Computational Electromagnetics

Computational Electromagnetics

An introduction to Maxwell solvers, numerical formulation, observables, and their connection to real electromagnetic design problems.

This page follows a clear progression: 1D FDTD → 1D FEM → 2D FDTD → 2D FEM. The emphasis is on the physical meaning of each formulation, how the numerical model is constructed, what kind of output it produces, and how that output is used in engineering and industrial practice.

Core principle. In computational electromagnetics, the real output is not a colorful field image by itself. The real output is a quantity that can support design, diagnosis, comparison, or performance evaluation.
Maxwell equations 1D FDTD 1D FEM 2D FDTD 2D FEM Boundary conditions Observables Validation

Main focus

  • Explain the governing ideas behind FDTD and FEM.
  • Show how simple 1D models grow into 2D scattering and wave problems.
  • Clarify the role of sources, materials, and absorbing boundaries.
  • Connect simulation output to reflection, transmission, resonance, and field response.

Where these methods are used

  • Antennas, radomes, and scattering control in RF systems.
  • Waveguides, resonators, filters, and microwave packaging.
  • Photonic structures, coatings, thin films, and optical materials.
  • EMC, shielding, sensing, and device-level field management.

Contents

1. Maxwell equations and constitutive relations 2. Why computational electromagnetics? 3. Observables, not decorative field maps 4. Numerical progression used in this page 5. From numerical model to engineering application 6. 1D finite-difference time-domain (FDTD) 7. 1D finite element method (FEM) 8. 2D finite-difference time-domain (FDTD) 9. 2D finite element method (FEM) 10. Boundary conditions and absorbing boundaries 11. Practical simulation workflow 12. Validation and common failure modes 13. Example codes and mini-projects

1. Maxwell equations and constitutive relations

All computational electromagnetics starts from Maxwell’s equations. What changes from solver to solver is how these equations are approximated and what class of geometry or observable is targeted.

\[ \nabla \times \mathbf{E} = -\frac{\partial \mathbf{B}}{\partial t}, \qquad \nabla \times \mathbf{H} = \mathbf{J} + \frac{\partial \mathbf{D}}{\partial t}, \] \[ \nabla \cdot \mathbf{D} = \rho, \qquad \nabla \cdot \mathbf{B} = 0. \]

For linear isotropic media:

\[ \mathbf{D} = \varepsilon \mathbf{E}, \qquad \mathbf{B} = \mu \mathbf{H}, \qquad \mathbf{J} = \sigma \mathbf{E}. \]

In optics and broadband RF, the medium is often spatially inhomogeneous and may also be dispersive, so \(\varepsilon\) becomes either position-dependent, frequency-dependent, or both. This is precisely where analytical solutions become restricted and numerical electromagnetics becomes necessary.

Figure 1. Physics to prediction
Physics Maxwell materials Numerics mesh solver Prediction R, T S, Q, fields From equations to engineering output governing equations → discretization → measurable response
The same logic appears in industrial simulation workflows: model the physics, discretize it appropriately, and extract a response that informs design decisions.

2. Why computational electromagnetics?

Analytical solutions to Maxwell’s equations exist only for limited classes of geometry, material distribution, and boundary condition. Once the structure becomes spatially inhomogeneous, temporally broadband, or geometrically irregular, the electromagnetic problem must be approximated numerically.

Computational electromagnetics is not merely about obtaining a field plot. It is about converting a physical boundary-value or initial-value problem into a discrete system whose numerical solution approximates a measurable electromagnetic response.

Start from 1D

In 1D, the full structure of the numerical problem remains visible: the equations, the grid or elements, the boundary treatment, and the extracted response.

Move to 2D

In 2D, the model begins to represent diffraction, scattering, wavefront distortion, and geometry-dependent field redistribution.

Compare FDTD and FEM

The comparison makes clear how time-domain propagation and frequency-domain boundary-value analysis answer different classes of engineering questions.

3. Observables, not decorative field maps

Raw fields are necessary, but they are almost never the final answer. The actual target is an observable that can support interpretation, design, or validation.

Reflection / Transmission

Used in slab problems, coatings, filters, and waveguide interfaces.

S-parameters

Natural language of port-based devices in RF and microwave engineering.

Resonance / Q

Used for cavities, resonators, and geometry-dependent storage of energy.

Field response

Useful when the quantity of interest is local, but only when interpreted with proper normalization and geometry context.

\[ A = 1 - R - T \]
Do not confuse visibility with validity. A bright hot spot in \(|\mathbf{E}|\) may say little about matching, efficiency, or loss unless the observable extraction is mathematically correct.
Figure 2. Field map versus observable
Raw field map qualitative, local, cluttered post-process Observable S21 S11 quantitative, comparable, defensible
The student must learn to move from raw fields to a quantity that can be compared across meshes, methods, or experiments.

