This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
docs:become [2020/04/22 15:43] pklapetek |
docs:become [2020/04/22 22:18] pklapetek |
||
---|---|---|---|
Line 91: | Line 91: | ||
The image below shows the normalized angular dependence of the diffraction from the (finite size) grating. | The image below shows the normalized angular dependence of the diffraction from the (finite size) grating. | ||
- | {{:docs:become_grating_2d_normalised.png?600|}} | + | {{:docs:result_twomethods.png?800|}} |
- | When inspected in detail, it can be seen that there is a slight asymmetry in the result which needs to be analyzed, probably due to wrong placement of far-field points. The average intensity of the s+1 and s-1 diffraction orders is 0.178 of the incident intensity for the default silver model (not the Become one). | + | When inspected in detail, it can be seen that there is a difference of about 3 percents between |
+ | the two methods output for the first diffraction maximum, similarly to the PEC case. | ||
+ | Again we expect that this difference is given by the difference in models - big single calculation | ||
+ | has to work with limited source size and repeated calculations work with infinitely periodic field | ||
+ | even if they are not infinitely periodic. | ||
+ | A bigger problem is that the diffracted intensity is about 0.147 of the incident intensity, | ||
+ | which is less than expected (the expected value is 0.178). Something has to be wrong. | ||
+ | The first suspect is the refractive index. | ||
As we use the PLRC metal handling we can not enter the refractive index directly to the calculation. Instead we need to use some dispersive model, in our case based on two critical points. We have fitted part of the optical database data by the model to get closer to the prescribed values, however there are still small differences (unless we restrict the fitting spectral region much more, which would lead to quite unrealistic dispersive model). The correspondence of the fitted refractive index and database data is shown below. | As we use the PLRC metal handling we can not enter the refractive index directly to the calculation. Instead we need to use some dispersive model, in our case based on two critical points. We have fitted part of the optical database data by the model to get closer to the prescribed values, however there are still small differences (unless we restrict the fitting spectral region much more, which would lead to quite unrealistic dispersive model). The correspondence of the fitted refractive index and database data is shown below. | ||
Line 101: | Line 108: | ||
To compare its impact on results, here is a list of different metal model settings and results: | To compare its impact on results, here is a list of different metal model settings and results: | ||
- | * metal setting: 6 1.03583 0 1.37537e+16 1.25733e+14 2.1735 -0.504659 7.60514e+15 4.28131e+15 0.554478 -1.48944 6.13809e+15 6.62262e+14 means n=(0.129 + 3.87i) and leads to first order diffraction of 0.172 | + | * the fitted metal setting: 6 1.03583 0 1.37537e+16 1.25733e+14 2.1735 -0.504659 7.60514e+15 4.28131e+15 0.554478 -1.48944 6.13809e+15 6.62262e+14 means n=(0.129 + 3.87i) and leads to first order diffraction intensity of 0.147 |
- | * default metal setting: 6 0.89583 0 13.8737e15 0.0207332e15 1.3735 -0.504659 7.59914e15 4.28431e15 0.304478 -1.48944 6.15009e15 0.659262e15 means n=(0.036 + 4.147i) and leads to first order diffraction of 0.178 | + | * default metal setting: 6 0.89583 0 13.8737e15 0.0207332e15 1.3735 -0.504659 7.59914e15 4.28431e15 0.304478 -1.48944 6.15009e15 0.659262e15 means n=(0.036 + 4.147i) and leads to first order diffraction intensity of 0.150. |
+ | Error in the reference value (PEC reflection) is below 0.001 percent. | ||
- | After improving nearly everything (CPML, periodic borders, source) the result is even worse, 0.148 for the fitted metal, spacing 10 nm. However, now the result is fully symmetric and does not depend on voxel size (0.147 for spacing 5 nm). | + | |
+ | Another suspect is the voxel spacing. However, at least for 5 nm and 10 nm voxel spacing the resulting diffraction maximum was nearly same (less than 0.5 percent difference). | ||
===== 3D calculation ===== | ===== 3D calculation ===== | ||
- | The 3D calculation was a simple extension of the 2D case, so the voxel spacing was again 5 nm in every direction and computational domain size was 240x240x100 voxels. | + | To be completed after we are happy with the 2D case, should be simpler as 3D code is used for years already. |
+ | ===== Summary ===== | ||
+ | The sensitivity of 2D calculations results on the settings (computational domain size, time step, near-to-far field transformation, et.c) is in the order of few percent, the dominant effect of this uncertainty is the near-to-far field transformation. This does not affect the cases when these conditions are kept same in a series of calculations (so relative changes can be calculated with much higher accuracy), however it certainly affects the absolute values, e.g. when comparing a single calculation to completely different calculation or experimental data. | ||
===== TODO ===== | ===== TODO ===== | ||
Line 119: | Line 130: | ||
The aspects that need further investigation or tuning: | The aspects that need further investigation or tuning: | ||
* far-field calculation postprocessing speedup, now very slow, printing many debugging data. | * far-field calculation postprocessing speedup, now very slow, printing many debugging data. | ||
- | * cross-check metal refractive index (now relying on pre-fitted database data, which is wrong) | + | * evaluate the voxel size vs. speed vs. accuracy for the benchmark, or anything else to benchmark |
- | * there is still slight asymmetry | + | |
- | * evaluate the voxel size vs. speed vs. accuracy for the benchmark. | + | |