Discussion Forum

Hi

for me a 1e-19 compared to 1 is so small that you must be seeing rounding errors, or the limit of the binary number (differential) precision.

check the doc, for the sign of the index or the equations, as indeed there are two conventions +and -, I do believe COSMOL uses the negative, but I always check twice with the doc before I start any optics simulations, as there are several of these conventions

--
Good luck
Ivar

Ivar
Thanx for ur quick reply.


Samiul

Hi var,

do you know some probably reasons for getting the negative confinemnt loss (I found this is strange and using 3.5a Femlab)...is the accuracy the probably reason? thanks for reply!!!:)


best regard
huab

Hi

I have no definitive answer, but my suspicion goes to steep gradients or dirac type BC or initial conditions. In fact it's only since I'm really playing with COMSOL that I understand the importance of using always functions that can be derived once or better at least twice. Far back my math and physics professors always banged on the table with : " do not forget the hypothesis that functions must be continuous" while explaining PDEs, no I see it in practice.

In 3.5 you have the Heaviside functions to help you, in v4 these are also given as pre-cooked step and rectangle functions . Still one need to adapt the mesh or the "FEM sampling", as I like to call it, to these steep gradients

--
Good luck
Ivar

Hi

I have no definitive answer, but my suspicion goes to steep gradients or dirac type BC or initial conditions. In fact it's only since I'm really playing with COMSOL that I understand the importance of using always functions that can be derived once or better at least twice. Far back my math and physics professors always banged on the table with : " do not forget the hypothesis that functions must be continuous" while explaining PDEs, no I see it in practice.

In 3.5 you have the Heaviside functions to help you, in v4 these are also given as pre-cooked step and rectangle functions . Still one need to adapt the mesh or the "FEM sampling", as I like to call it, to these steep gradients

--
Good luck
Ivar

Hi ivar,

thanks much for your detailed reply.. I have a try to change something than have a look...I learned from somewhere that the PML mesh is better to use quad mesh, I am not sure about that just hera form somewhere`... is that right? another question is that if the pml is too thin,is that has possibility to cause the result of negative confinement loss?

thanks

Best regards

Hi

I count around 10% of main model dameter, but if RF at least several wavelength thick.

Then indeed I mesh the interiour, then I finish the PML shell with a Sweep or quad with at least 5-10 elements in thickness

I center the model in the PML airea/volume, check the PML type and origine (default "0" is not always the middle of the domain

I find the easiest way to make a PML region is to use the "Layer" feature (v4) of the surrounding "air" domain

--
Good luck
Ivar

Hi Ivar,

I have read your answer, but I still don't know how to solve my problem. My problem is that the imaginary parts of mode effective index (i.e. Confinement Loss) are sometimes positive and sometimes negative. From the mode field distribution, I can know these two modes must be fundamental modes (same mode filed distribution with two polarization). They also have the same real part. But the imaginary parts are different, one is negative and another one is positive.

What I have done:
1, When I use comsol 3.5a, the imaginary parts are both negative. but use the same model in 4.3, the results are different.
2, I though the loss is too small ( 1E-5 dB/m ) to have rounding error. I adjusted the tolerance in Solver from 1E-6 to 1E-10. But the results didn't change.

I am very confused with this problem. Positive imaginary part means this mode in the fiber has Gain rather than Loss. It doesn't make sense for a passive optic device --- Photonic Crystal Fiber.

If you need, I can upload the COMSOL file.
Please help me. Can you tell me where the errors are?

Thanks very much.

@Leon Lyman

Did you solve the problem of having inconsistent imaginary values ? how ?