4. Numerical progression used in this page

The sequence below is intentionally narrow. The point is not to survey every electromagnetic solver, but to build numerical understanding through a controlled progression.

Method Domain formulation Discretization Main strength Main limitation Minimal example
1D FDTD Time domain Uniform staggered grid Broadband response from one pulse; excellent physical intuition Requires time stepping and absorbing boundaries for open problems Plane wave incident on dielectric slab
1D FEM Frequency domain Element-based weak form Clear boundary-value formulation; direct harmonic solution Broadband response requires repeated solves Layered dielectric interval
2D FDTD Time domain Structured Cartesian grid Natural for transient propagation, scattering, and pulse excitation Global meshing cost and CFL stability limit TM-like pulse scattering with and without ABC
2D FEM Frequency domain Triangular mesh Flexible geometry and local mesh refinement Less natural for broadband transient intuition 2D Helmholtz problem with dielectric inclusion
This progression is useful in both academia and industry: the solver is selected from the structure of the problem, the output that matters, and the type of design decision that must be made.

5. From numerical model to engineering application

The practical value of computational electromagnetics lies in the link between the numerical model and an engineering decision. A solver is used to estimate how a structure reflects, transmits, stores, radiates, concentrates, or suppresses electromagnetic energy. That estimated response then guides design choices in materials, geometry, thickness, spacing, packaging, or shielding.

RF and microwave

Antennas, radomes, waveguide components, filters, radar scattering, and EMC performance.

Electronics packaging

Interconnect coupling, signal integrity, package resonances, and field leakage in high-speed systems.

Optics and photonics

Thin films, resonant structures, absorbers, sensors, integrated photonics, and optical materials.

Materials and shielding

Field attenuation, absorber design, impedance matching, thermal-loading estimates, and shielding effectiveness.

In industrial work, the simulation is rarely the final product. The simulation supports a decision: which material stack to choose, where to refine a geometry, how to reduce reflection, how to control coupling, or whether a design meets a required performance window.
A solver does not automatically create engineering value. The value appears only when the model, the boundary conditions, the material data, and the extracted observable are all aligned with the real application.

6. 1D finite-difference time-domain (FDTD)

Method idea

Time domain • staggered grid

In 1D FDTD, electric and magnetic field components are staggered in both space and time. This produces the canonical leapfrog update scheme and makes wave propagation physically transparent to students at a very early stage.

\[ H_{i+\frac{1}{2}}^{n+\frac{1}{2}} = H_{i+\frac{1}{2}}^{n-\frac{1}{2}} + \frac{\Delta t}{\mu \Delta x}\left(E_{i+1}^{n} - E_i^{n}\right), \] \[ E_i^{n+1} = E_i^n + \frac{\Delta t}{\varepsilon_i \Delta x} \left(H_{i+\frac{1}{2}}^{n+\frac{1}{2}} - H_{i-\frac{1}{2}}^{n+\frac{1}{2}}\right). \]
\[ \Delta t \le \frac{\Delta x}{c} \]

Minimal example: dielectric slab

  • A Gaussian pulse is launched from the left.
  • The wave propagates through free space and meets a dielectric slab.
  • Part of the pulse reflects and part transmits.
  • Monitor signals are Fourier transformed to obtain \(R(\omega)\) and \(T(\omega)\), which are the same quantities used to assess layered devices, coatings, and material interfaces.
\[ s(t) = \exp\left[-\left(\frac{t-t_0}{\tau}\right)^2\right]\sin(\omega_0 t) \]
In FDTD, the mesh is not merely geometric. It controls phase error, dispersion, and the maximum stable time step.
Figure 3. 1D FDTD logic
slab incident pulse reflected transmitted monitor 1 monitor 2
A pulse, a slab, two monitors, and FFT post-processing already contain the full FDTD logic.
// 1D FDTD skeleton
for (n = 0; n < Nt; ++n) {
    update_H();
    update_E();
    inject_source();
    apply_boundary_conditions();
    record_monitors();
}
This type of model underlies practical work on dielectric stacks, transmission through layers, absorber concepts, and signal propagation through material interfaces.

