General tips for making a simulation run faster/use less memory

mesh

#1
  1. Set up the computational resources to run the simulation in parallel making use of available multi-core/multi-processor resources. This can be set up in the resource manager.

  2. Use symmetry when possible to reduce both memory requirements and simulation time

  3. Disable movie monitors when you don’t need them since they tend to cause the simulation to run more slowly.

  4. Reduce the amount of monitor data that is collected
    a. Only record the field components that are required. This can be set in the monitor’s “Data to record” tab.
    b. Use only as many frequency points as necessary (more than 100 or 200 frequency points is often unnecessary)
    c. Choose to use 1D or 2D monitors when possible rather than 3D monitors
    d. In regions with very fine meshing, you can use spatial down sampling to reduce the spatial resolution of the data returned by the monitor. This can be set in the monitor’s “Geometry” tab. For example, if “down sample x” is set to 2, this will average the data over every 2 mesh points in the x-direction, returning only half as much data.

  5. Start with a relatively coarse mesh for initial simulations. We typically recommend using a mesh accuracy of 2 in combination with mesh override regions where necessary to increase the resolution of the structure where there are finer features. There is more information about this here: https://kb.lumerical.com/en/index.html?ref_sim_obj_non-uniform_mesh.html


ERROR Lumerical deletes elements after running a simulation
Electric field profile for a rib waveguide
Clever way to mesh a structure - discretized structure gives unphysical results