Pole-Expansion of the T-Matrix Based on a Matrix-Valued AAA-Algorithm

This paper introduces a computationally efficient, open-source method that utilizes a matrix-valued adaptive Antoulas-Anderson (AAA) algorithm to represent the frequency-dependent T-matrix as a pole-expansion, thereby overcoming the high memory and computational costs of traditional discrete frequency sampling while preserving physical interpretability.

The Gist

Imagine you are trying to describe how a complex musical instrument, like a grand piano, reacts when you hit its keys.

The Old Way: The "Photo Album" Approach
Traditionally, if a physicist wanted to know how a tiny particle (like a speck of dust or a nano-structure) scatters light at different colors (frequencies), they would have to take a "snapshot" of its behavior at every single color.

• If they wanted to know the behavior at 100 colors, they had to run a super-computer simulation 100 times.
• If they wanted 1,000 colors, they had to run it 1,000 times.
• The Problem: This is like trying to describe a whole movie by taking a photo every single frame. It takes forever, fills up your hard drive, and you lose the "story" of why the movie looks the way it does. You just have a pile of disconnected photos.

The New Way: The "Magic Recipe" (This Paper)
The authors of this paper have invented a clever shortcut. Instead of taking thousands of photos, they use a mathematical "magic recipe" (called a Pole-Expansion) to describe the particle's behavior.

Think of the particle's reaction to light not as a random mess, but as a song made of a few specific notes (resonances) and a quiet background hum.

• The "Poles" (The Notes): These are the specific frequencies where the particle really "sings" or resonates. It's like finding the exact notes a guitar string vibrates at.
• The "Residues" (The Volume): This tells you how loud each note is.
• The "Background" (The Hum): This is the quiet, steady part of the sound that isn't a specific note.

The Innovation: The "Group Chef" (TensorAAA)
Here is the tricky part: A particle doesn't just react to one color of light; it reacts to many different types of light (different angles, polarizations, etc.). In math terms, this is a giant grid of numbers (a Matrix).

• The Old Problem: If you tried to find the "recipe" for every single number in that grid separately, you'd get slightly different "notes" for each one. It would be like asking 100 different chefs to write a recipe for the same cake, and they all use slightly different ingredients. The result is messy and confusing.
• The New Solution: The authors used a smart algorithm called tensorAAA. Imagine this as a "Group Chef" who looks at the entire grid of numbers at once. Instead of writing 100 different recipes, the Group Chef finds one single set of notes that explains the behavior of the entire instrument perfectly.

Why is this a Big Deal?

1. Speed: Instead of running the computer simulation 10,000 times to get a smooth curve, they only need to run it about 50 times. The algorithm then "fills in the blanks" perfectly. It's like drawing a smooth curve through a few dots instead of measuring every inch of the line.
2. Memory: You don't need to store thousands of photos. You just need to store the "recipe" (a few notes and their volumes). This saves massive amounts of computer space.
3. Understanding: Because the method finds the actual "notes" (resonances), scientists can now look at the recipe and say, "Ah! This particle is acting like a tiny electric dipole here, and a magnetic one there." It reveals the hidden physics that the old "photo album" method hid.

Real-World Example: The "Ghost" in the Machine
The paper uses this method to study "Bound States in the Continuum" (BICs). Imagine a sound that is trapped inside a room but somehow never leaks out, even though the room has no walls. These are "ghost" states that are incredibly hard to find because they are so narrow and precise.

Using their "Group Chef" algorithm, the researchers were able to tune a grid of tiny cylinders until two of these "ghost" states merged into one. They could see exactly how the electric and magnetic parts of the light were swapping places to create this perfect, dual state. Without this new method, finding this would have been like trying to find a needle in a haystack by looking at the haystack one grain of sand at a time.

In Summary
This paper gives scientists a smart, efficient, and insightful way to predict how tiny objects interact with light. It swaps a slow, memory-hogging "photo album" approach for a fast, elegant "musical recipe" that reveals the true nature of the light-matter dance.

Technical Summary

1. Problem Statement

The Transition Matrix (T-matrix) provides a complete description of a scatterer's linear scattering response, mapping incident field coefficients to scattered field coefficients in a vector spherical wave (VSW) basis. It is fundamental for modeling multiple-scattering phenomena, metasurfaces, and complex particle clusters.

However, obtaining the T-matrix over an extended spectral domain (frequency dispersion) presents significant challenges:

• Computational Cost: The conventional approach requires evaluating the T-matrix at a finely resolved set of discrete frequencies. For systems with sharp resonances (e.g., bound states in the continuum or high-Q resonances), this necessitates thousands of full-wave simulations, which is computationally prohibitive.
• Memory Storage: Storing a separate T-matrix for every frequency point results in excessive memory usage.
• Physical Interpretability: Discrete sampling obscures the physical origin of spectral features. It does not explicitly reveal the underlying resonant states (poles) or their multipolar composition.
• Limitations of Existing Methods: Traditional pole-expansion methods often require solving nonlinear eigenvalue problems (finding resonant states directly), which can be unstable or require a large number of modes (including spurious PML modes) to accurately reconstruct the scattering background.

2. Methodology

The authors propose a data-driven approach to represent the frequency-dependent T-matrix, $T(k)$, as a pole-expansion using a matrix-valued variant of the Adaptive Antoulas-Anderson (AAA) algorithm, termed tensorAAA.

