Learning Forum

[pengzheng97]

I made a optimization with lumopt, the result in 'working_dir' file as follows

[pengzheng97]

another result picture in here.

[Guilin Sun]

Do you mean you used the initial optimized object as the starting point for the 2nd optimization? if so, it seems the second one has some issues since it does not capture the optimal result at the beginning. Please make sure that the second one uses the same maximal, not the minimal optimization.

[pengzheng97]

No, what I mean is, I used 'script' to generate another simulation file, and added the same 'monitor', and then imported the optimized structure 'import' into this new file, the simulation results and the final optimized simulation file is different.

[Guilin Sun]

This is the same thing as I mentioned: you used the already-optimized objects as the beginning point of the second optimization, using "import". Right? please run the simulation before you do the optimization and see if the result is close to the first optimization. From what you sent, it seems it is far more away, from around 0.5 to 0.03. It might indicate that the first optimized result is very sensitive to the rounding of the structure. SO the first step is to compare the results between the original optimized and the imported structures.

[pengzheng97]

Yes, what I do is compare the results between the original optimized and the imported structures, the same waveguide, monitor, source and structure in the two simulation file, but the results are very different? I dont know why, I've had this problem before, but didn't notice it at the time.
thank you!

[Guilin Sun]

Before doing the second optimization, first do the final simulation again, by the use of the optimized structure, and see if the result is very close to the optimal one. Do not use other script, and use the original file as the template. Make sure to exclude as many as possible any possible mis-matches in the settings. Only when you can almost duplicate the result, would you do the second optimization.

[pengzheng97]

As you suggested, I added a structure to the 'forward_0' file optimized by lumopt. Compared with the original structure, the result of 'fom1' is the same.

But when I 'importnk2' this structure with the script the results are different



even they have the same structure


This really bothers me, can I send you the smulation files and scripts if needed?

thanks a lot!

[Guilin Sun]

importnk2(n,x,y,z);
SO you will need specify refractive index at the given point xyz. I believe there is problem with this assignment. What the refractive index you specified? did you use binary operation to extract where is the material and where is the background? Please use index monitor to compare them. In order to get the same result, the refractive index distribution and the values must be exactly the same.
Since your direct simulation works fine, the problem is from your import. Efforts should be made on the improvement of importnk2(n,x,y,z). Make sure the index profiles are the same.
I would suggest that you modify your first optimization to get better result, instead of using the first optimization as the initial point. This means the first optimization result is not optimal in you have any improvement. isn't it?

[pengzheng97]

The 'import_optimized' I added in the 'forward_0' simulation file was imported with importnk2, and the transmittance results obtained are the same as 'import'.
Maybe I found the difference, the refractive index in the 'forward_0' is
And in the script-made simulation is
one is 3.48, another one is 2.82, why it happen, I import it with the same script and 'structure.mat', I get the 'structure.mat' by transform the 'parameters_.npz' to '.mat' and get the 'eps' of the 'structure' type date in the matlab, the 'structure.mat' corresponding highest refractive index should be 3.48, why is it 2.82 in 'forward_0', is it the reason of VarFDTD?
should I send you these file to figure out the problem better?
thank you so much!

[Guilin Sun]

YouÒÇÇare using varFDTD, which simplifies the 3D simulation. When you got the first optimized result, it is 2D (3D collapses to 3D, and uses 3rd dimension confinement). However, if you import the result again to varFDTD, it recalculates the effective index via the varFDTD algorithm: Ansys Insight: MODE ÕêØÕ¡ªÞÇà´╝ÜÕà│õ║ÄvarFDTDþÜäÕÀÑõ¢£ÕăþÉåÚù«Úóÿ which might not be correct.
As I mentioned previously, if you do not trust the first optimization result, you should modify it instead of re-optimize it again. The second optimization means that the first result is not good, violates the optimization in the first run! I do not see any reason to do the 2nd optization using varFDDT or 2D FDTD. You may try 3D FDTD.