And Yet Another FEM-Based Mode Solver for Dielectric Waveguides

This paper introduces an open-source, full-vector finite element mode solver for dielectric waveguides that utilizes a mixed Nedelec-Lagrange discretization to accurately model hybrid modes while suppressing spurious solutions, achieving high precision and reproducibility in both MATLAB and Python implementations.

The Gist

Imagine you are an architect trying to design a tiny, invisible highway for light. This isn't a road for cars, but a waveguide—a microscopic channel carved into a chip that guides laser beams to carry data. Just like a real highway needs the right width and surface to keep traffic flowing smoothly, these light highways need precise dimensions to keep the light from leaking out or getting scrambled.

The problem? Calculating exactly how light behaves inside these tiny, weirdly shaped channels is incredibly hard. It's like trying to predict the path of a thousand bouncing billiard balls in a dark room, all while the walls are made of different materials.

This paper introduces a new, open-source calculator (a "mode solver") that helps engineers and students solve this puzzle. Here is the breakdown in simple terms:

1. The Problem: The "Ghost" Traffic

In the past, scientists used two main ways to calculate light paths:

• The Simple Way: They treated light like a single, straight line. This was fast but inaccurate for modern, high-tech chips where light behaves like a complex, twisting wave.
• The "Full-Vector" Way: They tried to model light in all its 3D complexity. This was accurate, but it often created "ghosts." In math terms, these are called spurious modes—fake solutions that look like light waves but don't actually exist in the real world. It's like your GPS telling you there's a bridge where there is only a river.

2. The Solution: The "Hybrid" Team

The author, Ergun Simsek, built a new tool that acts like a specialized construction crew using two different types of workers to build the solution:

• The Edge Workers (Nédélec elements): These workers are assigned to the edges of the tiny triangles that make up the map. They are experts at handling the "twisting" parts of the light wave (the magnetic field). They ensure the light flows smoothly around corners without creating those annoying "ghosts."
• The Node Workers (Lagrange elements): These workers stand at the points (corners) of the triangles. They handle the "straight" parts of the light wave (the electric field).

By mixing these two teams together, the calculator gets the best of both worlds: it's accurate enough to handle complex shapes and materials, but it filters out the fake "ghost" solutions.

3. The Tool: Open, Free, and Everywhere

One of the biggest strengths of this paper isn't just the math; it's the accessibility.

• No Black Boxes: Many powerful tools for this are expensive commercial software (like COMSOL) that you have to buy a license for. This tool is free and open-source.
• Language Friendly: It's written in Python and MATLAB, the two most popular languages for scientists.
• Cloud Ready: You don't need a supercomputer. You can run this code right in your web browser using Google Colab. It's like having a high-end physics lab in your pocket.

4. The Proof: Does it Work?

The author tested this new calculator against the "gold standard" (the expensive commercial software).

• The Result: The new tool was almost identical to the expensive one, with errors less than 0.05%.
• The Trade-off: It's slightly slower than the commercial giants (which have spent decades optimizing their code), but it's fast enough for research and teaching. It proves that you don't need a million-dollar budget to do high-quality photonics research.

5. Why This Matters

Think of this paper as releasing a free, high-quality blueprint for the next generation of optical chips.

• For Students: It's a transparent way to learn how light works without paying for expensive software.
• For Researchers: It allows them to test new ideas quickly and share their code with the world, speeding up innovation in things like faster internet, medical sensors, and quantum computers.

In a nutshell: This paper gives us a free, reliable, and "ghost-free" calculator to design the microscopic highways of the future, making advanced light technology accessible to anyone with a laptop and an internet connection.

Technical Summary

1. Problem Statement

The modal analysis of dielectric optical waveguides is fundamental to the design of integrated photonic devices. While scalar Finite Element Method (FEM) formulations are computationally efficient, they fail to accurately model hybrid modes in structures with high refractive index contrasts, strong anisotropy, or complex geometries.

Conversely, full-vector FEM formulations based on Maxwell's curl equations are necessary for accuracy but suffer from a significant numerical artifact: the emergence of spurious (non-physical) modes. These unphysical solutions contaminate the eigenvalue spectrum, making it difficult to distinguish true guided modes from numerical noise. Existing strategies to eliminate spurious modes (such as penalty functions or transverse-only formulations) often introduce other drawbacks, including unsuitability for lossy materials, increased computational complexity, or quadratic eigenvalue problems.

The paper addresses the need for a robust, open-source, and reproducible full-vector mode solver that effectively suppresses spurious modes while maintaining computational efficiency and accessibility for researchers and educators.

2. Methodology

The proposed solver utilizes a mixed Nédélec–Lagrange discretization of Maxwell's equations in the frequency domain.