7. 1D finite element method (FEM)

Method idea

Frequency domain • weak form

In 1D FEM, the governing equation is converted into a weak form and then assembled element by element into a global linear system. This is where students first see clearly that FEM is a boundary-value formulation rather than a time-marching scheme.

\[ \frac{d}{dx}\left(\frac{1}{\mu}\frac{dE}{dx}\right) + \omega^2 \varepsilon(x) E = f(x) \]
\[ \int_{\Omega} \frac{d v}{d x}\frac{1}{\mu}\frac{dE}{d x}\,dx - \omega^2 \int_{\Omega} v\,\varepsilon(x)\,E\,dx = \int_{\Omega} v\,f\,dx \]

Minimal example: layered dielectric interval

  • The domain is discretized into line elements.
  • Each element carries its own material parameter.
  • The local matrices are assembled into a sparse global system.
  • Boundary conditions complete the problem statement.
The contrast with FDTD is especially useful in practice: FDTD is often chosen for broadband pulse behavior, while FEM is often chosen when one operating frequency, a resonant condition, or a geometrically irregular device is the main concern.
Figure 4. 1D FEM logic
higher \(\varepsilon_r\) global matrix element by element assembly
FEM teaches that the mesh carries basis functions, material assignments, and boundary information into the final algebraic system.
// 1D FEM frequency-domain skeleton
assemble_global_matrix();
assemble_source_vector();
apply_boundary_conditions();
solve_linear_system();
postprocess_field();

8. 2D finite-difference time-domain (FDTD)

Method idea

2D time domain • propagation and scattering

Once the problem becomes genuinely two-dimensional, propagation is no longer just left-to-right. Waves can diffract, scatter from inclusions, and interact with the computational boundary in ways that are immediately visible.

\[ H_x^{n+\frac{1}{2}} = H_x^{n-\frac{1}{2}} - \frac{\Delta t}{\mu \Delta y}\left(E_z^n(i,j+1)-E_z^n(i,j)\right), \] \[ H_y^{n+\frac{1}{2}} = H_y^{n-\frac{1}{2}} + \frac{\Delta t}{\mu \Delta x}\left(E_z^n(i+1,j)-E_z^n(i,j)\right), \] \[ E_z^{n+1} = E_z^n + \frac{\Delta t}{\varepsilon}\left[ \frac{H_y^{n+\frac{1}{2}}(i,j)-H_y^{n+\frac{1}{2}}(i-1,j)}{\Delta x} - \frac{H_x^{n+\frac{1}{2}}(i,j)-H_x^{n+\frac{1}{2}}(i,j-1)}{\Delta y} \right]. \]

Minimal example: pulse scattering in 2D

  • A compact Gaussian source launches a transient field.
  • A dielectric inclusion or PEC object scatters the pulse.
  • With no absorbing boundary, edge reflections return into the domain.
  • With a simple ABC or PML-type layer, outgoing waves are suppressed.
This is the point at which students must stop treating boundary conditions as a coding detail. In an open-region problem, the boundary is part of the model.
Figure 5. 2D FDTD without ABC
outgoing waves hit the boundary and return What goes wrong • artificial standing waves • incorrect late-time energy • boundary reflections mistaken for physical scattering
Students should see this failure mode before they are given an absorbing boundary.
Figure 6. 2D FDTD with absorbing boundary
outgoing energy leaves the useful domain What improves • cleaner scattered field • better late-time behavior • boundary effects reduced • open-space logic restored
A pedagogical ABC already shows the essential idea, even before introducing full PML or CPML.
// 2D TM-like FDTD skeleton
for (n = 0; n < Nt; ++n) {
    update_Hx_Hy();
    update_Ez();
    inject_source();
    apply_boundary_conditions();   // none / Mur / PML-type
    record_snapshots_or_monitors();
}

9. 2D finite element method (FEM)

Method idea

2D frequency domain • triangular mesh

In 2D FEM, the geometry can be represented by triangles rather than a uniform Cartesian grid. This is where mesh flexibility becomes visible and students begin to understand why FEM is attractive for irregular geometries, boundary-value problems, resonators, and multiphysics coupling.

\[ \nabla \cdot \left( \nabla u \right) + k_0^2 \varepsilon_r(x,y)\,u = f(x,y) \]
\[ \int_{\Omega} \nabla v \cdot \nabla u\, d\Omega - k_0^2 \int_{\Omega} \varepsilon_r(x,y)\, v u\, d\Omega = \int_{\Omega} v f\, d\Omega \]

Minimal example: 2D Helmholtz problem with dielectric inclusion

  • The domain is split into triangles.
  • Material is assigned element by element from the element centroid.
  • A sparse linear system is assembled from local element matrices.
  • The solution reveals how geometry and material distribution shape the field.
The central lesson is not merely that FEM can solve the problem. It is that the geometry itself now participates in how the mesh is built and how the numerical error is distributed.
Figure 7. 2D FEM mesh logic
inclusion field response geometry and material are both embedded in the mesh
The triangular mesh is not decoration. It is the carrier of basis functions, material regions, and local approximation quality.
// 2D FEM skeleton
generate_triangular_mesh();
assemble_element_matrices();
assemble_global_sparse_system();
apply_boundary_conditions();
solve_frequency_domain_problem();
postprocess_field();

10. Boundary conditions and absorbing boundaries

Boundary conditions are not a secondary implementation detail. They are part of the mathematical statement of the electromagnetic problem. A finite computational domain must represent either a closed system, a symmetry plane, a periodic extension, or an approximation to open space.

