Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

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.

error of segregated solver

Please login with a confirmed email address before reporting spam

Dear Experts,

I used two models of electric currents and PDE of 2D problem.

The geometry is a rectangle with a circle in the center. The resistivity of the circle is very high. Therefore, the current density will be increased around the circle. The resistivity of the rectangle is related with the current density. Simply, if the current density is above super_jc, the resistivtiy is 10^6 ohm*m, or it is 1^-10 ohm*m.

For my physic model is a sequence, I use a segregated solver. I calculate current first and resistivity second. However, there is error

Undefined value found.
- Detail: Undefined value found in the equation residual vector.
There are 222 degrees of freedom giving NaN/Inf in the vector for the variable mod1.u

I put on my model and results. Is there anybody can tell me where the problem is?
Many Thanks!

Zhao


3 Replies Last Post 2010年12月14日 GMT-5 09:03
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 2010年12月14日 GMT-5 06:01
Hi

I believe those errors are rather linked to missing BC when you couple the physics

By the way isn't it easier (to read) to use a v4.1 "step" function for you shape of "u" ? but probably a "()^41" is still numerical stable

next your conductivities range 30 orders of magnitude !
but the binary number system hardly resolves correctly more than 6-8 taking into account that most physics equations are of second order type, this migh well lead to some errors too.
You might need to do some more tricks here and work in a log scale rather than a linear one

and one more: I do not really understand why you use a separate physics to write just u = my function, what about defining a variable in the main physics ?
But I havnt spent much time on your model so I might well have missed something (or your next quenching modelling step ;)

--
Good luck
Ivar
Hi I believe those errors are rather linked to missing BC when you couple the physics By the way isn't it easier (to read) to use a v4.1 "step" function for you shape of "u" ? but probably a "()^41" is still numerical stable next your conductivities range 30 orders of magnitude ! but the binary number system hardly resolves correctly more than 6-8 taking into account that most physics equations are of second order type, this migh well lead to some errors too. You might need to do some more tricks here and work in a log scale rather than a linear one and one more: I do not really understand why you use a separate physics to write just u = my function, what about defining a variable in the main physics ? But I havnt spent much time on your model so I might well have missed something (or your next quenching modelling step ;) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 2010年12月14日 GMT-5 08:42
Thanks for your reply!

I did not understand your meaning that "
but the binary number system hardly resolves correctly more than 6-8 taking into account that most physics equations are of second order type, this migh well lead to some errors too.
You might need to do some more tricks here and work in a log scale rather than a linear one "

In my physics, the transition of resistivity with current is very sharp. Is it the reason that caused the solver error?

You said that it was better work in a log scale. Do you mean that I design the geometry in log scale?

Many thanks!

Zhao

Thanks for your reply! I did not understand your meaning that " but the binary number system hardly resolves correctly more than 6-8 taking into account that most physics equations are of second order type, this migh well lead to some errors too. You might need to do some more tricks here and work in a log scale rather than a linear one " In my physics, the transition of resistivity with current is very sharp. Is it the reason that caused the solver error? You said that it was better work in a log scale. Do you mean that I design the geometry in log scale? Many thanks! Zhao

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 2010年12月14日 GMT-5 09:03
Hi

No, not the geometry, its your conductivity change going from 1E-10 to +E+20 that frightens me, the transition is indeed very steep, the numerical solver needs to do operations, take differences etc, long this gradient, if your gradient is too steep the binary representation of the real number, which has its limits, might result in numerical errors and you end up with wrong results.

The log scale idea is to replace sigma by log sigma, an correct the other equations to have a consistent system. But this means rewriting several equations, not sure the solver woud understand etc. So its a long way.

To start with try out with a smaller change from 1E-5 to 1E+5 or something thereabout and then increase until it crashes

--
Good luck
Ivar
Hi No, not the geometry, its your conductivity change going from 1E-10 to +E+20 that frightens me, the transition is indeed very steep, the numerical solver needs to do operations, take differences etc, long this gradient, if your gradient is too steep the binary representation of the real number, which has its limits, might result in numerical errors and you end up with wrong results. The log scale idea is to replace sigma by log sigma, an correct the other equations to have a consistent system. But this means rewriting several equations, not sure the solver woud understand etc. So its a long way. To start with try out with a smaller change from 1E-5 to 1E+5 or something thereabout and then increase until it crashes -- Good luck Ivar

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.