| T O P I C R E V I E W |
| lisa |
Posted - Aug 10 2012 : 22:15:20
Hello. I am new to FastCap2/FasterCap. I am wondering what is the difference between the two programs given the following capacitor problems I put to it:
1. I made a parallel plate capacitor with two infinitely thin plates (1m x 1m x 0m) separated by 0.1 m. I ran the list file with fastcap2. I then ran the list file with fastercap and generated a new list file compatible with fastcap2. I then ran this list file with fastcap2 again (pretty much following the instructions of the "Comparing FasterCap and FastCap2 results" tutorial video). Following the instructions from the Maxwell's Capacitance Matrix document, I got the capacitance values:
FastCap2 (original file)
C12=C21=88.64 pF
C11=C22=16.16 pF
FasterCap (original file, makes new file)
C12=C21=105.0 pF
C11=C22=22.5 pF
FastCap2 (new file made by FasterCap)
C12=C21=103.5 pF
C11=C22=22.7 pF
The values themselves aren't really of interest. I just noted that, like in the video tutorial, the capacitance values given with the FasterCap file are quite close (within 1.4% and 1.3%), but these values are very different from the original values calculated by FastCap2.
2. I followed the same procedure as above only instead of using infinitely thin plates, I gave them a finite thickness (1 m x 1m x 0.01 m) separated by 0.1 m. This time, the capacitance values were:
FastCap2 (original file)
C12=C21=109.5 pF
C11=C22=21.4 pF
FasterCap (original file, makes new file)
C12=C21=138.0 pF
C11=C22=16.1 pF
FastCap2 (new file made by FasterCap)
C12=C21=115.8 pF
C11=C22=23.1 pF
Now that I am looking at real, 3D plates, FasterCap and FastCap2 agree much less (only within 19.2% and 43.5%), and FastCap is much more self-consistent (despite getting the updated file from FasterCap).
So I guess my question is: Why are these two scenarios different? And, which capacitance should I trust?
Thanks so much in advance. Below, I've included some code so you can run what I did, in case my explanation wasn't clear.
Lisa
0 2Dplate plate.qui
*1m x 1m in xy plane, infinitely thin in z direction
*
Q plate1 0.0 0.0 0.0 1.0 0.0 0.0 1.0 1.0 0.0 0.0 1.0 0.0
G parallel plate capacitor
C plate.qui 1 0 0 0
C plate.qui 1 0 0 0.1
0 1mX1mX0.01m cube (n=1 e=0.1) plate_z_1m.qui
* xo = 0, yo = 0, zo = 0
* view from -x, -y, +z
* front left
*
*3D plate
*
Q 1 1.00000e+000 1.00000e+000 0.00000e+000 1.00000e+000 0.00000e+000 0.00000e+000 1.00000e+000 0.00000e+000 1.00000e-002 1.00000e+000 1.00000e+000 1.00000e-002
* front right
Q 1 0.00000e+000 1.00000e+000 0.00000e+000 1.00000e+000 1.00000e+000 0.00000e+000 1.00000e+000 1.00000e+000 1.00000e-002 0.00000e+000 1.00000e+000 1.00000e-002
* back left
Q 1 1.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 1.00000e-002 1.00000e+000 0.00000e+000 1.00000e-002
* back right
Q 1 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 1.00000e+000 0.00000e+000 0.00000e+000 1.00000e+000 1.00000e-002 0.00000e+000 0.00000e+000 1.00000e-002
* bottom
Q 1 0.00000e+000 0.00000e+000 0.00000e+000 1.00000e+000 0.00000e+000 0.00000e+000 1.00000e+000 1.00000e+000 0.00000e+000 0.00000e+000 1.00000e+000 0.00000e+000
* top
Q 1 0.00000e+000 0.00000e+000 1.00000e-002 1.00000e+000 0.00000e+000 1.00000e-002 1.00000e+000 1.00000e+000 1.00000e-002 0.00000e+000 1.00000e+000 1.00000e-002
G bigplatecap
*
*bottom plate
C plate_z_1m.qui 1 0 0 0
*
*top plate
C plate_z_1m.qui 1 0 0 0.1
|
| 3 L A T E S T R E P L I E S (Newest First) |
| Enrico |
Posted - Oct 31 2013 : 10:24:28
You can find the answer to your question also in other posts, since from time to time somebody asks it, so you can also find other examples.
