Fundamentals | Pillar

Methods for modelling optical phenomena: ray tracing, FDTD, BPM, EME

Modelling optical phenomena is the numerical reproduction of light propagation and its interaction with matter — from macroscopic lens systems to photonic nanostructures. There is no single method that works everywhere: the choice between ray tracing and full electrodynamics decides the accuracy, the computational cost, and which effects you will even see in the result. This guide organises the methods — ray tracing, FDTD, BPM, EME, mode solver, RCWA and system simulation — and shows when to reach for which.

Methods for modelling optical phenomena: ray tracing, FDTD, BPM, EME

Two worlds: geometrical and wave optics

A first approximation when choosing a method comes from the question: how large are the elements of the system relative to the wavelength of light? That is a good engineering rule of thumb, but not a complete physical criterion.
- When the elements are much larger than the wavelength (lenses, reflectors, objectives), it is usually enough to treat light as geometric rays — the domain of ray tracing. Even then, wave effects can be significant: diffraction at apertures, resolution limited by the pupil, periodic structures. In those places ray tracing is supplemented with approximate models or, locally, with a diffraction method.
- When the sizes are on the order of the wavelength (waveguides, gratings, metasurfaces, nanostructures), a wave method is needed — amplitude, phase, polarisation, diffraction and interference become important. This does not always mean a full solution of Maxwell's equations by FDTD: for many such tasks BPM, EME, RCWA, mode solvers, or physical-optics methods are more appropriate and cheaper.
- Between these extremes lies physical / diffraction optics — scalar diffraction on the macro scale (laser-beam propagation, apertures, the Airy spot, coupling to a fiber).
A safer rule than the size-to-wavelength ratio alone: the more significant the wave effects are relative to the geometric scale of the problem, the more a wave or diffraction method is needed — and the choice of a specific wave method depends on the geometry, the directionality of propagation, and whether back-reflections matter.
We cover the fundamentals of ray tracing separately → What is ray tracing?. Here we focus on the map of methods.

Ray tracing (geometrical optics)

Tracing millions of rays through a 3D model. Two modes:
- Sequential — rays pass through surfaces in a fixed order. For imaging systems: aberrations, MTF, diffraction PSF analysis. Tool: OSLO; sequential ray tracing (Sequence Editor) is now also offered by TracePro.
- Non-sequential — rays hit any surfaces in any order, reflecting and scattering multiple times; the Monte Carlo method. For illumination, stray light, and optical efficiency. Tool: TracePro.
When: macro-optics, illumination, imaging systems, stray light analysis. Ray tracing is not a full-wave solution of Maxwell's equations — phenomena that require amplitude and phase information need appropriate models or wave methods. TracePro does, however, provide selected models and functions that let you account for specific diffraction effects (e.g. Aperture Diffraction), as well as ray splitting, reverse ray tracing, importance sampling, and scattering and fluorescence models.

FDTD — the full Maxwell's equations

The Finite-Difference Time-Domain method discretises space onto a Yee grid and propagates the E and H fields step by step in time. It solves Maxwell's equations directly and handles very complex geometries and materials — dispersive and nonlinear; a single run covers a broad band (the Fourier transform of the impulse response).
Cost: the grid resolution must faithfully reproduce the wavelength in the medium (on the order of ten to a few tens of points per wave), and the time step is limited by a stability condition (CFL). In three-dimensional FDTD the computational effort can grow very quickly as the grid is refined — with the spatial and time steps reduced together, the scaling can be roughly fourth-order. The actual size of the region you can compute depends on the resolution, the dimensionality (2D/3D), the materials, the number of time steps, and the available memory and hardware (CPU/GPU).
When: nanostructures, photonic crystals, metasurfaces, plasmonics, 2D/3D gratings, strong back-scattering, resonators. Tool: OptiFDTD; OptiOmega is a GPU-accelerated 3D FDTD solver with a mode solver, intended for larger tasks.
How it works and when exactly to use it — full description → How does the FDTD method work, and when should you use it?.

