Rigorous foundations of adaptive mode tracking in single-parametric Hermitian eigenvalue problems: existence theorems, error indicators, and application to SAFE dispersion analysis

This paper establishes a rigorous theoretical framework and proposes an adaptive sampling algorithm for robust mode tracking in single-parametric Hermitian eigenvalue problems, specifically addressing challenges in SAFE dispersion analysis such as mode veering and degeneracy through novel error indicators and symmetry-aware subspace metrics.

The Gist

Imagine you are trying to map the "musical notes" a metal pipe or a composite wing can play. In physics, these notes are called dispersion curves. They tell you how sound waves travel through a structure at different speeds and frequencies.

To find these notes, scientists use a powerful computer tool called SAFE (Semi-Analytical Finite Element). Think of SAFE as a high-tech piano tuner that calculates the notes one by one.

However, there's a major problem: The notes get confused.

The Problem: The "Veering" Phenomenon

Imagine two runners on a track. Usually, they stay in their own lanes. But sometimes, they get very close to each other. In the world of waves, this is called mode veering.

When two wave modes get close:

1. They don't actually crash into each other (they "avoid" crossing).
2. Instead, they swap identities very quickly. It's like two dancers spinning around each other so fast that if you blink, you can't tell who is who anymore.

The Old Way (Uniform Sampling):
Previous methods tried to map these waves by taking steps of the exact same size, like walking across a field with a ruler.

• The Issue: In the easy parts of the field, you take big, lazy steps. But in the "dancing" zone (veering), those big steps are too wide. You skip right over the confusion, and your map ends up with the wrong notes connected to the wrong paths.
• The Fix (Old School): To be safe, people used tiny steps everywhere. This worked, but it was like using a microscope to look at the whole sky. It took forever and wasted a lot of computer power.

The Solution: The "Smart Step" Algorithm

This paper introduces a smart, adaptive algorithm. Instead of walking with a fixed ruler, imagine you are a hiker with a magic compass that tells you exactly when the terrain is getting tricky.

Here is how the new method works, using simple analogies:

1. The "Confidence Meter" (Error Indicator)

The algorithm doesn't just guess; it checks its own work. After it takes a step, it asks: "Am I 100% sure I'm still following the same wave?"

• It uses a tool called MAC (Modal Assurance Criterion), which is like a fingerprint scanner. It compares the "fingerprint" of the wave at the start of the step with the fingerprint at the end.
• If the fingerprints match perfectly (score near 100%), the step was safe.
• If the fingerprints look different (score drops), the algorithm knows: "Oh no! We just hit a 'veering' zone where the waves are spinning wildly. I need to slow down!"

2. The "Smart Refinement" (Adaptive Sampling)

When the confidence meter drops, the algorithm doesn't panic. It simply cuts the step in half and tries again.

• Analogy: Imagine you are walking through a foggy forest. In the clear parts, you walk fast. When you hit a thick patch of fog (the veering zone), you stop, take a tiny step, check your map, take another tiny step, and so on. Once you clear the fog, you go back to walking fast.
• Result: You get a perfect map of the whole forest, but you only took tiny steps where they were actually needed. This saves massive amounts of time and computing power.

3. The "Twin Dance" (Symmetry and Degeneracy)

There is a special case, like in a perfectly round steel pipe. Here, two waves are so identical (twins) that they are mathematically indistinguishable. They can spin around each other infinitely without changing the physics.

• The Old Problem: Trying to track them as two separate individuals is impossible because the computer gets dizzy and loses track.
• The New Solution: The algorithm realizes, "These two aren't separate; they are a single team." Instead of tracking two dancers, it tracks the entire dance floor (the subspace). It ensures the group stays together, even if the individual dancers spin around. This is called Subspace Tracking.

Why Does This Matter?

• Speed: It finds the answers much faster because it doesn't waste time taking tiny steps in easy areas.
• Accuracy: It never gets lost in the confusing "veering" zones, ensuring engineers don't design bridges or planes based on wrong data.
• Reliability: It gives a "confidence score" for every part of the map. If the score is low, the computer knows to double-check that specific spot.

The Bottom Line

This paper is like upgrading from a blindfolded walker with a fixed stride to a smart hiker with a GPS and a confidence meter. It knows exactly when to speed up and when to slow down, ensuring that the map of how sound travels through materials is always accurate, no matter how complex or confusing the terrain gets.

The authors have even made their "magic compass" (the code) free for everyone to use, so engineers can build safer, better structures without getting lost in the math.

Technical Summary

1. Problem Statement

The Semi-Analytical Finite Element (SAFE) method is a standard tool for computing guided wave dispersion curves in waveguides with arbitrary cross-sections. It reduces 3D wave propagation to a parameterized family of Hermitian eigenvalue problems. However, a critical bottleneck remains: robust mode tracking.

When computing dispersion curves, eigenpairs (eigenvalues and eigenvectors) must be correctly linked across consecutive wavenumber steps. This becomes notoriously difficult in mode veering regions (avoided crossings), where:

• Eigenvalues approach each other closely without intersecting.
• Eigenvectors undergo rapid rotation over a narrow parameter interval.
• Standard similarity measures, such as the Modal Assurance Criterion (MAC), fail if the wavenumber step size ($\Delta k$) is too coarse, leading to mode misassignment (jumping between branches).

