Troubleshooting Lumerical FDTD Accuracy: Settings That Prevent Wrong Results

What to do when an FDTD solve returns something impossible, such as transmission above one. Covers rechecking the basic settings, isolating the cause by simplifying the model in stages, visualising the fields, revisiting material fitting, and confirming with a convergence test.

Published18 October 2021

The main causes of inaccurate results `and errors in FDTD simulation`

FDTD simulation, it is not unusual to set up and run a simulation and get an unexpected result that looks inaccurate. Such results may be physically impossible (transmission greater than one, for instance) or may differ from results in another source such as a paper or another simulation. This may indicate that something in the simulation setup is wrong.

There are many possible causes of an inaccurate result, and these problems do not always produce the same symptom each time they occur. That means troubleshooting a simulation cannot be handled by one universal procedure. What follows are tips and strategies that should help identify the cause when an FDTD simulation returns an inaccurate result or an error.

When something goes wrong in an FDTD simulation, the following steps can help identify the cause of the error.

Rechecking the basic settings that determine whether the analysis is sound

Key Lumerical FDTD setting (Key settings in FDTD) and confirm that the simulation settings described there are correct. That article explains the settings that matter most in Lumerical FDTD simulation.

Consistency of mesh size, simulation time and simulation region

It explains, for example, how mesh, simulation time and simulation region are frequently the cause of errors in a Lumerical FDTD simulation.

Making use of official resources and the community

There are various online resources that help with troubleshooting a simulation. For example, Ansys Learning Forum (ALF) carries customer support cases on troubleshooting. Another user has very likely encountered the same problem, so when something goes wrong in a simulation we recommend searching ALF for a similar case.Application Gallery contains many simulation examples and explains how to approach various types of simulation. These examples include tips on setting up a particular type of simulation correctly.

Isolating the error by simplifying the model in stages

FWhen starting an FDTD simulation, we recommend beginning with a simple structure alongside the source and monitors. If the result for that simple structure is accurate, you can then increase the complexity of the simulation gradually, confirming at each step that the settings are correct. The same approach works for troubleshooting. When the full simulation gives an inaccurate result, reducing the complexity in stages lets you identify which part of the simulation or geometry is producing the error. For example:

Start by approximating the structure with a 2D simulation before extending to a full 3D FDTD simulation. A 2D simulation runs faster, which shortens the time spent developing the simulation.
Start with a single-wavelength simulation using non-dispersive materials before moving to a broadband simulation.
If you are measuring transmission or reflection from a patterned substrate, first simulate the unpatterned substrate and compare the FDTD result against the result from the stackrt command.

Detecting unexpected behaviour by visualising the fields

Movie or DFT monitors let you view the fields in the simulation, which helps identify where the error occurs. You can use them, for example, to find where the fields are diverging in a diverging simulation, or to judge whether the source is injecting the input pulse correctly.

Wavelength range limits and material fitting in broadband analysis

FDTD can produce broadband simulation results, but in some situations a simulation over a wide band does not work. It may be difficult to obtain a good material fit , not all wavelengths may be absorbed properly at the PML boundary, or the source may not inject light correctly (this last point can sometimes be improved by using multifrequency injection ). For these reasons the band should include only the wavelengths for which results can be computed. When using a wide band (more than a few hundred nanometres in optical simulation), splitting the band and running several simulations can resolve the problem.

Consistency between the injected source profile and the computational domain

It is common for the source field to be computed incorrectly or truncated by the edge of the source object. We recommend rechecking the field profile of the source before running an FDTD simulation. When you do, confirm that the source span is large enough that the mode field amplitude at the edge of the source is close to zero (around 10^{^-3}). Checking each component of the source electric field profile can also help confirm that the polarisation is correct. In the mode source below, for example, the span is too small to contain the whole mode field.

Securing final accuracy with a convergence test

Once the settings look broadly correct, you should carry out convergence testing 4 to finalise the simulation settings and confirm that the results are accurate. This is a particularly important step when comparing data and results against another source such as a published paper or a different simulation.