About Autoshutoff Level

[bilal.barut.2011]

Hi, I have some questions regarding the autoshutoff level. In the knowledge base it is written that the autoshutoff level was an indication of how much energy left in the simulation region. Therefore;

1. Can we say that all simulations including some metal structures and open boundaries (PML) has to reach the autoshutoff minimum for sufficiently long simulation times? Because I think all energy has to leave the simulation region through open boundaries eventually.

2. Can we see the autoshutoff level as an indication of reliability of the simulation (because of question 1 above)? What shall I do if the simulation ends when autoshutoff level at 1.2 or 0.8 for example.

3. When I check the log files of some simulations I see the autoshutoff level first decreases and then starts to increase. Why the energy inside the simulation region starts to increase? I cannot be sure whether such simulations will diverge if I set the simulation time to a much higher value. Some of my simulations take 20-30 hours and it really hard to increase the simulation time and try to see what will happen.

Oscillations of autoshutoff level really confuses me. I would be very glad if someone helps.

[mbenes]

Q1: Can we say that all simulations including some metal structures and open boundaries (PML) has to reach the autoshutoff minimum for sufficiently long simulation times? Because I think all energy has to leave the simulation region through open boundaries eventually.

A: Yes, generally speaking, the simulation time must be long enough to reach the autoshutoff level if there is PML and/or absorbing material involved. There are some exceptions such as high Q cavities and bandstructure calculations where we can get reliable results without reaching the autoshutoff, but those are rather odd cases with very specific methodology.

Q2: Can we see the autoshutoff level as an indication of reliability of the simulation (because of question 1 above)? What shall I do if the simulation ends when autoshutoff level at 1.2 or 0.8 for example.

A: Unless this is expected behaviour(e.g. high Q simulation/resonant cavities), this would indicate that your simulation is either too short and you should use longer simulation time or that the PML might not be properly absorbing the incident light(e.g. light entering PML under a steep angle)

Q3:> When I check the log files of some simulations I see the autoshutoff level first decreases and then starts to increase. Why the energy inside the simulation region starts to increase? I cannot be sure whether such simulations will diverge if I set the simulation time to a much higher value. Some of my simulations take 20-30 hours and it really hard to increase the simulation time and try to see what will happen.

A: Small up and down movement of the autoshutoff level is quite common. As you mentioned, the purpose of autoshutoff is to estimate the remaining energy in the simulation region. The calculation is simplified to avoid severely impacting the calculation time. As a result, it does not account for the material as the light moves through the simulation region. hence you might observe that sometimes the autoshutoff goes from 0.1 to 0.2 and it does not necessarily mean something is wrong. Of course, if the autoshutoff goes from 0.1 to 5, you are most likely looking at diverging simulation

I would recommend the following article to read about high Q simulations that nicely explains why it is not necessary to wait for full field decay if you chose the appropriate methodology(decay slope based calculation).

[bilal.barut.2011]

Dear Miroslav Benes,

I really appreciate your help, thank you very much.

I try to simulate a split ring resonator. I need to see the response of only one resonator. If I use PML boundary conditions for my structure autoshutoff level never goes to a small value (maybe 0.04 or 0.01). But If I use periodic boundary conditions and set the FDTD region very wide (to eliminate the coupling effects) autoshutoff level goes to a very small value. What is the reason for this? Is what I do correct, can I simulate only one resonator using periodic boundaries with a very large FDTD region?

One more question. I want to fabricate a THz resonator on Si/SiO2 substrate. But below 0.5 there is no experimental data in the Lumerical. Can we model Si and SiO2 with constant parameters below 0.5 THz (like we model gold with PEC at these freq. values)

[fgomez]

Hi,

These are some comments and suggestions:

I try to simulate a split ring resonator. I need to see the response of only one resonator. If I use PML boundary conditions for my structure autoshutoff level never goes to a small value (maybe 0.04 or 0.01). But If I use periodic boundary conditions and set the FDTD region very wide (to eliminate the coupling effects) autoshutoff level goes to a very small value. What is the reason for this? Is what I do correct, can I simulate only one resonator using periodic boundaries with a very large FDTD region?

What kind of source are you using in the single resonator simulation? A regular plane wave source? If that is the case you will have edge effects, which might be part of the issue. My suggestion would be to try a TFSF source, which allows you to calculate things like absorption, scattering and extinction cross sections of a single structure.

One additional suggestion: when you use PML boundaries for the single resonator simulation, make sure the distance between the structure and the PML is at least half the maximum wavelength in order to avoid evanescent fields at the PML.

One more question. I want to fabricate a THz resonator on Si/SiO2 substrate. But below 0.5 there is no experimental data in the Lumerical. Can we model Si and SiO2 with constant parameters below 0.5 THz (like we model gold with PEC at these freq. values)

I think it is probably OK to assume a constant refractive index. However, it is worth checking the literature to see if there is any refractive index data available in that regime, which you could use to create a sampled data material.

[wxlim1]

I am also quite confused about the auto shutoff level.

Am i right to say that there are 2 situations that indicates the energy has fully run through the simulation region?

1. I have a simulation that runs down to the auto shutoff level, but it terminates early at 9% with comments,
   “Early termination of simulation, the autoshutoff criteria are satisfied.” .

2. I have a simulation that runs down to 99%, but i see the auto shutoff level at 0.00176482.

[fgomez]

Hi @wxlim1,

In the first situation, the auto shutoff was triggered, as reported by the comment in the log file. You might want to take a look at this post for additional information:

Did my simulation stop before the auto shutoff minimum was reached?

When you run a simulation in FDTD the job manager window shows the auto shutoff level, which is a measure of the energy left in the simulation window. This value is used to determine if the calculation can be stopped before the end of the simulation time. In some cases it might happen that the job manager shows the value of the auto shutoff to be 1 or some value smaller than 1 but larger than the minimum from the simulation settings (1e-5 by default) and then the simulation suddenly stops. This …

In the second case, if the early termination comment does not appear in the log file, the simulation ran for the entire simulation time and the auto shutoff was not triggered. An auto shutoff level of ~0.001 at the end of the simulation might not be small enough (the default auto shutoff level to shut down the simulation is 1e-5), although this depends on the simulation. If you expect fields to decay significantly inside the simulation region (this is the typical behavior unless you have some high Q cavity or some structure that traps light very efficiently), you might want to try increasing the simulation time to check if the auto shutoff level decreases beyond ~0.001.

Hope this helps!

[HaddenJ]

Silly question about the autoshutoff level… is it possible to output the autoshutoff level in a sweep? I have noticed that in some of my sweeps some of the results converge, and some don’t. Obviously I can open each log file and find the shutoff level (or even write a script to) - but it that seems like a crazy low-tech solution. I noticed there is a command called simulationdiverged - which works in scripts/in the command window - but maddeningly it doesn’t seem to work in analysis groups - so it doesn’t seem to be able to access it within something which would be available to sweep.

Perhaps I’m missing something obvious - but presumably knowing whether a simulation converged/diverged would be useful to know within analysis scripts - there must be a way to get to it?

[trobertson]

Hey @HaddenJ

Maybe use getresult('FDTD','status'); this should give you an integer for the status after running the simulation. 1 = run to completion, 2 = ended due to auto-shutoff, 3 = diverged.

This post discusses similar topics.

[resolved] How to extract system run time

Hi, Sorry if this has been posted or it’s really obvious but I haven’t been able to find it. If I’m using the auto shutoff and the simulation finished before the allocated simulation time(fs) I’d like to be able to know the real/system time for the simulation. Is there a way to extract the duration and memory cost to the scripting environment? Thanks, Ken

Regards,