Thank you for reaching out.
As you mentioned, in this case we are pushing the limits of our simulations and dipolepower won’t be reliable in this case. You can find the approached that inspired me here with useful links and also our KB example here.
In short, the box approach is the only reliable approach and it should work, however implementing it might be challenging and requires some considerations. I tired to do some convergence testing in 2D with surrounding air, source wavelength at 1550 nm and then running a sweep over mesh sizes for different trnas-box sizes. At this point, mesh2 was disabled to do some quick runs. Here is my simulation file:
test_dipole_BK.fsp (293.9 KB)
The mesh override region (mesh1) overlaps with the trnas-box analysis that is used to extract the dipole transmission data (i.e. mesh1 override has the same geometry as trans-box). So, lets assume that you want to measure the transmitted power for three different box sizes (10nm10nm, 5nm5nm, 2nm*2nm) while sweeping over mesh size. So for example if your mesh size is 0.2 nm, you will have 50, 25 and 10 mesh cells in your trans-box region, respectively. Here are the results:
region 1: Many thanks to @kchow, it looks like the mesh discrepancy between a dense mesh around the dipole (mesh1) and relatively coarse mesh for the remaining simulation region (mesh2 was disabled) results into some reflection or numerical errors, and as a result the measured transmission is less than 1 here. While this is not a limit to trans-box approach, it seems that problem comes from mesh discrepancy. We run simulations by defining a second override mesh (mesh2) to decrease the mesh discrepancy in the FDTD simulation region and we could get results close to 1 in region 1 using the trans-box (my attached simulation file includes the modifications, but it might take you a while to run it as it uses a very fine mesh). This means that even the trans-box approach works fine for very small mesh sizes, you need to have an extra attention on other issues that might arise. However, with proper meshing and consideration, region 1 should show transmission of 1.
region2: In our setup, this data are more reliable, and we can get results around 1. The small peaks are probably caused because of some other numerical errors.
region3: The mesh becomes very coarse that trans-box sits inside the dipole source region (generally has a size of 2 mesh * 2 mesh), and trans-box monitors can’t capture all the transmitted light. This is more obvious for a 2 nm *2 nm box.
I got a similar results for a trans-box of 1 nm * 1 nm (I guess this is what you are trying to simulate) as 2 nm * 2 nm case:
I wasn’t quite sure what QY stands for but I guess as long as you have enough mesh cells inside your transbox and you do a proper meshing for the rest of simulation region, you should get appropriate and reliable results.
I hope this could give you some hint performing your simulation.