Author 
Topic 

zhc
China
3 Posts 
Posted  Oct 11 2021 : 15:49:34

Dear Enrico: Hi!When I run the Fastcap in my Linux,I used the input file in the sample.But I found that there is a big difference between the running result and the reference value. I want to ask if this is normal?
***************** the input file in Sample ***************** * 2D  coupled microstrips * * capacitance should be about C11 = 91.65 pF/m, C12 = 8.22 pF/m * * See C. Wei, R. F. Harrington, J. R. Mautz, T. K. Sarkar, "Multiconductor transmission lines * in multilayered dielectric media", IEEE Transactions on MTT, Vol. 32, No. 4, Apr 1985
C microstrip_top_0.03.txt 1.0 0.025 0.01 + C microstrip_bottom_0.03.txt 2.0 0.025 0.01
C microstrip_top_0.03.txt 1.0 0.025 0.01 + C microstrip_bottom_0.03.txt 2.0 0.025 0.01
D diel_microstrips_0.03.txt 2.0 1.0 0.0 0.01 0.0 0.0
C gnd_plane.txt 2.0 0.0 0.0
********* the running result ********* Capacitance matrix is: Dimension 2 x 2 g1_microstrip 8.79634e11 7.86901e12 g2_microstrip 7.87271e12 8.79361e11
According to the running result : C11 = 80.1pF/m C12 = 7.87pf/m
But in the input file :C11 = 91.65 pF/m, C12 = 8.22 pF/m
Is this normal?
Looking forward to your reply!Thanks!
from:zhang 

Michele Ancis
Austria
9 Posts 
Posted  Oct 11 2021 : 16:56:23

Hi Zhang,
while we wait for Enrico to reply, I am also learning to use FasterCap so my results refer to that software. I assume yours too, because FasterCap2 is the only software that solves 2D problems like yours. I have just run the file: coupled_microstrips_2d.lst
and my *last* iteration gives:
Capacitance matrix is: Dimension 2 x 2 g1_microstrip 9.33154e011 8.59175e012 g2_microstrip 8.58945e012 9.3312e011
Which is relatively close to the solution indicated in the file header, and is also different from yours.
A couple of thoughts:
1  I let FasterCap decide the simulation parameters automatically ("Automatically Calculate settings" is the topleft check box)
2  I don't look at the intermediate results of the simulation. Actually, the third iteration gives rise to a warning for nonnegative offdiagonal elements
3  I look at the last, final iteration, in my case iteration number 10.
4  I am running FasterCap on Windows, although I would ideally like to have it in Linux
Regards, Michele 


zhc
China
3 Posts 
Posted  Oct 12 2021 : 07:34:30

Hi Michele#65292;
Thanks! I run the inputfile in the Windows and the result is same of your.But we can see the result is also different with the solution indicated in the file header.Maybe automatic settings lead to the difference.
Let us look forward to Enrico's reply!
Regards, Zhang 


Enrico
447 Posts 
Posted  Oct 20 2021 : 12:16:01

Dear both,
about how to read the result values vs. the reference values in the comments at the top of the input file: you need to consider this is a Maxwell capacitance matrix. C11 in this case is the element at position (1,1) of the Maxwell capacitance matrix, NOT the selfcapacitance only (so you should not subtract the element at position (1,2) from the one at position (1,1) to compare with C11)
about the difference in values: please remember that the cited paper is dated 1985. If you look up the paper, you will see that it explicity states that the conductors have been refined with (only) 16 subsections and the dielectric interface with 10 subsections. This reflects the computational capabilities at the time. Now if you play with the manual settings of FasterCap to simulate the same coarse geometry, you will get approximately the same results. However note that in this paper the results of this example are compared with the results from other two papers, and the authors of the paper observed already at the time that "the difference should become smaller as we increase the number of our subsections"  so they had the doubt of not having really reached convergence. We can check that, so if we increase the number of sections (try playing with "m" parameter in manual mode, lowering it from 0.1 to 0.01 in steps of 0.01) you will see that the results have in fact not converged yet when at C11=91.65pF/m C12=8.22pF/m but continue to change until more or less reaching C11=93pF/m and C12=8.6pF/m and stabilizing there.
Best Regards, Enrico



Michele Ancis
Austria
9 Posts 
Posted  Oct 22 2021 : 17:44:24

Hi Enrico,
thanks first of all. I must say I can' t fully understand your comments about how to read the C11 and C12 values reported in the header of your example file, in comparison to the values output of FasterCap.
To me, they are *both* the same quantities, i.e. the elements of Maxwell's cap matrix.
In other words, FasterCap outputs
C11 , C12 C21 , C22
And we know that the ideal solutions in this case would give C21 = C12 and C11 = C22
So I would compare C11 from file header with the first row, first column number of FasterCap's output, and so on.

While I can imagine that there are errors tied to the discretization, I think here the interesting thing is that you get two different results by running on different Operating Systems. To me, this is a hint of the fact that the problem conditions are not homogeneous between systems, so that if you "tweak" the simulation in Linux, you will end up getting the same values as the same structure simulated in Windows, but this is not a given condition right off the bat. This is interesting to me because in the last instance I would like to run all these tools from the Linux platform, not Windows.
Regards, Michele 


Enrico
447 Posts 
Posted  Nov 12 2021 : 12:56:53

Hi Michele and Zhang,
on Michele's first point, we are actually saying the same thing. My comment was related to the calculation by Zhang in the first post of the thread, where he says: quote:
********* the running result ********* Capacitance matrix is: Dimension 2 x 2 g1_microstrip 8.79634e11 7.86901e12 g2_microstrip 7.87271e12 8.79361e11
According to the running result : C11 = 80.1pF/m C12 = 7.87pf/m
I.e. he is calculating C11 by subtracting element C12 from the capacitance matrix result (87.96pF  7.86pF = 80pF), which is not correct. C11 is 87.96pF.
On Michele's second point, I actually run the simulation under Windows and under Kubuntu on the same machine, and got the same results. Zhang, on which platform / OS did you run the simulation in Linux? Can you copy & paste here the output (so I can check the iterations?). In some cases there might be small differences in the numerical accuracy due to the different math libraries linked on different OS / or version of math libraries / or machine precision. Very tiny differences may still cause the Frobenious norm to trigger an additonal iteration, if very close to the threshold. This is not a normal case and anyway if this happens you should be close to convergence.
Best Regards, Enrico




Topic 


