Learning Forum

[ejlee0911]

Hello.

I have to simulate optical cavities with a dipole source. When I set LC attribution for the bottom, a problem showed up.

I got this message below.

[Kyle]

It's possible that the memory could be the cause of this crash, even if the available memory is larger than the estimate. Can you try reducing the memory requirement of the simulation by removing monitors, reducing the number of mesh points or the span of the simulation, reducing the number of frequency points being recorded, etc. then running it again?

[ejlee0911]

Thank you, kjohnson. I had tried the lists you recommended. And I had reduced the memory, especially doing the span of the simulation.
I have some questions about PML type.

I thought it was still used much time when the structure is a simple cavity (only metal-insulator-metal).
I used a dipole source with a single wavelength and the dipole source is in the insulator. For simulation, I increase the simulation time (20000 fs) and reduce the auto shutoff min (5e-05). And doing the convergence test, I found the number of PML layers must be 42 at least because the alpha value is not 0. If alpha was set 0, simulation diverged. For simulation optimization, I changed the PML type. For associated files in some Lumerical support-articles (examples in circular polarization and phase convention (https://support.lumerical.com/hc/en-us/articles/1500006150981-Circular-polarization-and-phase-convention) and understanding dipoles in non-homogeneous materials (https://support.lumerical.com/hc/en-us/articles/360034382834-Understanding-dipoles-in-non-homogeneous-materials)) use the uniaxial anisotropic PML. I apply this PML type to my simulation. Using this type I got the results much faster than stretched coordinate PML. But in information (https://support.lumerical.com/hc/en-us/articles/360034382674-PML-boundary-conditions-in-FDTD-and-MODE), a legacy uniaxial anisotropic PML seems not to recommend. And same in the other forum (https://forum.ansys.com/index.php?p=/discussion/29899/why-is-the-fdtd-solusion-of-the-microdisk-resonator-not-converged). If a legacy is not recommended, what boundary condition is suitable for this case?

[ejlee0911]

I apologize for the misunderstanding convergence test.
Simulated again and there is no difference between PML types. It shows identical results so I think it is okay to use a legacy uniaxial anisotropic PML. As I mentioned above, I wonder why do not recommend using that PML type?

[Kyle]

From the release notes from when that PML type was released:
The newstretched coordinate PML(SCPML) uses a state of the art formulation [1] that incorporates many recent advances and delivers a number of advantages including better absorption and greater numerical stability when compared to the uniaxial anisotropic PML (UPML) provided in earlier versions of the software.
The new SCPML tends to have better performance than the UPML, but If the results are identical when you switch between them then it shouldn't matter which one you use.