Forum Message

 

 

We have moved the forum to https://forum.ngsolve.org . This is an archived version of the topics until 05/05/23. All the topics were moved to the new forum and conversations can be continued there. This forum is just kept as legacy to not invalidate old links. If you want to continue a conversation just look for the topic in the new forum.

Notice

The forum is in read only mode.

Apparent accuracy problem

More
2 years 7 months ago #3909 by point41
Hello,

I ran into an apparent accuracy problem with NGSolve, I wonder if there is some mistake in the way I am using NGSolve or else if there is a workaround for this.

To debug this, I made a simple test case with a known analytic solution: capacitance calculation of a parallel plate capacitor. The attached .py file creates two parallel plates 1 distance unit apart with an area 1x1 unit^2 each. The solution of the Laplace equation is a simple linear potential distribution, the field (-grad u) is a constant 1 between the plates and 0 outside.

The potential distribution calculated by NGSolve looks ok, the field defined as gfE.Set( -grad(gfu)) is less clean (a bit noisy) but otherwise doesn't look wrong. However, when I calculate surface integrals of normal field at the two electrodes I get ~0.5 instead of 1. This is how I calculate the integrals in the attached .py:

charge = (Integrate(sqrt((gfE*normals)**2), mesh, definedon=mesh.Boundaries(electrodes)))

Any idea what is going on and how to improve the accuracy of this calculation?

Thanks and regards,

Valery



File Attachment:

File Name: calculate_...08-18.py
File Size:5 KB
More
2 years 7 months ago #3910 by point41
Replied by point41 on topic Apparent accuracy problem
It seems that the problem is with the way integration is carried out. Integrate(sqrt((gfE*normals)**2), mesh, definedon=mesh.Boundaries(electrodes)) calculates the normal field integral along the interface between the electrode (Dirichlet boundary condition) and dielectric. Of course, inside the electrode the electric field is zero, so the interface integral seems to use the average of electric field 0 inside the electrode and non-zero field E on the dielectric side. Overall the effect is to lower the calculated field integral by 2x.

Is there a way to limit integration of normal field to just the dielectric side of the boundary and skip the electrode side? If not, I guess just multiplying the integral by 2x should produce the same result to reverse the apparent averaging of field outside E and inside 0 of the electrodes.

Best regards,

Valery
More
2 years 7 months ago #3911 by joachim
Replied by joachim on topic Apparent accuracy problem
Hi Valery,

it makes perfect sense what you get. The integration does the right thing.

You choose the H(div) - space (i.e. face elements) for the electric field, which admits a unique normal-component En on every face. The interpolation gfE.Set (grad(u)) does an arithmetic averaging of the normal component from both sides, the dielectric, and the electrode (where grad(u) is zero). From that you get half the value.

Solutions:
(A) Define the space for the E-field only in the dielectric, see attachment
or
(B) use only the dielectric region for the interpolation to the E-field. Then no averaging (with 0) happens.

I also gave a material name to the dielectric in you example file.

Best,
Joachim
The following user(s) said Thank You: point41
More
2 years 7 months ago #3913 by point41
Replied by point41 on topic Apparent accuracy problem
Hi Joachim,

Thanks very much for your help, your solution A makes sense and works well.

Best regards,

Valery
Time to create page: 0.150 seconds