Core Concept: Pole-Expansion

Instead of storing discrete samples, the T-matrix is approximated as a sum of resonant contributions (poles) and a slowly varying background:
$$T(k) \approx \hat{T}(k) = \sum_{n} \frac{R_n}{k - k_n} + BG(k)$$
Where:

• $k_n$ are the complex poles (resonant frequencies).
• $R_n$ are matrix-valued residues (containing information about the radiating fields and coupling strengths).
• $BG(k)$ is a background term.

The Matrix-Valued AAA Algorithm (tensorAAA)

Standard scalar AAA algorithms approximate individual scalar transfer functions. Applying them element-wise to a T-matrix is inefficient and physically inconsistent because:

1. Different matrix elements would yield slightly different pole locations, obscuring the fact that a single physical resonant state couples to multiple radiation channels.
2. It requires storing and evaluating separate expansions for every matrix element.

The Innovation: The authors modify the AAA algorithm to handle multidimensional (tensor-valued) transfer functions.

• Shared Poles: The algorithm constructs a single set of poles ($k_n$) shared across all matrix elements of the T-matrix. This ensures physical consistency, as the poles represent the scatterer's intrinsic resonant states.
• Matrix-Valued Residues: The residues ($R_n$) are full matrices, capturing the coupling between different multipole orders and polarizations.
• Optimization: The algorithm uses a generalized barycentric rational form where scalar weights are optimized to minimize the error across all matrix entries simultaneously (stacking rows in the optimization problem).
• Input: It requires only a small number of direct T-matrix evaluations (samples) at discrete frequencies.

Analytic Continuation

To ensure the validity of the pole-expansion, the authors utilize the analytic continuation of the VSW decomposition integrals. By replacing complex conjugation in the surface integrals with incoming Hankel functions, they ensure the T-matrix elements are holomorphic in the complex plane, allowing for rigorous rational approximation.

3. Key Contributions

1. tensorAAA Algorithm: Development of a matrix-valued rational approximation algorithm that extracts a joint set of poles for the entire T-matrix, ensuring physical consistency and computational efficiency.
2. Open-Source Tools: Release of the diffaaable Python package (updated with tensorAAA) and a utility package baryTmat for storing and retrieving T-matrices in barycentric rational form.
3. Physical Insight Extraction: Demonstration that the residues of the pole-expansion directly yield the normalized Quasinormal Modes (QNMs) and their multipolar composition without solving eigenvalue problems or performing volume integrals.
4. Efficiency Benchmarking: Comprehensive comparison showing that the method achieves high accuracy with orders of magnitude fewer frequency samples than traditional discretization.

4. Results

The method was validated on two distinct scatterer geometries:

• Case 1: Tetrahedron of Spheres (Broken Symmetry)

  • A cluster of four spheres with different radii arranged in a tetrahedron.
  • Result: The T-matrix is densely populated due to broken symmetries. The tensorAAA converged rapidly (using ~50 samples) to a relative error below the numerical noise of the underlying FEM solver.
  • Efficiency: The method reduced the required frequency evaluations from ~90,000 (for fine sampling) to ~40.

• Case 2: Dielectric Cylinder (High Symmetry)

  • A cylinder representing a meta-atom in a metasurface.
  • Result: The T-matrix is sparse due to cylindrical symmetry. The method successfully identified the poles corresponding to z-oriented electric and magnetic dipoles.
  • Quasi-Dual BICs: The authors used the method to investigate Quasi-Bound States in the Continuum (qBICs). By tuning the lattice constant of a cylinder array, they forced two symmetry-protected BICs (one electric, one magnetic) to coincide.
  • Multipolar Analysis: Analysis of the residues revealed that the coinciding BICs possess an "almost dual" correspondence: their multipolar compositions are identical but with electric and magnetic multipoles interchanged. This level of detail was extracted directly from the T-matrix residues.

Performance Metrics:

• Storage: The pole-expansion requires storing $2(b^2+1)n$ floating-point numbers (where $b$ is the basis size and $n$ is the number of poles), compared to $2b^2 N_{dense}$ for discrete sampling. This offers a massive data efficiency gain ($N_{dense}/n$).
• Evaluation Speed: Once the expansion is computed, evaluating the T-matrix at arbitrary frequencies takes sub-millisecond times, compared to tens of seconds for full-wave solvers.

5. Significance

This work bridges the gap between computational efficiency and physical interpretability in nanophotonics and scattering theory.

• For Simulation: It drastically reduces the computational cost of frequency-dispersive T-matrix calculations, making high-resolution spectral analysis of complex multiple-scattering systems (like metasurfaces and disordered media) feasible.
• For Physics: It provides a direct route to extract Quasinormal Modes and their multipolar content from scattering data. This allows researchers to "see" the internal structure of resonances (e.g., identifying the specific multipole orders contributing to a BIC) without solving complex eigenvalue problems.
• Generalizability: While demonstrated for electromagnetic scattering, the framework is applicable to any wave phenomenon described by a transition matrix, including acoustics, seismology, and quantum scattering.

In summary, the paper introduces a robust, open-source framework that transforms the T-matrix from a computationally expensive lookup table into a compact, physically meaningful analytical function, enabling new insights into resonant scattering phenomena.