Discussion Forum

Hi

You need materials for the domains for the main physics (and for all DOMAINS) but boundaries do not heritate material properties, so when you use "surface physics" attached to boundaries, you need to redefine a new material node for the specific BOUNDARIES where you apply particular BC physics

--
Good luck
Ivar

thank you very much

Mr. Kjelberg,

I have one more question about the materials.

When I add an insulator which has a conductivity of 0 to my model which is working in ACDC module, the program does not calculate, it gives an error of singular matrix.
However when I change the conductivity to a number quite close to 0 such as 1e-16, the program calculates safe and sound.

I did not see the reason. What is that?

I need to use air because of my model, however just because of this reason I can't get air as an insulator. Is it safe to use an insulator which does not have a conductivity of 0 in such a situation?

Thank you

Hi

which physics module are you using ? I'm pretty sure its MEF which solves for vecror A and scalar V.
If you put "0" as conductivity for a voltage field, you cannot define V hence COMSOL cannot identify the "unique solution" therefore , from the moment you give a small conductvity, it can again solve the equations homogenously.

Now I would not use 1e-16 as the conductivity of most materials are a few times 1E7 and the relative difference between 1E-16 and 1E7 becomes very great. This is not a physics issue but a math or rather a binary representation of real number issue. Your matrices become ill conditionned with relative ratios of small and large number above 1:1E7 or there about.
In fact you can check how you computer calculates, by looking at the value of "eps", the smallest integer COMSOL detect your floating point processor can distinguis a real number from 1 (in double precision). Now as most physics equations are of second order, the true ratio limitation is rather sqrt(eps), or 6-7 digits at best, today.
And notice the separation between numbers in a binary representation is not regular, and becomes coarser the larger the values

--
Good luck
Ivar

Thank you, Mr.Kjelberg.
Yes, I'm using MEF. Your response was indeed helpful. Thanks again.

Ceren

My questions come in bits and pieces. I am sorry but as I think, I come across some points that I cannot explain. I am sorry to take your time, Mr.Kjelberg.

You said that it is not smart to give so much small number since the ratio will excess the sqrt(eps), that will make the matrix system ill-conditioned.
However, when we have an intrinsic silicon in our model, the default value of conductivity is 1e-12 S/m. So, if the model consists of both copper and silicon materials, wouldn't be again the same ill-conditioned system?

How can I control my variables then? That means, my numerical values could be totally wrong, right?


thank you again
Ceren

Hi

normally COMSOL will tell you if it's "too wrong" or just not converge, or take long to -.
One way is to run a few cases with S=1 or 1000 and compare to a S=1E-12 case, and check how different your results are.

by the way this does not only apply t the conductivity, it's a general issue. That is why COMSOL has some algorithms to define "scaling" values for all he dependent variables (you see them in the log), but you can also tweak these manually, often it helps converging, but that means too that you know in advance more or less where the solution is going, or you have already done a few runs and you are fine tuning your model

I always leave COMSOL's default, then i take a detailed look ,and decide later to change or adapt, if I have some doubts or that I need to validate better my results

--
Good luck
Ivar

thanks a lot, Mr.Kjelberg.

--
physics undergrad