I have been trying to simulate the bending loss for a TM mode. I am not sure if the method is correct since the results are not convincing for me because I get the T_forward larger than the T_total in the expansion monitor.

I calculate the bending loss by dividing the T_forward by the input monitor transmission.

Please find attached the file I used to test the bend_loss. bend_radius_1.fsp (1.7 MB)

The reason why you are getting T_forward to be larger than T_total is because you are getting some non-negligible T_backward caused by PML back reflections. I noticed that you are using the “stabilized” PML profile, which is only recommended for situations with stability issues . This PML profile enhances the stability at the price of reducing the absorption per layer so it is common to end up with significant back-reflections from the PML if not enough PML layers are used. However, in this case we don’t have stability problems so we can use the “standard” profile, which is more absorptive. For more information about PML profiles please visit:

After switching to the standard profile, back reflections were reduced dramatically and now T_backward~1e-7 and T_total > T_forward. Please take a look at:.bend_radius_1_modFG.fsp (1.5 MB)
I also increased the z span of the simulation region to make sure the modal fields decay enough when they reach the PML. To speed up the simulation I reduced the mesh accuracy to 2.

Even though you can use this FDTD simulation to calculate the bending loss, there are more efficient ways for simple circular bends. You can use the Eigenmode solver FDE in MODE as described in this KB example. Using FDE is much faster and accurate because you can use a fine mesh to resolve the modes correctly. On the other hand, if you refine the mesh in the FDTD simulation the calculation time increases much faster. The results from your FDTD simulation show that the bending loss is very small, so to get accurate values you would need to refine the mesh and check the convergence of your results. You can do this much faster with the FDE method. FDTD would be useful for more complicated bends, for example if the curvature radius of the bend is not constant.