In short, no the 'B' element was never implemented. The physical structures you want to simulate do have a thickness, and assuming zero thickness is only a simplification. You can, most of the times, provide a corresponding model; there are only few cases I encountered in the simulation space where the aspect ratios are so high that you cannot reasonably model the structure as full 3D.
The only tuning you have to do is to reduce the '-d' value in FasterCap, usually 0.1 is enough. Do not lower the value too much, otherwise the matrix will not be compressed at all and your simulation will last 'forever'.
Now, going through your questions one by one:
1. The dielectric 'D' statement models an interface, so there no real 'in' and 'out' from FastCap2 / FasterCap perspective. It is your responsibility as a user to guarantee that the dielectric is consistent (e.g. you don't have two parallel, adjacent panels, specifying different permittivity values on the same side). Said that, for convenience we say that the first permittivity value following the 'D' statement is the <outperm> while the second is <inperm>, but this is just to tell one from the other. Your reference point is assumed to lie on the <outperm> side, if you want to revert it, i.e. the reference point lies on the <inperm> side, you can add a '-' at the end of the line.
2. Absolutely not. The panels, conductor or dielectrics, MUST NEVER overlap.
3.1 You would need a 'B' element, as you suspect. So the answer is 'no'.
3.2 Please see FasterCap's sample file 'capacitor.lst'. You must run it in FasterCap, and you can read the comments inside the file for the details. You can also dump the refined output and run the simulaiton in FastCap2 as well. If you change the thickness, lowering it, don't forget to tune the -d value.
I also recommend reading the white paper "The Treatment of Dielectrics in FasterCap" that you can find in the 'Literature' section of the web site. It is not mandatory to be able to use FastCap2 / FasterCap, and it is a bit technical, but helps a lot to understand the mechanics under the hood.
Best Regards,
Enrico
|
| Max |
Posted - Oct 31 2013 : 02:17:51
Hi Enrico,
I'm just new here and get started with FastCap. I find this example very interesting and I read the examples in FasterCap & FastCap2 but still have some questions about representation of the contact surface of conductors and dieletric.
There is an example shows how to represent reference point of a dielectric. But I didn't find any examples of how to use "B" command. So I'll list my questions one by one:
1. If I understand the example correctly, I should just use a reference point which is inside the dieletric.
2. Is it possible to get a resonable result if I try to simulate a parallel capacitor with two infinitesimal thin conductors on the top and bottom and between them filled with a dieletric?
Example:
*lst file
C cond.qui 1.0 0 0 0
C cond.qui 1.0 0 0 1
D diele.qui 3.5 1.0 0 0 0 0.5 0.5 0.5
-------------------------------------------
*condutor.qui
*condutor
Q condDw 0 0 0 0 1 0 1 1 0 1 0 0
-------------------------------------------
*diele.qui
*top and bottom
Q diele 0 0 1 0 1 1 1 1 1 1 0 1
Q diele 0 0 0 0 1 0 1 1 0 1 0 0
*sides
Q diele 0 0 0 1 0 0 1 0 1 0 0 1
Q diele 1 0 0 1 0 1 1 1 1 1 1 0
Q diele 1 1 0 1 1 1 0 1 1 0 1 0
Q diele 0 0 0 0 0 1 0 1 1 0 1 0
3.
(1) If it does possible, how should I represent the contact surface between the conductor and the dieletric (Does the surface belong to conductor or dieletric)?
(2) If the simulation result with infinitesimal conductors ist not accurate, which means that I should use a conductor with resonable dimension (Length x Width x Height), how should I treat the contact surface? Could you please give me an example? In which case should I use "B ...." command?
Thank you very much!
Best regards,
Max |
| Enrico |
Posted - Aug 17 2012 : 00:43:11
Let's go one by one, and start with the simple thin plate capacitor.
Before that, however, let me re-cap here the main differences betweed FastCap2 and FasterCap, that will somewhat guide us in the discussion.
