I'm NOT sure I have a clean 4.3a version
--
Good luck
Ivar
indeed I see 2 things:
1) changing from Global -intop1(...) to +intp1(...) does indeed displace the initial scaling values strangely, as if there is an index pointer issue and "T" gets the scaling value of "u" and ODE gets the scaling values of "T" !?
This looks very strange and suspicious, I agree, but still ...?
2) I never like to rename COMSOL's internal variables, so I replaced your w by my_w, rx by my_rx and wZ by my_wZ in the formulas and in the equations inside the physics node, then I could no longer get a solution, which indicates for me that the internal variables of COMSOL of similar names (even if not explicitly used for 2D might be making issues too: so my advice: never use variable names that can be mixed up with COMSOL's default ones (fur full 3D mode)
3) to come back to case 1) by changing the sign in the global equation you might change the sign estimate of the derivative (as both time initial conditions are all "0"), which could give some difference in convergence time, but in principle should not give any difference in results, if unique ?
Now I'm still scratching my head to fully understand what your formula tweaking is really doing, and as you have not added them in the energy etc ( COMSOL has very many other variables that should, in your case, be updated too ;)
I would propose, try a 3D model with some more realistic initial conditions, perhaps set up first by a stationary solver case, then run your time series. For the time series of diffusion problems, I mostly use time series in powers:
i.e. 2^{range(-8,0.5,8)}
I did my tests in 4.3a, but I did NOT rebuild fully your model, so I'm NOT sure I have a clean 4.3a version
--
Good luck
Ivar