PEC / PMC

Useful for ideal conductors or symmetry reductions, but physically inappropriate for open radiation problems.

Simple ABC

Useful pedagogically in 1D or simple 2D examples to show how boundary reflections can be reduced, though not perfectly eliminated.

PML / CPML

The standard tool for truncating open domains. Its job is to absorb outgoing waves without sending them back into the useful physical region.

Critical point. In engineering work, a poor boundary treatment can contaminate scattering, reflection, and radiation predictions. Open-region behavior must be represented deliberately, not assumed.

11. Practical simulation workflow

Figure 8. Workflow from physics to defensible data
Geometry what is built? Materials what is assumed? Discretization how is it resolved? Solver how is it solved? Observable what is compared? Errors in one stage propagate into the next stage
This is the workflow students should internalize: geometry → materials → discretization → solver → observable, with validation attached to the entire chain.
  1. Define geometry and its important scale hierarchy.
  2. Assign physically justified material models.
  3. Choose source, ports, and boundary conditions.
  4. Select the formulation that matches the problem class.
  5. Choose discretization consistent with wavelength and smallest critical geometric scale.
  6. Run the solver and extract observables rather than only plotting fields.
  7. Perform convergence, cross-checks, or analytical comparison.

12. Validation and common failure modes

Validation is not a decorative final slide. It is part of the method.

Analytical checks

  • 1D Fresnel reflection at a dielectric interface
  • Layered-medium intuition for the 1D slab
  • Mode or resonance checks in simple cavities

Convergence checks

  • mesh refinement
  • time-step sensitivity
  • domain-size sensitivity
  • boundary-distance sensitivity

Cross-validation

  • 1D FDTD vs analytical interface result
  • 1D FDTD vs 1D FEM for the same material profile
  • 2D FDTD without ABC vs with ABC
Model failure. Wrong \(\varepsilon\), ignored loss, unrealistic geometry, or a problem statement that does not match the intended experiment.
Discretization failure. Too coarse a mesh, unstable time step, poor element quality, or insufficient resolution at interfaces.
Boundary failure. Artificial reflection from truncated boundaries or a boundary condition that does not represent the physical setting.
Post-processing failure. Wrong normalization, early signal truncation, or comparing decorative field maps instead of observables.

Minimal convergence logic

\[ \delta_Q^{(k)} = \frac{|Q_{k+1}-Q_k|}{|Q_{k+1}| + 10^{-16}} \]

Replace \(Q\) by any meaningful observable: resonant frequency, transmission peak, reflected power, or local field norm. The point is not endless refinement. The point is to show that the chosen discretization no longer changes the answer in a consequential way.

Figure 9. Convergence behavior
mesh density observable value converged value
The student should learn to trust the plateau, not the first plausible plot.

13. Example codes and mini-projects

These example programs are intentionally compact. They show the numerical structure of each method clearly and can be used as a starting point for more realistic RF, microwave, photonics, EMC, shielding, sensing, and packaging problems.

Code 1

1D FDTD slab. Gaussian pulse, dielectric slab, monitor signals, and FFT-based reflection/transmission extraction.

Download fdtd_1d_slab.py

Code 2

1D FEM layered medium. Linear elements, piecewise material profile, global sparse solve, and field visualization.

Download fem_1d_layered.py

Code 3

2D FDTD without ABC. Shows how edge reflection contaminates an open-region transient problem when no absorbing treatment is used.

Download fdtd_2d_no_abc.py

Code 4

2D FDTD with Mur ABC. A simple absorbing-boundary implementation that already shows the logic of open-region truncation.

Download fdtd_2d_mur_abc.py

Code 5

2D FEM Helmholtz. Triangular-mesh assembly for a dielectric inclusion in a bounded domain.

Download fem_2d_helmholtz.py

Suggested extensions

  • extract reflection and transmission
  • compare no-ABC and ABC results
  • change dielectric contrast
  • refine mesh or time step
  • connect the result to a device use case
These codes are small enough to study line by line, but they already map directly onto practical workflows used in absorber design, waveguide analysis, antenna environments, EMC studies, photonic components, and package-level field prediction.

Back to Teaching