FasterCap has the following features, on top of what FastCap2 does:
- Support of lossy dielectric mediums
- Automatic mesh refinement
- Handling of very large models
- Out-of-core capability
- Speed-independency from non-uniform geometries in space
- Hierarchical input files support
- Charge density output for visualization in FasterModel
So let's look at the first results you got:
FastCap2 (original file)
C12=C21=88.64 pF
C11=C22=16.16 pF
FasterCap (original file, makes new file)
C12=C21=105.0 pF
C11=C22=22.5 pF
FastCap2 (new file made by FasterCap)
C12=C21=103.5 pF
C11=C22=22.7 pF
Looking at the numbers, and checking your input files, I see that the FastCap2 'original file' run was performed on the plain, non-discretized capacitor plates. This is a very crude approximation, since FastCap2 assumes constant charge over every input panel. So to consider varying charge over the plates, you need to discretize them. FastCap2, on the contrary from FasterCap, is not able to automatically handle the discretization step, nor to refine the geometry until convergence in the results is achieved.
On this topic, you can reference the original FastCap User's Manual, paragraph 2.4.1, where it is suggested in general to go on refining the discretization until the results differ of about 1% (the declared accuracy).
On the other hand, when you run FasterCap in automatic mode, it will perform two operations:
1) discretize the geometry up to a certain extent. This is not a simple meshing, but an intelligent refinement. First, there are different levels of discretization, of which you can visually appreciate only the finer one; but internally FasterCap considers the interactions of higher-level panels far from each other as direct. The benefit of this hierarchical discretization is more and more appreciable when you have structures at some distance from each other or stretiching along one dimension; this is what the fifth feature mentioned ("Speed-independency from non-uniform geometries in space") refers to. Second, curvature is considered and the refinement is more accurate where the curvature is sharp (again, over the whole hierarchy, so you may not see it looking at the finest level only).
2) solve the system, increase the discretization, and solve it again until convergence within 1% (or the desired threshold) is achieved. This process is costly, since it involves one full solve pass each time, but it is the same you would do manually with FastCap2. This is also why sometimes people comparing FastCap2 and FasterCap think that FasterCap is slower than FastCap2. A fair comparison must be done on the time needed to achieve the desired result within the required error threshold, at the minimum discretization level (including the "Direct potential interaction coefficient to mesh refinement ratio (-d)" value; more on this later on), not on the overall time to get the result, which includes many solve steps.
So going back to your example, the FasterCap-refined geometry leads to a converged result of 22.5 pF. This is done automatically thanks to the second feature mentioned ("Automatic mesh refinement"), as discussed. Now if you take the discretized geometry output by FasterCap and feed it to FastCap2, FastCap2 will benefit from a correct level of discretization to model the charge density within the required accuracy, and provide as well a converged result of 22.7 pF.
We can also have a visual idea of the difference between the first FastCap2 result and the converged result, by looking at the charge density pictures. This can be done using the last feature mentioned ("Charge density output for visualization in FasterModel"). In the first picture here below, you can see the charge density output by FasterCap for the non-discretized case of two plates. The charge density is uniform over the whole panel.
Then we can look at the charge density picture of the geometry providing the converged result. Charge is crowded along the edges, as expected. You could not simulate it with a single panel with constant charge.
Now let's discuss the second example. In this case, since you added a thickness and the sides of the electrodes, the starting geometry is somewhat more discretized. Let's look at the converged result. You can see that the charge is crowded along the sides. So in this case also the first run with FastCap2 does not give you a bad approximation of the converged solution.
Why then FasterCap is not so close to FastCap2's solution on the refined geometry? The full blown explanation is in another thread in the Forum ("microstrip coupler mesh refinement problem"), and I invite you to refer to that one. However, in summary, you have some 'extreme' ratio between the area of the electrodes and the distance of the top and bottom plates of the same electrode (about 1:100). To consider this ratio, you need to decrease the famous "Direct potential interaction coefficient to mesh refinement ratio (-d)" parameter. Looking into FasterCap Help you will find the explanation for the use of this parameter:
quote:
(it) controls the degree of matrix compression. FasterCap stores the panels set and their relationships ('links') in a hierarchical structure, representing the interaction matrix. The '-d' parameter specifies the compression of the links with respect to the panels mesh information as specified with the 'Mesh relative refinement value (-m)' parameter (either manually set or automatically calculated in the 'Auto' mode).
If you set this parameter to '1', you will get:
FasterCap (original file, -d = 1)
C12=C21=117.4 pF
C11=C22=22.9 pF
FastCap2 (new file made by FasterCap, -d = 1, 1344 panels)
C12=C21=116.1 pF
C11=C22=23.0 pF
your FastCap2 result (new file made by FasterCap, -d = 1, 832 panels)
C12=C21=115.8 pF
C11=C22=23.1 pF
I hope I have helped you
Best Regards,
Enrico
|
|
|