BPM — the Beam Propagation Method

Solves an approximate (paraxial or wide-angle) wave equation, propagating the beam step by step along one axis. It assumes that propagation is mainly in one direction and that the refractive-index contrast is small.
Advantage: orders of magnitude faster than FDTD for long structures. Limitation: it captures back-reflections, large angles, and strong contrast poorly or not at all.
When: long waveguides, directional couplers, Y-branches, tapers, MMI structures, mode analysis in fibers. Tool: OptiBPM.

EME — eigenmode expansion

EME (eigenmode expansion) splits the structure into sections that are invariant along the propagation direction; in each section the field is expanded onto a basis of eigenmodes, and the transitions between sections are described by scattering matrices (S-matrix). Unlike BPM, the method accounts for back-reflections and can be much more efficient than direct methods for long structures — particularly when the geometry changes in a limited number of sections. The cost depends among other things on the number of sections, the number of modes included, and how the structure varies.
When: adiabatic tapers, ring resonators, couplers, devices with reflections. Limitation: not very efficient for structures that are highly irregular along the propagation axis — FDTD is appropriate there.
Optiwave provides mode analysis in OptiMode and the propagation tools OptiBPM and OptiFDTD; EME as a method can be realised within specific design flows. It is worth knowing — for many long devices it can be more accurate than BPM and cheaper than FDTD.

Mode solver (FEM)

It does not propagate the field — it determines the eigenmodes, their propagation constants, and losses in a given waveguide or fiber cross-section. It can provide mode profiles and parameters used as input for propagation methods, including BPM and EME, and is the basis for analysing bend losses, inter-mode coupling, birefringence, and waveguide dispersion.
When: characterising a waveguide before a propagation simulation; designing specialty fibers (polarisation-maintaining, microstructured). Tool: OptiMode; for telecom fibers — OptiFiber.

RCWA — periodic structures

Rigorous Coupled-Wave Analysis (also known as the Fourier Modal Method) expands the field and the material profile into a Fourier series in the plane of periodicity and solves the eigenvalue problem layer by layer. Very efficient for periodic structures: diffraction gratings, sub-wavelength gratings, periodic metasurfaces, photonic-crystal layers.
RCWA is particularly intended for periodic structures; this does not, however, mean a simple split "periodic = RCWA, aperiodic = FDTD" — there are other methods for aperiodic structures too, and FDTD with periodic boundary conditions also computes periodic structures. In the portfolio: fiber gratings (Bragg / FBG and long-period) are modelled by OptiGrating; general periodic structures — OptiFDTD with periodic boundary conditions.

Physical and diffraction optics

Selected diffraction methods — the scalar Fresnel–Kirchhoff integral, angular-spectrum propagation, Gaussian Beam Decomposition — describe diffraction and interference on the macro scale without solving the full Maxwell's equations (physical optics also includes non-scalar methods). They form a bridge between ray tracing and electrodynamics: laser-beam propagation, apertures, spot analysis, depth of focus, coupling to a fiber.
When: systems with apertures on the order of thousands of wavelengths, where diffraction matters but FDTD is infeasible. Available among other things as a supplement to ray tracing in OSLO (diffraction PSF/MTF analysis).

System and circuit simulation

At the highest level of abstraction you do not model the field, but the signal parameters:
- OptiSystem — whole transmission links (WDM/DWDM, coherent, 400G+): power budget, OSNR, BER, Q-factor, dispersion, fiber nonlinear effects, EDFA amplifiers.
- OptiSPICE — simulation of optoelectronic circuits: it combines models of optical and electronic components (lasers, modulators, photodiodes, amplifiers) in a single circuit-level simulation.
When: assessing the performance of the whole system, once the geometry of the individual component is already fixed.

The hybrid approach and STOP analysis

A complete project usually combines methods:
- component (FDTD / EME) → section or circuit (BPM, OptiSPICE) → system (OptiSystem); each level verifies a different aspect.
- STOP (Structural–Thermal–Optical Performance) — coupling mechanical and thermal simulation (FEM) with optical: stresses and thermal expansion of the mounts can change the geometry of the system and the refractive index, which can lead to a degradation of imaging parameters such as MTF (the analysis may also show that the effect is negligible). Key in aerospace, space, and defence optics.
- CAD integrationRayViz lets you apply optical properties and run ray tracing directly in the SOLIDWORKS environment, and the model can then be passed to TracePro for a full analysis.

