Converged bending losses

Hello everyone,

I have problem with bending losses and converging my results to reasonable values.
My structure is given in attached file with parameters: r_core = 25 um, n1=1.45245, n2=1.450567055.
LMA_50_um_step_index.lms (244.0 KB)

Here is my converging test for simulation span region for 500 mesh and bend radius 5 cm.

Bend radius is quite small so I expected large losses. What is more for span 120 and 130 there is no more fundamental mode propagating. I prevented this by putting complex refractive index in “search” in FDE. But then there were two polarizations of fundamental mode with losses x dB/m and -x dB/m for TM and TE polarization respectively. It is similar behavior for other modes as well. Why is that happening and which mode is correct? What does negative value mean? And how can I choose correct simulation span for various structures (I am going to use script which will be sweeping core index and calculate bend losses)?

And last question: What is the best procedure for calculating bending losses? Sweeping with straight mode in d-card and coupling with bent waveguide (PCF LMA example on Lumerical site)? Or just setting bending radius and finding propagation loss of mode is enough? I noticed that the both methods give the same values, but I’d like to make sure.
I am aware that core size is quite big and maybe the problem comes from physics itself.

I would be grateful for any help.


I checked the simulation file and made a modification to set up the mesh based on maximum mesh step in combination with a mesh override region so that as the span of the simulation region is increased, the mesh will still have the same resolution near the core of the structure where the fields change more quickly over space.

Here is the modified file:
modified_LMA_50_um_step_index.lms (276.4 KB)

I started with the 130 nm span and was able to find the fundamental mode by using a combination of search near index 1.4527 with 50 trial modes. I think you were not able to find the fundamental mode before since as the size of the simulation region gets larger, more and more artificial modes propagating in the PML region are supported, so increasing the number of trial modes is necessary.

Since the refractive index contrast between the core and cladding is low, the fields are not well confined and extend very far out into the cladding region, and this makes it challenging to simulate because in order to prevent the interaction between the fields of the mode and the PML the span needs to be made very large. Here are the results I got:

X span (nm)		loss (dB/cm) 
130		        2.0723e-10
300		        1.1260e-10
500		        9.4642e-11
1000		    7.9867e-11
1500		    7.5830e-11
2000		    7.3946e-11

I found that increasing the y span didn’t make much difference to the result so I kept the y span at 100 nm.

To simulate the larger x spans, I did increase the radius of the cladding structure assuming that the cladding is much larger.

There can sometimes be negative values of loss due to the PML material introducing some small amount of gain, and this error due to PML is reduced as the span of the simulation region is increased, as discussed here:

To get the total loss for a bend, it’s important to include both the propagation loss within the bend as well as the mismatch due to light coupling from the straight to bent segments of the structure. The approach for including both sources of loss is discussed in this example:

Hopefully this helps!