Existing approaches suffer from three main limitations:

1. Uniform Sampling: Using a globally fine grid to resolve veering is computationally expensive and inefficient.
2. Lack of Confidence Metrics: Most methods provide deterministic assignments without quantifying the reliability of the tracking.
3. Symmetry-Protected Degeneracy: In structures with continuous symmetry (e.g., pipes), eigenvalues may coincide over entire intervals. In these degenerate subspaces, individual eigenvectors are not uniquely defined, causing pointwise MAC tracking to fail arbitrarily.

2. Methodology

The authors propose a rigorous theoretical framework combined with an adaptive numerical algorithm.

A. Theoretical Foundations

• Perturbation Analysis: The authors derive an explicit expression for the eigenvector derivative ($q'_i$) with respect to the wavenumber $k$:
  $$q'_i = \sum_{j \neq i} \frac{q_j^H K' q_i}{\lambda_i - \lambda_j} q_j$$
  This reveals that eigenvector sensitivity is inversely proportional to the eigenvalue gap ($\lambda_i - \lambda_j$). In veering regions, the gap is small, causing rapid eigenvector variation.
• Symmetry Protection (Wigner–von Neumann Rule): The paper proves that for single-parameter Hermitian systems, true crossings only occur if modes belong to different symmetry classes (coupling term vanishes). In these cases, eigenvector derivatives remain bounded, and coarse sampling suffices.
• Existence Theorem (Theorem 1): The authors prove that for any mode veering region with a positive eigenvalue gap, there exists a sufficiently small step size $\Delta k_{max}(k)$ that guarantees the correct mode is identified (i.e., self-MAC > cross-MAC).
• Subspace Tracking: For symmetry-protected degeneracy (e.g., axisymmetric pipes), the authors introduce a Subspace MAC. Instead of tracking individual vectors, the algorithm treats degenerate pairs as a single invariant subspace, using a rotation-invariant similarity measure (Frobenius norm of the subspace projection).

B. Adaptive Algorithm

The proposed algorithm automates the refinement of the wavenumber grid based on a posteriori error estimation:

1. Initialization: Start with a coarse uniform grid.
2. Tracking & Error Estimation: Compute eigenpairs and apply the Hungarian algorithm for optimal mode assignment. Calculate a MAC separation metric ($D_i$) and a global error indicator $\varepsilon(k, \Delta k) = 1 - \min_i D_i$.
   • $\varepsilon \approx 0$: High confidence (correct tracking).
   • $\varepsilon \to 1$: Low confidence (risk of misassignment).
3. Adaptive Refinement: If $\varepsilon$ exceeds a tolerance (e.g., 0.05) and the step size is above a minimum threshold, the interval is bisected, and a new eigenvalue solve is performed at the midpoint.
4. Degeneracy Handling: The algorithm automatically detects degenerate eigenvalue clusters and switches from pointwise MAC to Subspace MAC for those specific modes.

3. Key Contributions

1. Rigorous Existence Theorem: Proves that a sufficiently fine grid always exists to resolve mode veering, providing a mathematical guarantee for the feasibility of adaptive tracking.
2. Novel Error Indicator: Introduces $\varepsilon(k, \Delta k)$, a quantitative metric that not only drives adaptive refinement but also provides a confidence measure for the final dispersion curves.
3. Unified Subspace Framework: Solves the degeneracy problem in axisymmetric structures by treating degenerate pairs as invariant subspaces, enabling robust tracking where pointwise methods fail.
4. Adaptive Sampling Strategy: Demonstrates that adaptive refinement can achieve perfect tracking accuracy with significantly fewer computational resources compared to uniform fine sampling or continuation-based methods.

4. Numerical Results & Validation

The method was validated on four diverse waveguide geometries and compared against SAFEDC (uniform sampling) and the Dispersion Calculator (DC) (continuation-based):

• Symmetric Laminated Plate: Successfully handled symmetry-protected crossings without refinement. In veering regions, the adaptive method corrected misassignments found in coarse sampling, achieving accuracy comparable to dense uniform sampling (98 points vs. 350 points).
• Unsymmetric Laminated Plate: A case of pervasive veering with no symmetry protection. The adaptive method correctly tracked all modes (185 points), whereas uniform sampling (350 points) still exhibited residual errors, and the continuation method (DC) failed to detect all modes due to strong coupling.
• L-Shaped Bar: An arbitrary cross-section with no symmetries. The adaptive method achieved perfect tracking (191 points) compared to uniform sampling (250 points) which had residual errors.
• Steel Pipe: Demonstrated the necessity of Subspace MAC. Pointwise MAC failed in the low-wavenumber regime due to arbitrary rotation of degenerate eigenvectors. Switching to subspace tracking resolved this, achieving high confidence across the entire spectrum.

Efficiency: In all cases, the adaptive method used 30–50% fewer grid points than the dense uniform sampling required to achieve comparable or better accuracy.

5. Significance

This work bridges the gap between theoretical spectral analysis and practical computational efficiency in guided wave dispersion analysis.

• Theoretical Impact: It establishes the mathematical conditions under which mode tracking is guaranteed to succeed, moving beyond heuristic approaches.
• Practical Impact: The adaptive algorithm eliminates the need for users to manually guess critical step sizes or refine grids in "suspected" veering regions. It provides a "confidence score" for the results, which is crucial for non-destructive evaluation (NDE) and structural health monitoring (SHM).
• Software Availability: The authors have released the implementation as an open-source Python tool (DisperPy), making these advanced capabilities accessible to the broader research community.

In summary, the paper provides a mathematically rigorous, automated, and efficient solution to the long-standing challenge of mode tracking in SAFE dispersion analysis, particularly in complex scenarios involving mode veering and symmetry-induced degeneracy.