• Mathematical Formulation:

  • The solver starts with the source-free Maxwell's curl equations for a waveguide invariant along the $z$-axis, assuming harmonic time dependence ($e^{j\omega t}$) and longitudinal propagation ($e^{-j\beta z}$).
  • Instead of solving the strong form directly, the method derives a weak formulation by splitting the electric field into transverse ($E_t$) and longitudinal ($E_z$) components.
  • Hybrid Discretization:
    • Transverse Field ($E_t$): Discretized using lowest-order Nédélec edge elements. These elements ensure tangential continuity across element boundaries and inherently satisfy the divergence-free condition, which is crucial for suppressing spurious modes.
    • Longitudinal Field ($E_z$): Discretized using first-order Lagrange (nodal/P1) elements.
  • This combination results in a coupled system of equations that preserves the self-adjointness of the operator while accurately modeling hybrid modes.

• Numerical Implementation:

  • Meshing: The transverse domain is triangulated using a Delaunay algorithm with specific seeding along material interfaces to ensure interface-aligned edges.
  • Matrix Assembly: The weak form leads to a generalized eigenvalue problem $A\mathbf{x} = \lambda B\mathbf{x}$, where $\lambda = \beta^2/k_0^2 = n_{eff}^2$. The element matrices are assembled using a symmetric three-point Gaussian quadrature rule, which is exact for the polynomial degrees involved.
  • Boundary Conditions:
    • Open/Radiative Boundaries: $E_z$ is set to zero on boundary nodes (first-order absorbing condition).
    • PEC Boundaries: Tangential electric field components (edge DOFs) are set to zero.
  • Solver: The reduced generalized eigenvalue problem is solved using the shift-invert spectral transformation with the ARPACK-based implicitly restarted Arnoldi method (via scipy.sparse.linalg.eigs in Python and equivalent in MATLAB). This allows for the efficient extraction of the lowest eigenvalues (highest effective indices) corresponding to guided modes.

• Post-Processing:

  • The solver reconstructs the full vector field from the edge and nodal coefficients.
  • It calculates the TE fraction ($f_{TE}$) to characterize the polarization state of each mode.
  • It computes overlap integrals for mode orthogonality and coupling analysis.

3. Key Contributions

• Open-Source Implementation: The solver is implemented in both MATLAB and Python, with specific support for cloud-based platforms like Google Colab. This enhances accessibility and reproducibility compared to proprietary commercial tools.
• Spurious Mode Suppression: By combining Nédélec edge elements for transverse fields with nodal elements for longitudinal fields, the formulation effectively eliminates spurious solutions without resorting to penalty functions that may fail in lossy or complex scenarios.
• Reproducibility and Flexibility: The code includes a built-in mesh generator capable of handling complex geometries (e.g., trapezoidal or irregular cross-sections) and supports various material models (including Sellmeier equations).
• Educational Utility: The clear structure and availability of the code make it a valuable tool for teaching computational electromagnetics and integrated photonics.

4. Numerical Results and Validation

The solver was validated against COMSOL Multiphysics, a commercial standard, using two distinct case studies:

• Case Study 1 (Si₃N₄ Waveguide):

  • Structure: Rectangular Si₃N₄ core ($n \approx 1.996$) in SiO₂ cladding ($n \approx 1.444$).
  • Results: The effective indices ($n_{eff}$) calculated by the FEM solver matched COMSOL with relative errors below 0.004% for the first four modes.
  • Convergence: A log-log plot of error vs. mesh quality (points per wavelength) demonstrated systematic error reduction as the mesh was refined, confirming the expected convergence behavior of the FEM.

• Case Study 2 (High-Index Contrast Silicon Waveguide):

  • Structure: Silicon core ($n=3.5$) in SiO₂ cladding, representing a high-index-contrast scenario where hybrid modes are prominent.
  • Results: The solver achieved relative errors below 0.05% compared to COMSOL across four modes.
  • Performance: The MATLAB implementation completed a simulation with ~241,000 degrees of freedom in approximately 9.5 seconds, comparable to the ~9 seconds taken by COMSOL for a similar mesh size.

5. Significance and Conclusion

This work provides a reliable, flexible, and accessible alternative to commercial solvers for integrated photonics research. While the computational performance does not yet match highly optimized commercial codes (which use adaptive meshing and advanced preconditioning), the proposed solver offers:

1. High Accuracy: Capable of resolving hybrid modes in high-index-contrast waveguides with errors on the order of $10^{-3}\%$.
2. Robustness: Effective suppression of spurious modes through the mixed Nédélec–Lagrange formulation.
3. Accessibility: Being open-source and cloud-compatible, it lowers the barrier to entry for students and researchers who may not have access to expensive commercial licenses.

The paper concludes that this tool is well-suited for educational purposes and research into integrated photonic devices, with future work planned to include advanced preconditioning, parallelization, and support for anisotropic and nonlinear materials.