Quick cheat sheet

Task / phenomenonTypical scaleMethodTool
Imaging, aberrations, MTFmm–cmsequential ray tracingOSLO; also TracePro (Sequence Editor)
Illumination, stray light, optical efficiencymm and largernon-sequential ray tracing (Monte Carlo)TracePro
Nanostructure, metasurface, photonic crystalnm–µmFDTDOptiFDTD; OptiOmega (GPU, 3D)
Long waveguide, coupler, taper, MMIµm–mmBPMOptiBPM
Taper / resonator with reflections, long deviceµm–mmEME (+ mode solver)OptiMode + OptiBPM / OptiFDTD
Modes, bend losses, waveguide dispersionµmmode solver (FEM)OptiMode, OptiFiber
Diffraction grating, periodic metasurfacenm–µmRCWA / FDTDOptiFDTD
Fiber grating (FBG, long-period)µm–mmfiber-grating analysisOptiGrating
Laser-beam propagation, macro diffractionmm–cmphysical / diffraction opticsOSLO (diffraction analysis)
Transmission link (DWDM, 400G)system levelsystem simulationOptiSystem
Optoelectronic circuitcircuit levelcircuit-level simulationOptiSPICE

How to choose the method and the tool

Step by step — the scale of the system, the goal (imaging / illumination / transmission), the design stage, CAD integration, and the licensing model — we set out in a separate guide → How to choose optical simulation software?. If you are not sure where the boundary between methods runs for your system, describe the problem to us — we will point to the method and the package.

Často kladené otázky

How does ray tracing differ from wave methods?

Ray tracing (geometrical optics) treats light as rays and works well for elements much larger than the wavelength — objectives, reflectors, illumination systems. Wave methods (FDTD, BPM, EME) are based on Maxwell's equations — exactly or approximately — and capture amplitude, phase, diffraction, and interference, which is essential on the micro and nano scale. The starting point is the ratio of element size to wavelength, but ultimately what decides is how significant the wave effects are for the given problem.

When FDTD, and when BPM?

FDTD solves Maxwell's equations directly on a discretised grid and handles very complex geometries and materials (dispersive, nonlinear) and back-reflections. In three-dimensional FDTD the computational cost grows quickly as the grid is refined — with the spatial and time steps reduced together, the scaling can be roughly fourth-order — and the size of the region that can be computed depends on the resolution, the dimensionality, the materials, the simulation length, and the hardware. BPM is orders of magnitude faster for long, weakly guiding structures that propagate mainly in one direction (waveguides, couplers), but it omits strong back-reflections and large angles.

What is EME and how does it differ from BPM?

EME (eigenmode expansion) splits the structure into sections that are uniform along the propagation direction and joins them with scattering matrices. Unlike BPM, it accounts for back-reflections and can be much more efficient than direct methods for long structures, particularly when the geometry changes in a limited number of sections. It loses its advantage for structures that are highly irregular along the propagation axis, where FDTD is appropriate.

Which phenomena are the hardest to simulate?

Nonlinear phenomena in fibers, the interaction of light with strongly scattering media (e.g. biological tissue), and regions on the boundary between geometrical and wave optics, where both methods must remain physically consistent. They require hybrid algorithms and dense grids, which raises the computational complexity.

Do wave simulations require a GPU?

Not always. The hardware requirements depend on the size of the model, the grid resolution, whether the analysis is 2D or 3D, the simulation length, and the solver and whether it computes on the CPU or the GPU. For three-dimensional FDTD, GPU acceleration can substantially shorten the computation time — Optiwave provides such a solver in OptiOmega. This is not, however, a universal rule for all wave methods and all tools; for large photonic models, compute clusters or the cloud are also used.

Does iSymulacje help choose the computational method?

Yes — the SPECTROPOL / iSymulacje technical support in Polish covers choosing the method and module for a specific physical problem, help with installation and licence configuration, and interpreting the results. The portfolio comprises Lambda Research Corporation (TracePro, OSLO, RayViz) and Optiwave (OptiFDTD, OptiBPM, OptiMode, OptiGrating, OptiSPICE, OptiSystem, OptiFiber, OptiInstrument, OptiOmega).

Máte otázky k tejto téme?

Naši inžinieri vám radi pomôžu s výberom vhodného nástroja a prístupu k simulácii.